Update to last common bleve (#3986)

vendor/github.com/RoaringBitmap/roaring/AUTHORS

@@ -0,0 +1,10 @@
+# This is the official list of roaring authors for copyright purposes. 
+
+Todd Gruben (@tgruben),
+Daniel Lemire (@lemire),
+Elliot Murphy (@statik),
+Bob Potter (@bpot),
+Tyson Maly (@tvmaly),
+Will Glynn (@willglynn),
+Brent Pedersen (@brentp)
+Maciej Biłas (@maciej)

vendor/github.com/RoaringBitmap/roaring/CONTRIBUTORS

@@ -0,0 +1,12 @@
+# This is the official list of roaring contributors 
+
+Todd Gruben (@tgruben),
+Daniel Lemire (@lemire),
+Elliot Murphy (@statik),
+Bob Potter (@bpot),
+Tyson Maly (@tvmaly),
+Will Glynn (@willglynn),
+Brent Pedersen (@brentp),
+Jason E. Aten (@glycerine),
+Vali Malinoiu (@0x4139),
+Forud Ghafouri (@fzerorubigd)

vendor/github.com/RoaringBitmap/roaring/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2016 by the authors
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

vendor/github.com/RoaringBitmap/roaring/LICENSE-2.0.txt

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2016 by the authors
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

vendor/github.com/RoaringBitmap/roaring/Makefile

@@ -0,0 +1,121 @@
+.PHONY: help all test format fmtcheck vet lint     qa deps clean nuke rle backrle ser fetch-real-roaring-datasets
+
+
+
+
+
+
+
+
+# Display general help about this command
+help:
+	@echo ""
+	@echo "The following commands are available:"
+	@echo ""
+	@echo "    make qa          : Run all the tests"
+	@echo "    make test        : Run the unit tests"
+	@echo ""
+	@echo "    make format      : Format the source code"
+	@echo "    make fmtcheck    : Check if the source code has been formatted"
+	@echo "    make vet         : Check for suspicious constructs"
+	@echo "    make lint        : Check for style errors"
+	@echo ""
+	@echo "    make deps        : Get the dependencies"
+	@echo "    make clean       : Remove any build artifact"
+	@echo "    make nuke        : Deletes any intermediate file"
+	@echo ""
+	@echo "    make fuzz-smat   : Fuzzy testing with smat"
+	@echo "    make fuzz-stream : Fuzzy testing with stream deserialization"
+	@echo "    make fuzz-buffer : Fuzzy testing with buffer deserialization"
+	@echo ""
+
+# Alias for help target
+all: help
+test:
+	go test
+	go test -race -run TestConcurrent*
+# Format the source code
+format:
+	@find ./ -type f -name "*.go" -exec gofmt -w {} \;
+
+# Check if the source code has been formatted
+fmtcheck:
+	@mkdir -p target
+	@find ./ -type f -name "*.go" -exec gofmt -d {} \; | tee target/format.diff
+	@test ! -s target/format.diff || { echo "ERROR: the source code has not been formatted - please use 'make format' or 'gofmt'"; exit 1; }
+
+# Check for syntax errors
+vet:
+	GOPATH=$(GOPATH) go vet ./...
+
+# Check for style errors
+lint:
+	GOPATH=$(GOPATH) PATH=$(GOPATH)/bin:$(PATH) golint ./...
+
+
+
+
+
+# Alias to run all quality-assurance checks
+qa: fmtcheck test vet lint
+
+# --- INSTALL ---
+
+# Get the dependencies
+deps:
+	GOPATH=$(GOPATH) go get github.com/smartystreets/goconvey/convey
+	GOPATH=$(GOPATH) go get github.com/willf/bitset
+	GOPATH=$(GOPATH) go get github.com/golang/lint/golint
+	GOPATH=$(GOPATH) go get github.com/mschoch/smat
+	GOPATH=$(GOPATH) go get github.com/dvyukov/go-fuzz/go-fuzz
+	GOPATH=$(GOPATH) go get github.com/dvyukov/go-fuzz/go-fuzz-build
+	GOPATH=$(GOPATH) go get github.com/glycerine/go-unsnap-stream
+	GOPATH=$(GOPATH) go get github.com/philhofer/fwd
+	GOPATH=$(GOPATH) go get github.com/jtolds/gls
+
+fuzz-smat:
+	go test -tags=gofuzz -run=TestGenerateSmatCorpus
+	go-fuzz-build -func FuzzSmat github.com/RoaringBitmap/roaring
+	go-fuzz -bin=./roaring-fuzz.zip -workdir=workdir/ -timeout=200
+
+
+fuzz-stream:
+	go-fuzz-build -func FuzzSerializationStream github.com/RoaringBitmap/roaring
+	go-fuzz -bin=./roaring-fuzz.zip -workdir=workdir/ -timeout=200
+
+
+fuzz-buffer:
+	go-fuzz-build -func FuzzSerializationBuffer github.com/RoaringBitmap/roaring
+	go-fuzz -bin=./roaring-fuzz.zip -workdir=workdir/ -timeout=200
+
+# Remove any build artifact
+clean:
+	GOPATH=$(GOPATH) go clean ./...
+
+# Deletes any intermediate file
+nuke:
+	rm -rf ./target
+	GOPATH=$(GOPATH) go clean -i ./...
+
+rle:
+	cp rle.go rle16.go
+	perl -pi -e 's/32/16/g' rle16.go
+	cp rle_test.go rle16_test.go
+	perl -pi -e 's/32/16/g' rle16_test.go
+
+backrle:
+	cp rle16.go rle.go
+	perl -pi -e 's/16/32/g' rle.go
+	perl -pi -e 's/2032/2016/g' rle.go
+
+ser: rle
+	go generate
+
+cover:
+	go test -coverprofile=coverage.out
+	go tool cover -html=coverage.out
+
+fetch-real-roaring-datasets:
+	# pull github.com/RoaringBitmap/real-roaring-datasets -> testdata/real-roaring-datasets
+	git submodule init
+	git submodule update

vendor/github.com/RoaringBitmap/roaring/README.md

@@ -0,0 +1,246 @@
+roaring [![Build Status](https://travis-ci.org/RoaringBitmap/roaring.png)](https://travis-ci.org/RoaringBitmap/roaring) [![Coverage Status](https://coveralls.io/repos/github/RoaringBitmap/roaring/badge.svg?branch=master)](https://coveralls.io/github/RoaringBitmap/roaring?branch=master) [![GoDoc](https://godoc.org/github.com/RoaringBitmap/roaring?status.svg)](https://godoc.org/github.com/RoaringBitmap/roaring) [![Go Report Card](https://goreportcard.com/badge/RoaringBitmap/roaring)](https://goreportcard.com/report/github.com/RoaringBitmap/roaring)
+=============
+
+This is a go version of the Roaring bitmap data structure. 
+
+
+
+Roaring bitmaps are used by several major systems such as [Apache Lucene][lucene] and derivative systems such as [Solr][solr] and
+[Elasticsearch][elasticsearch], [Metamarkets' Druid][druid], [LinkedIn Pinot][pinot], [Netflix Atlas][atlas],  [Apache Spark][spark], [OpenSearchServer][opensearchserver], [Cloud Torrent][cloudtorrent], [Whoosh][whoosh],  [Pilosa][pilosa],  [Microsoft Visual Studio Team Services (VSTS)][vsts], and eBay's [Apache Kylin][kylin].
+
+[lucene]: https://lucene.apache.org/
+[solr]: https://lucene.apache.org/solr/
+[elasticsearch]: https://www.elastic.co/products/elasticsearch
+[druid]: http://druid.io/
+[spark]: https://spark.apache.org/
+[opensearchserver]: http://www.opensearchserver.com
+[cloudtorrent]: https://github.com/jpillora/cloud-torrent
+[whoosh]: https://bitbucket.org/mchaput/whoosh/wiki/Home
+[pilosa]: https://www.pilosa.com/
+[kylin]: http://kylin.apache.org/
+[pinot]: http://github.com/linkedin/pinot/wiki
+[vsts]: https://www.visualstudio.com/team-services/
+[atlas]: https://github.com/Netflix/atlas
+
+Roaring bitmaps are found to work well in many important applications:
+
+> Use Roaring for bitmap compression whenever possible. Do not use other bitmap compression methods ([Wang et al., SIGMOD 2017](http://db.ucsd.edu/wp-content/uploads/2017/03/sidm338-wangA.pdf))
+
+
+The ``roaring`` Go library is used by
+* [Cloud Torrent](https://github.com/jpillora/cloud-torrent): a self-hosted remote torrent client
+* [runv](https://github.com/hyperhq/runv): an Hypervisor-based runtime for the Open Containers Initiative
+* [InfluxDB](https://www.influxdata.com)
+* [Pilosa](https://www.pilosa.com/)
+* [Bleve](http://www.blevesearch.com)
+
+This library is used in production in several systems, it is part of the [Awesome Go collection](https://awesome-go.com).
+
+
+There are also  [Java](https://github.com/RoaringBitmap/RoaringBitmap) and [C/C++](https://github.com/RoaringBitmap/CRoaring) versions.  The Java, C, C++ and Go version are binary compatible: e.g,  you can save bitmaps
+from a Java program and load them back in Go, and vice versa. We have a [format specification](https://github.com/RoaringBitmap/RoaringFormatSpec).
+
+
+This code is licensed under Apache License, Version 2.0 (ASL2.0).
+
+Copyright 2016-... by the authors.
+
+
+### References
+
+- Daniel Lemire, Owen Kaser, Nathan Kurz, Luca Deri, Chris O'Hara, François Saint-Jacques, Gregory Ssi-Yan-Kai, Roaring Bitmaps: Implementation of an Optimized Software Library, Software: Practice and Experience 48 (4), 2018 [arXiv:1709.07821](https://arxiv.org/abs/1709.07821)
+-  Samy Chambi, Daniel Lemire, Owen Kaser, Robert Godin,
+Better bitmap performance with Roaring bitmaps,
+Software: Practice and Experience 46 (5), 2016.
+http://arxiv.org/abs/1402.6407 This paper used data from http://lemire.me/data/realroaring2014.html
+- Daniel Lemire, Gregory Ssi-Yan-Kai, Owen Kaser, Consistently faster and smaller compressed bitmaps with Roaring, Software: Practice and Experience 46 (11), 2016. http://arxiv.org/abs/1603.06549
+
+
+### Dependencies
+
+Dependencies are fetched automatically by giving the `-t` flag to `go get`.
+
+they include
+  - github.com/smartystreets/goconvey/convey
+  - github.com/willf/bitset
+  - github.com/mschoch/smat
+  - github.com/glycerine/go-unsnap-stream
+  - github.com/philhofer/fwd
+  - github.com/jtolds/gls
+
+Note that the smat library requires Go 1.6 or better.
+
+#### Installation
+
+  - go get -t github.com/RoaringBitmap/roaring
+
+
+### Example
+
+Here is a simplified but complete example:
+
+```go
+package main
+
+import (
+    "fmt"
+    "github.com/RoaringBitmap/roaring"
+    "bytes"
+)
+
+
+func main() {
+    // example inspired by https://github.com/fzandona/goroar
+    fmt.Println("==roaring==")
+    rb1 := roaring.BitmapOf(1, 2, 3, 4, 5, 100, 1000)
+    fmt.Println(rb1.String())
+
+    rb2 := roaring.BitmapOf(3, 4, 1000)
+    fmt.Println(rb2.String())
+
+    rb3 := roaring.New()
+    fmt.Println(rb3.String())
+
+    fmt.Println("Cardinality: ", rb1.GetCardinality())
+
+    fmt.Println("Contains 3? ", rb1.Contains(3))
+
+    rb1.And(rb2)
+
+    rb3.Add(1)
+    rb3.Add(5)
+
+    rb3.Or(rb1)
+
+    // computes union of the three bitmaps in parallel using 4 workers  
+    roaring.ParOr(4, rb1, rb2, rb3)
+    // computes intersection of the three bitmaps in parallel using 4 workers  
+    roaring.ParAnd(4, rb1, rb2, rb3)
+
+
+    // prints 1, 3, 4, 5, 1000
+    i := rb3.Iterator()
+    for i.HasNext() {
+        fmt.Println(i.Next())
+    }
+    fmt.Println()
+
+    // next we include an example of serialization
+    buf := new(bytes.Buffer)
+    rb1.WriteTo(buf) // we omit error handling
+    newrb:= roaring.New()
+    newrb.ReadFrom(buf)
+    if rb1.Equals(newrb) {
+    	fmt.Println("I wrote the content to a byte stream and read it back.")
+    }
+}
+```
+
+If you wish to use serialization and handle errors, you might want to
+consider the following sample of code:
+
+```go
+	rb := BitmapOf(1, 2, 3, 4, 5, 100, 1000)
+	buf := new(bytes.Buffer)
+	size,err:=rb.WriteTo(buf)
+	if err != nil {
+		t.Errorf("Failed writing")
+	}
+	newrb:= New()
+	size,err=newrb.ReadFrom(buf)
+	if err != nil {
+		t.Errorf("Failed reading")
+	}
+	if ! rb.Equals(newrb) {
+		t.Errorf("Cannot retrieve serialized version")
+	}
+```
+
+Given N integers in [0,x), then the serialized size in bytes of
+a Roaring bitmap should never exceed this bound:
+
+`` 8 + 9 * ((long)x+65535)/65536 + 2 * N ``
+
+That is, given a fixed overhead for the universe size (x), Roaring
+bitmaps never use more than 2 bytes per integer. You can call
+``BoundSerializedSizeInBytes`` for a more precise estimate.
+
+
+### Documentation
+
+Current documentation is available at http://godoc.org/github.com/RoaringBitmap/roaring
+
+### Goroutine safety
+
+In general, it should not generally be considered safe to access
+the same bitmaps using different goroutines--they are left
+unsynchronized for performance. Should you want to access
+a Bitmap from more than one goroutine, you should
+provide synchronization. Typically this is done by using channels to pass
+the *Bitmap around (in Go style; so there is only ever one owner),
+or by using `sync.Mutex` to serialize operations on Bitmaps.
+
+### Coverage
+
+We test our software. For a report on our test coverage, see
+
+https://coveralls.io/github/RoaringBitmap/roaring?branch=master
+
+### Benchmark
+
+Type
+
+         go test -bench Benchmark -run -
+         
+To run benchmarks on [Real Roaring Datasets](https://github.com/RoaringBitmap/real-roaring-datasets)
+run the following:
+
+```sh
+go get github.com/RoaringBitmap/real-roaring-datasets
+BENCH_REAL_DATA=1 go test -bench BenchmarkRealData -run -
+```
+
+### Iterative use
+
+You can use roaring with gore:
+
+- go get -u github.com/motemen/gore
+- Make sure that ``$GOPATH/bin`` is in your ``$PATH``.
+- go get github/RoaringBitmap/roaring
+
+```go
+$ gore
+gore version 0.2.6  :help for help
+gore> :import github.com/RoaringBitmap/roaring
+gore> x:=roaring.New()
+gore> x.Add(1)
+gore> x.String()
+"{1}"
+```
+
+
+### Fuzzy testing
+
+You can help us test further the library with fuzzy testing:
+
+         go get github.com/dvyukov/go-fuzz/go-fuzz
+         go get github.com/dvyukov/go-fuzz/go-fuzz-build
+         go test -tags=gofuzz -run=TestGenerateSmatCorpus
+         go-fuzz-build github.com/RoaringBitmap/roaring
+         go-fuzz -bin=./roaring-fuzz.zip -workdir=workdir/ -timeout=200
+
+Let it run, and if the # of crashers is > 0, check out the reports in
+the workdir where you should be able to find the panic goroutine stack
+traces.
+
+### Alternative in Go
+
+There is a Go version wrapping the C/C++ implementation https://github.com/RoaringBitmap/gocroaring
+
+For an alternative implementation in Go, see https://github.com/fzandona/goroar
+The two versions were written independently.
+
+
+### Mailing list/discussion group
+
+https://groups.google.com/forum/#!forum/roaring-bitmaps

vendor/github.com/RoaringBitmap/roaring/arraycontainer.go

@@ -0,0 +1,960 @@
+package roaring
+
+import (
+	"fmt"
+)
+
+//go:generate msgp -unexported
+
+type arrayContainer struct {
+	content []uint16
+}
+
+func (ac *arrayContainer) String() string {
+	s := "{"
+	for it := ac.getShortIterator(); it.hasNext(); {
+		s += fmt.Sprintf("%v, ", it.next())
+	}
+	return s + "}"
+}
+
+func (ac *arrayContainer) fillLeastSignificant16bits(x []uint32, i int, mask uint32) {
+	for k := 0; k < len(ac.content); k++ {
+		x[k+i] = uint32(ac.content[k]) | mask
+	}
+}
+
+func (ac *arrayContainer) getShortIterator() shortIterable {
+	return &shortIterator{ac.content, 0}
+}
+
+func (ac *arrayContainer) getManyIterator() manyIterable {
+	return &manyIterator{ac.content, 0}
+}
+
+func (ac *arrayContainer) minimum() uint16 {
+	return ac.content[0] // assume not empty
+}
+
+func (ac *arrayContainer) maximum() uint16 {
+	return ac.content[len(ac.content)-1] // assume not empty
+}
+
+func (ac *arrayContainer) getSizeInBytes() int {
+	return ac.getCardinality() * 2
+}
+
+func (ac *arrayContainer) serializedSizeInBytes() int {
+	return ac.getCardinality() * 2
+}
+
+func arrayContainerSizeInBytes(card int) int {
+	return card * 2
+}
+
+// add the values in the range [firstOfRange,endx)
+func (ac *arrayContainer) iaddRange(firstOfRange, endx int) container {
+	if firstOfRange >= endx {
+		return ac
+	}
+	indexstart := binarySearch(ac.content, uint16(firstOfRange))
+	if indexstart < 0 {
+		indexstart = -indexstart - 1
+	}
+	indexend := binarySearch(ac.content, uint16(endx-1))
+	if indexend < 0 {
+		indexend = -indexend - 1
+	} else {
+		indexend++
+	}
+	rangelength := endx - firstOfRange
+	newcardinality := indexstart + (ac.getCardinality() - indexend) + rangelength
+	if newcardinality > arrayDefaultMaxSize {
+		a := ac.toBitmapContainer()
+		return a.iaddRange(firstOfRange, endx)
+	}
+	if cap(ac.content) < newcardinality {
+		tmp := make([]uint16, newcardinality, newcardinality)
+		copy(tmp[:indexstart], ac.content[:indexstart])
+		copy(tmp[indexstart+rangelength:], ac.content[indexend:])
+
+		ac.content = tmp
+	} else {
+		ac.content = ac.content[:newcardinality]
+		copy(ac.content[indexstart+rangelength:], ac.content[indexend:])
+
+	}
+	for k := 0; k < rangelength; k++ {
+		ac.content[k+indexstart] = uint16(firstOfRange + k)
+	}
+	return ac
+}
+
+// remove the values in the range [firstOfRange,endx)
+func (ac *arrayContainer) iremoveRange(firstOfRange, endx int) container {
+	if firstOfRange >= endx {
+		return ac
+	}
+	indexstart := binarySearch(ac.content, uint16(firstOfRange))
+	if indexstart < 0 {
+		indexstart = -indexstart - 1
+	}
+	indexend := binarySearch(ac.content, uint16(endx-1))
+	if indexend < 0 {
+		indexend = -indexend - 1
+	} else {
+		indexend++
+	}
+	rangelength := indexend - indexstart
+	answer := ac
+	copy(answer.content[indexstart:], ac.content[indexstart+rangelength:])
+	answer.content = answer.content[:ac.getCardinality()-rangelength]
+	return answer
+}
+
+// flip the values in the range [firstOfRange,endx)
+func (ac *arrayContainer) not(firstOfRange, endx int) container {
+	if firstOfRange >= endx {
+		//p("arrayContainer.not(): exiting early with ac.clone()")
+		return ac.clone()
+	}
+	return ac.notClose(firstOfRange, endx-1) // remove everything in [firstOfRange,endx-1]
+}
+
+// flip the values in the range [firstOfRange,lastOfRange]
+func (ac *arrayContainer) notClose(firstOfRange, lastOfRange int) container {
+	if firstOfRange > lastOfRange { // unlike add and remove, not uses an inclusive range [firstOfRange,lastOfRange]
+		//p("arrayContainer.notClose(): exiting early with ac.clone()")
+		return ac.clone()
+	}
+
+	// determine the span of array indices to be affected^M
+	startIndex := binarySearch(ac.content, uint16(firstOfRange))
+	//p("startIndex=%v", startIndex)
+	if startIndex < 0 {
+		startIndex = -startIndex - 1
+	}
+	lastIndex := binarySearch(ac.content, uint16(lastOfRange))
+	//p("lastIndex=%v", lastIndex)
+	if lastIndex < 0 {
+		lastIndex = -lastIndex - 2
+	}
+	currentValuesInRange := lastIndex - startIndex + 1
+	spanToBeFlipped := lastOfRange - firstOfRange + 1
+	newValuesInRange := spanToBeFlipped - currentValuesInRange
+	cardinalityChange := newValuesInRange - currentValuesInRange
+	newCardinality := len(ac.content) + cardinalityChange
+	//p("new card is %v", newCardinality)
+	if newCardinality > arrayDefaultMaxSize {
+		//p("new card over arrayDefaultMaxSize, so returning bitmap")
+		return ac.toBitmapContainer().not(firstOfRange, lastOfRange+1)
+	}
+	answer := newArrayContainer()
+	answer.content = make([]uint16, newCardinality, newCardinality) //a hack for sure
+
+	copy(answer.content, ac.content[:startIndex])
+	outPos := startIndex
+	inPos := startIndex
+	valInRange := firstOfRange
+	for ; valInRange <= lastOfRange && inPos <= lastIndex; valInRange++ {
+		if uint16(valInRange) != ac.content[inPos] {
+			answer.content[outPos] = uint16(valInRange)
+			outPos++
+		} else {
+			inPos++
+		}
+	}
+
+	for ; valInRange <= lastOfRange; valInRange++ {
+		answer.content[outPos] = uint16(valInRange)
+		outPos++
+	}
+
+	for i := lastIndex + 1; i < len(ac.content); i++ {
+		answer.content[outPos] = ac.content[i]
+		outPos++
+	}
+	answer.content = answer.content[:newCardinality]
+	return answer
+
+}
+
+func (ac *arrayContainer) equals(o container) bool {
+
+	srb, ok := o.(*arrayContainer)
+	if ok {
+		// Check if the containers are the same object.
+		if ac == srb {
+			return true
+		}
+
+		if len(srb.content) != len(ac.content) {
+			return false
+		}
+
+		for i, v := range ac.content {
+			if v != srb.content[i] {
+				return false
+			}
+		}
+		return true
+	}
+
+	// use generic comparison
+	bCard := o.getCardinality()
+	aCard := ac.getCardinality()
+	if bCard != aCard {
+		return false
+	}
+
+	ait := ac.getShortIterator()
+	bit := o.getShortIterator()
+	for ait.hasNext() {
+		if bit.next() != ait.next() {
+			return false
+		}
+	}
+	return true
+}
+
+func (ac *arrayContainer) toBitmapContainer() *bitmapContainer {
+	bc := newBitmapContainer()
+	bc.loadData(ac)
+	return bc
+
+}
+func (ac *arrayContainer) iadd(x uint16) (wasNew bool) {
+	// Special case adding to the end of the container.
+	l := len(ac.content)
+	if l > 0 && l < arrayDefaultMaxSize && ac.content[l-1] < x {
+		ac.content = append(ac.content, x)
+		return true
+	}
+
+	loc := binarySearch(ac.content, x)
+
+	if loc < 0 {
+		s := ac.content
+		i := -loc - 1
+		s = append(s, 0)
+		copy(s[i+1:], s[i:])
+		s[i] = x
+		ac.content = s
+		return true
+	}
+	return false
+}
+
+func (ac *arrayContainer) iaddReturnMinimized(x uint16) container {
+	// Special case adding to the end of the container.
+	l := len(ac.content)
+	if l > 0 && l < arrayDefaultMaxSize && ac.content[l-1] < x {
+		ac.content = append(ac.content, x)
+		return ac
+	}
+
+	loc := binarySearch(ac.content, x)
+
+	if loc < 0 {
+		if len(ac.content) >= arrayDefaultMaxSize {
+			a := ac.toBitmapContainer()
+			a.iadd(x)
+			return a
+		}
+		s := ac.content
+		i := -loc - 1
+		s = append(s, 0)
+		copy(s[i+1:], s[i:])
+		s[i] = x
+		ac.content = s
+	}
+	return ac
+}
+
+// iremoveReturnMinimized is allowed to change the return type to minimize storage.
+func (ac *arrayContainer) iremoveReturnMinimized(x uint16) container {
+	ac.iremove(x)
+	return ac
+}
+
+func (ac *arrayContainer) iremove(x uint16) bool {
+	loc := binarySearch(ac.content, x)
+	if loc >= 0 {
+		s := ac.content
+		s = append(s[:loc], s[loc+1:]...)
+		ac.content = s
+		return true
+	}
+	return false
+}
+
+func (ac *arrayContainer) remove(x uint16) container {
+	out := &arrayContainer{make([]uint16, len(ac.content))}
+	copy(out.content, ac.content[:])
+
+	loc := binarySearch(out.content, x)
+	if loc >= 0 {
+		s := out.content
+		s = append(s[:loc], s[loc+1:]...)
+		out.content = s
+	}
+	return out
+}
+
+func (ac *arrayContainer) or(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.orArray(x)
+	case *bitmapContainer:
+		return x.orArray(ac)
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+		return x.orArray(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) orCardinality(a container) int {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.orArrayCardinality(x)
+	case *bitmapContainer:
+		return x.orArrayCardinality(ac)
+	case *runContainer16:
+		return x.orArrayCardinality(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) ior(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.iorArray(x)
+	case *bitmapContainer:
+		return a.(*bitmapContainer).orArray(ac)
+		//return ac.iorBitmap(x) // note: this does not make sense
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+		return ac.iorRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) iorArray(value2 *arrayContainer) container {
+	value1 := ac
+	len1 := value1.getCardinality()
+	len2 := value2.getCardinality()
+	maxPossibleCardinality := len1 + len2
+	if maxPossibleCardinality > arrayDefaultMaxSize { // it could be a bitmap!
+		bc := newBitmapContainer()
+		for k := 0; k < len(value2.content); k++ {
+			v := value2.content[k]
+			i := uint(v) >> 6
+			mask := uint64(1) << (v % 64)
+			bc.bitmap[i] |= mask
+		}
+		for k := 0; k < len(ac.content); k++ {
+			v := ac.content[k]
+			i := uint(v) >> 6
+			mask := uint64(1) << (v % 64)
+			bc.bitmap[i] |= mask
+		}
+		bc.cardinality = int(popcntSlice(bc.bitmap))
+		if bc.cardinality <= arrayDefaultMaxSize {
+			return bc.toArrayContainer()
+		}
+		return bc
+	}
+	if maxPossibleCardinality > cap(value1.content) {
+		newcontent := make([]uint16, 0, maxPossibleCardinality)
+		copy(newcontent[len2:maxPossibleCardinality], ac.content[0:len1])
+		ac.content = newcontent
+	} else {
+		copy(ac.content[len2:maxPossibleCardinality], ac.content[0:len1])
+	}
+	nl := union2by2(value1.content[len2:maxPossibleCardinality], value2.content, ac.content)
+	ac.content = ac.content[:nl] // reslice to match actual used capacity
+	return ac
+}
+
+// Note: such code does not make practical sense, except for lazy evaluations
+func (ac *arrayContainer) iorBitmap(bc2 *bitmapContainer) container {
+	bc1 := ac.toBitmapContainer()
+	bc1.iorBitmap(bc2)
+	*ac = *newArrayContainerFromBitmap(bc1)
+	return ac
+}
+
+func (ac *arrayContainer) iorRun16(rc *runContainer16) container {
+	bc1 := ac.toBitmapContainer()
+	bc2 := rc.toBitmapContainer()
+	bc1.iorBitmap(bc2)
+	*ac = *newArrayContainerFromBitmap(bc1)
+	return ac
+}
+
+func (ac *arrayContainer) lazyIOR(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.lazyIorArray(x)
+	case *bitmapContainer:
+		return ac.lazyIorBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+		return ac.lazyIorRun16(x)
+
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) lazyIorArray(ac2 *arrayContainer) container {
+	// TODO actually make this lazy
+	return ac.iorArray(ac2)
+}
+
+func (ac *arrayContainer) lazyIorBitmap(bc *bitmapContainer) container {
+	// TODO actually make this lazy
+	return ac.iorBitmap(bc)
+}
+
+func (ac *arrayContainer) lazyIorRun16(rc *runContainer16) container {
+	// TODO actually make this lazy
+	return ac.iorRun16(rc)
+}
+
+func (ac *arrayContainer) lazyOR(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.lazyorArray(x)
+	case *bitmapContainer:
+		return a.lazyOR(ac)
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+		return x.orArray(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) orArray(value2 *arrayContainer) container {
+	value1 := ac
+	maxPossibleCardinality := value1.getCardinality() + value2.getCardinality()
+	if maxPossibleCardinality > arrayDefaultMaxSize { // it could be a bitmap!
+		bc := newBitmapContainer()
+		for k := 0; k < len(value2.content); k++ {
+			v := value2.content[k]
+			i := uint(v) >> 6
+			mask := uint64(1) << (v % 64)
+			bc.bitmap[i] |= mask
+		}
+		for k := 0; k < len(ac.content); k++ {
+			v := ac.content[k]
+			i := uint(v) >> 6
+			mask := uint64(1) << (v % 64)
+			bc.bitmap[i] |= mask
+		}
+		bc.cardinality = int(popcntSlice(bc.bitmap))
+		if bc.cardinality <= arrayDefaultMaxSize {
+			return bc.toArrayContainer()
+		}
+		return bc
+	}
+	answer := newArrayContainerCapacity(maxPossibleCardinality)
+	nl := union2by2(value1.content, value2.content, answer.content)
+	answer.content = answer.content[:nl] // reslice to match actual used capacity
+	return answer
+}
+
+func (ac *arrayContainer) orArrayCardinality(value2 *arrayContainer) int {
+	return union2by2Cardinality(ac.content, value2.content)
+}
+
+func (ac *arrayContainer) lazyorArray(value2 *arrayContainer) container {
+	value1 := ac
+	maxPossibleCardinality := value1.getCardinality() + value2.getCardinality()
+	if maxPossibleCardinality > arrayLazyLowerBound { // it could be a bitmap!^M
+		bc := newBitmapContainer()
+		for k := 0; k < len(value2.content); k++ {
+			v := value2.content[k]
+			i := uint(v) >> 6
+			mask := uint64(1) << (v % 64)
+			bc.bitmap[i] |= mask
+		}
+		for k := 0; k < len(ac.content); k++ {
+			v := ac.content[k]
+			i := uint(v) >> 6
+			mask := uint64(1) << (v % 64)
+			bc.bitmap[i] |= mask
+		}
+		bc.cardinality = invalidCardinality
+		return bc
+	}
+	answer := newArrayContainerCapacity(maxPossibleCardinality)
+	nl := union2by2(value1.content, value2.content, answer.content)
+	answer.content = answer.content[:nl] // reslice to match actual used capacity
+	return answer
+}
+
+func (ac *arrayContainer) and(a container) container {
+	//p("ac.and() called")
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.andArray(x)
+	case *bitmapContainer:
+		return x.and(ac)
+	case *runContainer16:
+		if x.isFull() {
+			return ac.clone()
+		}
+		return x.andArray(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) andCardinality(a container) int {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.andArrayCardinality(x)
+	case *bitmapContainer:
+		return x.andCardinality(ac)
+	case *runContainer16:
+		return x.andArrayCardinality(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) intersects(a container) bool {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.intersectsArray(x)
+	case *bitmapContainer:
+		return x.intersects(ac)
+	case *runContainer16:
+		return x.intersects(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) iand(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.iandArray(x)
+	case *bitmapContainer:
+		return ac.iandBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return ac.clone()
+		}
+		return x.andArray(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) iandBitmap(bc *bitmapContainer) container {
+	pos := 0
+	c := ac.getCardinality()
+	for k := 0; k < c; k++ {
+		// branchless
+		v := ac.content[k]
+		ac.content[pos] = v
+		pos += int(bc.bitValue(v))
+	}
+	ac.content = ac.content[:pos]
+	return ac
+
+}
+
+func (ac *arrayContainer) xor(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.xorArray(x)
+	case *bitmapContainer:
+		return a.xor(ac)
+	case *runContainer16:
+		return x.xorArray(ac)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) xorArray(value2 *arrayContainer) container {
+	value1 := ac
+	totalCardinality := value1.getCardinality() + value2.getCardinality()
+	if totalCardinality > arrayDefaultMaxSize { // it could be a bitmap!
+		bc := newBitmapContainer()
+		for k := 0; k < len(value2.content); k++ {
+			v := value2.content[k]
+			i := uint(v) >> 6
+			bc.bitmap[i] ^= (uint64(1) << (v % 64))
+		}
+		for k := 0; k < len(ac.content); k++ {
+			v := ac.content[k]
+			i := uint(v) >> 6
+			bc.bitmap[i] ^= (uint64(1) << (v % 64))
+		}
+		bc.computeCardinality()
+		if bc.cardinality <= arrayDefaultMaxSize {
+			return bc.toArrayContainer()
+		}
+		return bc
+	}
+	desiredCapacity := totalCardinality
+	answer := newArrayContainerCapacity(desiredCapacity)
+	length := exclusiveUnion2by2(value1.content, value2.content, answer.content)
+	answer.content = answer.content[:length]
+	return answer
+
+}
+
+func (ac *arrayContainer) andNot(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.andNotArray(x)
+	case *bitmapContainer:
+		return ac.andNotBitmap(x)
+	case *runContainer16:
+		return ac.andNotRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) andNotRun16(rc *runContainer16) container {
+	acb := ac.toBitmapContainer()
+	rcb := rc.toBitmapContainer()
+	return acb.andNotBitmap(rcb)
+}
+
+func (ac *arrayContainer) iandNot(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return ac.iandNotArray(x)
+	case *bitmapContainer:
+		return ac.iandNotBitmap(x)
+	case *runContainer16:
+		return ac.iandNotRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (ac *arrayContainer) iandNotRun16(rc *runContainer16) container {
+	rcb := rc.toBitmapContainer()
+	acb := ac.toBitmapContainer()
+	acb.iandNotBitmapSurely(rcb)
+	*ac = *(acb.toArrayContainer())
+	return ac
+}
+
+func (ac *arrayContainer) andNotArray(value2 *arrayContainer) container {
+	value1 := ac
+	desiredcapacity := value1.getCardinality()
+	answer := newArrayContainerCapacity(desiredcapacity)
+	length := difference(value1.content, value2.content, answer.content)
+	answer.content = answer.content[:length]
+	return answer
+}
+
+func (ac *arrayContainer) iandNotArray(value2 *arrayContainer) container {
+	length := difference(ac.content, value2.content, ac.content)
+	ac.content = ac.content[:length]
+	return ac
+}
+
+func (ac *arrayContainer) andNotBitmap(value2 *bitmapContainer) container {
+	desiredcapacity := ac.getCardinality()
+	answer := newArrayContainerCapacity(desiredcapacity)
+	answer.content = answer.content[:desiredcapacity]
+	pos := 0
+	for _, v := range ac.content {
+		answer.content[pos] = v
+		pos += 1 - int(value2.bitValue(v))
+	}
+	answer.content = answer.content[:pos]
+	return answer
+}
+
+func (ac *arrayContainer) andBitmap(value2 *bitmapContainer) container {
+	desiredcapacity := ac.getCardinality()
+	answer := newArrayContainerCapacity(desiredcapacity)
+	answer.content = answer.content[:desiredcapacity]
+	pos := 0
+	for _, v := range ac.content {
+		answer.content[pos] = v
+		pos += int(value2.bitValue(v))
+	}
+	answer.content = answer.content[:pos]
+	return answer
+}
+
+func (ac *arrayContainer) iandNotBitmap(value2 *bitmapContainer) container {
+	pos := 0
+	for _, v := range ac.content {
+		ac.content[pos] = v
+		pos += 1 - int(value2.bitValue(v))
+	}
+	ac.content = ac.content[:pos]
+	return ac
+}
+
+func copyOf(array []uint16, size int) []uint16 {
+	result := make([]uint16, size)
+	for i, x := range array {
+		if i == size {
+			break
+		}
+		result[i] = x
+	}
+	return result
+}
+
+// flip the values in the range [firstOfRange,endx)
+func (ac *arrayContainer) inot(firstOfRange, endx int) container {
+	if firstOfRange >= endx {
+		return ac
+	}
+	return ac.inotClose(firstOfRange, endx-1) // remove everything in [firstOfRange,endx-1]
+}
+
+// flip the values in the range [firstOfRange,lastOfRange]
+func (ac *arrayContainer) inotClose(firstOfRange, lastOfRange int) container {
+	//p("ac.inotClose() starting")
+	if firstOfRange > lastOfRange { // unlike add and remove, not uses an inclusive range [firstOfRange,lastOfRange]
+		return ac
+	}
+	// determine the span of array indices to be affected
+	startIndex := binarySearch(ac.content, uint16(firstOfRange))
+	if startIndex < 0 {
+		startIndex = -startIndex - 1
+	}
+	lastIndex := binarySearch(ac.content, uint16(lastOfRange))
+	if lastIndex < 0 {
+		lastIndex = -lastIndex - 1 - 1
+	}
+	currentValuesInRange := lastIndex - startIndex + 1
+	spanToBeFlipped := lastOfRange - firstOfRange + 1
+
+	newValuesInRange := spanToBeFlipped - currentValuesInRange
+	buffer := make([]uint16, newValuesInRange)
+	cardinalityChange := newValuesInRange - currentValuesInRange
+	newCardinality := len(ac.content) + cardinalityChange
+	if cardinalityChange > 0 {
+		if newCardinality > len(ac.content) {
+			if newCardinality > arrayDefaultMaxSize {
+				//p("ac.inotClose() converting to bitmap and doing inot there")
+				bcRet := ac.toBitmapContainer()
+				bcRet.inot(firstOfRange, lastOfRange+1)
+				*ac = *bcRet.toArrayContainer()
+				return bcRet
+			}
+			ac.content = copyOf(ac.content, newCardinality)
+		}
+		base := lastIndex + 1
+		copy(ac.content[lastIndex+1+cardinalityChange:], ac.content[base:base+len(ac.content)-1-lastIndex])
+		ac.negateRange(buffer, startIndex, lastIndex, firstOfRange, lastOfRange+1)
+	} else { // no expansion needed
+		ac.negateRange(buffer, startIndex, lastIndex, firstOfRange, lastOfRange+1)
+		if cardinalityChange < 0 {
+
+			for i := startIndex + newValuesInRange; i < newCardinality; i++ {
+				ac.content[i] = ac.content[i-cardinalityChange]
+			}
+		}
+	}
+	ac.content = ac.content[:newCardinality]
+	//p("bottom of ac.inotClose(): returning ac")
+	return ac
+}
+
+func (ac *arrayContainer) negateRange(buffer []uint16, startIndex, lastIndex, startRange, lastRange int) {
+	// compute the negation into buffer
+	outPos := 0
+	inPos := startIndex // value here always >= valInRange,
+	// until it is exhausted
+	// n.b., we can start initially exhausted.
+
+	valInRange := startRange
+	for ; valInRange < lastRange && inPos <= lastIndex; valInRange++ {
+		if uint16(valInRange) != ac.content[inPos] {
+			buffer[outPos] = uint16(valInRange)
+			outPos++
+		} else {
+			inPos++
+		}
+	}
+
+	// if there are extra items (greater than the biggest
+	// pre-existing one in range), buffer them
+	for ; valInRange < lastRange; valInRange++ {
+		buffer[outPos] = uint16(valInRange)
+		outPos++
+	}
+
+	if outPos != len(buffer) {
+		panic("negateRange: internal bug")
+	}
+
+	for i, item := range buffer {
+		ac.content[i+startIndex] = item
+	}
+}
+
+func (ac *arrayContainer) isFull() bool {
+	return false
+}
+
+func (ac *arrayContainer) andArray(value2 *arrayContainer) container {
+	desiredcapacity := minOfInt(ac.getCardinality(), value2.getCardinality())
+	answer := newArrayContainerCapacity(desiredcapacity)
+	length := intersection2by2(
+		ac.content,
+		value2.content,
+		answer.content)
+	answer.content = answer.content[:length]
+	return answer
+}
+
+func (ac *arrayContainer) andArrayCardinality(value2 *arrayContainer) int {
+	return intersection2by2Cardinality(
+		ac.content,
+		value2.content)
+}
+
+func (ac *arrayContainer) intersectsArray(value2 *arrayContainer) bool {
+	return intersects2by2(
+		ac.content,
+		value2.content)
+}
+
+func (ac *arrayContainer) iandArray(value2 *arrayContainer) container {
+	length := intersection2by2(
+		ac.content,
+		value2.content,
+		ac.content)
+	ac.content = ac.content[:length]
+	return ac
+}
+
+func (ac *arrayContainer) getCardinality() int {
+	return len(ac.content)
+}
+
+func (ac *arrayContainer) rank(x uint16) int {
+	answer := binarySearch(ac.content, x)
+	if answer >= 0 {
+		return answer + 1
+	}
+	return -answer - 1
+
+}
+
+func (ac *arrayContainer) selectInt(x uint16) int {
+	return int(ac.content[x])
+}
+
+func (ac *arrayContainer) clone() container {
+	ptr := arrayContainer{make([]uint16, len(ac.content))}
+	copy(ptr.content, ac.content[:])
+	return &ptr
+}
+
+func (ac *arrayContainer) contains(x uint16) bool {
+	return binarySearch(ac.content, x) >= 0
+}
+
+func (ac *arrayContainer) loadData(bitmapContainer *bitmapContainer) {
+	ac.content = make([]uint16, bitmapContainer.cardinality, bitmapContainer.cardinality)
+	bitmapContainer.fillArray(ac.content)
+}
+func newArrayContainer() *arrayContainer {
+	p := new(arrayContainer)
+	return p
+}
+
+func newArrayContainerFromBitmap(bc *bitmapContainer) *arrayContainer {
+	ac := &arrayContainer{}
+	ac.loadData(bc)
+	return ac
+}
+
+func newArrayContainerCapacity(size int) *arrayContainer {
+	p := new(arrayContainer)
+	p.content = make([]uint16, 0, size)
+	return p
+}
+
+func newArrayContainerSize(size int) *arrayContainer {
+	p := new(arrayContainer)
+	p.content = make([]uint16, size, size)
+	return p
+}
+
+func newArrayContainerRange(firstOfRun, lastOfRun int) *arrayContainer {
+	valuesInRange := lastOfRun - firstOfRun + 1
+	this := newArrayContainerCapacity(valuesInRange)
+	for i := 0; i < valuesInRange; i++ {
+		this.content = append(this.content, uint16(firstOfRun+i))
+	}
+	return this
+}
+
+func (ac *arrayContainer) numberOfRuns() (nr int) {
+	n := len(ac.content)
+	var runlen uint16
+	var cur, prev uint16
+
+	switch n {
+	case 0:
+		return 0
+	case 1:
+		return 1
+	default:
+		for i := 1; i < n; i++ {
+			prev = ac.content[i-1]
+			cur = ac.content[i]
+
+			if cur == prev+1 {
+				runlen++
+			} else {
+				if cur < prev {
+					panic("then fundamental arrayContainer assumption of sorted ac.content was broken")
+				}
+				if cur == prev {
+					panic("then fundamental arrayContainer assumption of deduplicated content was broken")
+				} else {
+					nr++
+					runlen = 0
+				}
+			}
+		}
+		nr++
+	}
+	return
+}
+
+// convert to run or array *if needed*
+func (ac *arrayContainer) toEfficientContainer() container {
+
+	numRuns := ac.numberOfRuns()
+
+	sizeAsRunContainer := runContainer16SerializedSizeInBytes(numRuns)
+	sizeAsBitmapContainer := bitmapContainerSizeInBytes()
+	card := ac.getCardinality()
+	sizeAsArrayContainer := arrayContainerSizeInBytes(card)
+
+	if sizeAsRunContainer <= minOfInt(sizeAsBitmapContainer, sizeAsArrayContainer) {
+		return newRunContainer16FromArray(ac)
+	}
+	if card <= arrayDefaultMaxSize {
+		return ac
+	}
+	return ac.toBitmapContainer()
+}
+
+func (ac *arrayContainer) containerType() contype {
+	return arrayContype
+}

vendor/github.com/RoaringBitmap/roaring/arraycontainer_gen.go

@@ -0,0 +1,134 @@
+package roaring
+
+// NOTE: THIS FILE WAS PRODUCED BY THE
+// MSGP CODE GENERATION TOOL (github.com/tinylib/msgp)
+// DO NOT EDIT
+
+import "github.com/tinylib/msgp/msgp"
+
+// DecodeMsg implements msgp.Decodable
+func (z *arrayContainer) DecodeMsg(dc *msgp.Reader) (err error) {
+	var field []byte
+	_ = field
+	var zbzg uint32
+	zbzg, err = dc.ReadMapHeader()
+	if err != nil {
+		return
+	}
+	for zbzg > 0 {
+		zbzg--
+		field, err = dc.ReadMapKeyPtr()
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "content":
+			var zbai uint32
+			zbai, err = dc.ReadArrayHeader()
+			if err != nil {
+				return
+			}
+			if cap(z.content) >= int(zbai) {
+				z.content = (z.content)[:zbai]
+			} else {
+				z.content = make([]uint16, zbai)
+			}
+			for zxvk := range z.content {
+				z.content[zxvk], err = dc.ReadUint16()
+				if err != nil {
+					return
+				}
+			}
+		default:
+			err = dc.Skip()
+			if err != nil {
+				return
+			}
+		}
+	}
+	return
+}
+
+// EncodeMsg implements msgp.Encodable
+func (z *arrayContainer) EncodeMsg(en *msgp.Writer) (err error) {
+	// map header, size 1
+	// write "content"
+	err = en.Append(0x81, 0xa7, 0x63, 0x6f, 0x6e, 0x74, 0x65, 0x6e, 0x74)
+	if err != nil {
+		return err
+	}
+	err = en.WriteArrayHeader(uint32(len(z.content)))
+	if err != nil {
+		return
+	}
+	for zxvk := range z.content {
+		err = en.WriteUint16(z.content[zxvk])
+		if err != nil {
+			return
+		}
+	}
+	return
+}
+
+// MarshalMsg implements msgp.Marshaler
+func (z *arrayContainer) MarshalMsg(b []byte) (o []byte, err error) {
+	o = msgp.Require(b, z.Msgsize())
+	// map header, size 1
+	// string "content"
+	o = append(o, 0x81, 0xa7, 0x63, 0x6f, 0x6e, 0x74, 0x65, 0x6e, 0x74)
+	o = msgp.AppendArrayHeader(o, uint32(len(z.content)))
+	for zxvk := range z.content {
+		o = msgp.AppendUint16(o, z.content[zxvk])
+	}
+	return
+}
+
+// UnmarshalMsg implements msgp.Unmarshaler
+func (z *arrayContainer) UnmarshalMsg(bts []byte) (o []byte, err error) {
+	var field []byte
+	_ = field
+	var zcmr uint32
+	zcmr, bts, err = msgp.ReadMapHeaderBytes(bts)
+	if err != nil {
+		return
+	}
+	for zcmr > 0 {
+		zcmr--
+		field, bts, err = msgp.ReadMapKeyZC(bts)
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "content":
+			var zajw uint32
+			zajw, bts, err = msgp.ReadArrayHeaderBytes(bts)
+			if err != nil {
+				return
+			}
+			if cap(z.content) >= int(zajw) {
+				z.content = (z.content)[:zajw]
+			} else {
+				z.content = make([]uint16, zajw)
+			}
+			for zxvk := range z.content {
+				z.content[zxvk], bts, err = msgp.ReadUint16Bytes(bts)
+				if err != nil {
+					return
+				}
+			}
+		default:
+			bts, err = msgp.Skip(bts)
+			if err != nil {
+				return
+			}
+		}
+	}
+	o = bts
+	return
+}
+
+// Msgsize returns an upper bound estimate of the number of bytes occupied by the serialized message
+func (z *arrayContainer) Msgsize() (s int) {
+	s = 1 + 8 + msgp.ArrayHeaderSize + (len(z.content) * (msgp.Uint16Size))
+	return
+}

vendor/github.com/RoaringBitmap/roaring/bitmapcontainer.go

@@ -0,0 +1,982 @@
+package roaring
+
+import (
+	"fmt"
+	"unsafe"
+)
+
+//go:generate msgp -unexported
+
+type bitmapContainer struct {
+	cardinality int
+	bitmap      []uint64
+}
+
+func (bc bitmapContainer) String() string {
+	var s string
+	for it := bc.getShortIterator(); it.hasNext(); {
+		s += fmt.Sprintf("%v, ", it.next())
+	}
+	return s
+}
+
+func newBitmapContainer() *bitmapContainer {
+	p := new(bitmapContainer)
+	size := (1 << 16) / 64
+	p.bitmap = make([]uint64, size, size)
+	return p
+}
+
+func newBitmapContainerwithRange(firstOfRun, lastOfRun int) *bitmapContainer {
+	bc := newBitmapContainer()
+	bc.cardinality = lastOfRun - firstOfRun + 1
+	if bc.cardinality == maxCapacity {
+		fill(bc.bitmap, uint64(0xffffffffffffffff))
+	} else {
+		firstWord := firstOfRun / 64
+		lastWord := lastOfRun / 64
+		zeroPrefixLength := uint64(firstOfRun & 63)
+		zeroSuffixLength := uint64(63 - (lastOfRun & 63))
+
+		fillRange(bc.bitmap, firstWord, lastWord+1, uint64(0xffffffffffffffff))
+		bc.bitmap[firstWord] ^= ((uint64(1) << zeroPrefixLength) - 1)
+		blockOfOnes := (uint64(1) << zeroSuffixLength) - 1
+		maskOnLeft := blockOfOnes << (uint64(64) - zeroSuffixLength)
+		bc.bitmap[lastWord] ^= maskOnLeft
+	}
+	return bc
+}
+
+func (bc *bitmapContainer) minimum() uint16 {
+	for i := 0; i < len(bc.bitmap); i++ {
+		w := bc.bitmap[i]
+		if w != 0 {
+			r := countTrailingZeros(w)
+			return uint16(r + i*64)
+		}
+	}
+	return MaxUint16
+}
+
+// i should be non-zero
+func clz(i uint64) int {
+	n := 1
+	x := uint32(i >> 32)
+	if x == 0 {
+		n += 32
+		x = uint32(i)
+	}
+	if x>>16 == 0 {
+		n += 16
+		x = x << 16
+	}
+	if x>>24 == 0 {
+		n += 8
+		x = x << 8
+	}
+	if x>>28 == 0 {
+		n += 4
+		x = x << 4
+	}
+	if x>>30 == 0 {
+		n += 2
+		x = x << 2
+	}
+	return n - int(x>>31)
+}
+
+func (bc *bitmapContainer) maximum() uint16 {
+	for i := len(bc.bitmap); i > 0; i-- {
+		w := bc.bitmap[i-1]
+		if w != 0 {
+			r := clz(w)
+			return uint16((i-1)*64 + 63 - r)
+		}
+	}
+	return uint16(0)
+}
+
+type bitmapContainerShortIterator struct {
+	ptr *bitmapContainer
+	i   int
+}
+
+func (bcsi *bitmapContainerShortIterator) next() uint16 {
+	j := bcsi.i
+	bcsi.i = bcsi.ptr.NextSetBit(bcsi.i + 1)
+	return uint16(j)
+}
+func (bcsi *bitmapContainerShortIterator) hasNext() bool {
+	return bcsi.i >= 0
+}
+
+func newBitmapContainerShortIterator(a *bitmapContainer) *bitmapContainerShortIterator {
+	return &bitmapContainerShortIterator{a, a.NextSetBit(0)}
+}
+
+func (bc *bitmapContainer) getShortIterator() shortIterable {
+	return newBitmapContainerShortIterator(bc)
+}
+
+type bitmapContainerManyIterator struct {
+	ptr    *bitmapContainer
+	base   int
+	bitset uint64
+}
+
+func (bcmi *bitmapContainerManyIterator) nextMany(hs uint32, buf []uint32) int {
+	n := 0
+	base := bcmi.base
+	bitset := bcmi.bitset
+
+	for n < len(buf) {
+		if bitset == 0 {
+			base += 1
+			if base >= len(bcmi.ptr.bitmap) {
+				bcmi.base = base
+				bcmi.bitset = bitset
+				return n
+			}
+			bitset = bcmi.ptr.bitmap[base]
+			continue
+		}
+		t := bitset & -bitset
+		buf[n] = uint32(((base * 64) + int(popcount(t-1)))) | hs
+		n = n + 1
+		bitset ^= t
+	}
+
+	bcmi.base = base
+	bcmi.bitset = bitset
+	return n
+}
+
+func newBitmapContainerManyIterator(a *bitmapContainer) *bitmapContainerManyIterator {
+	return &bitmapContainerManyIterator{a, -1, 0}
+}
+
+func (bc *bitmapContainer) getManyIterator() manyIterable {
+	return newBitmapContainerManyIterator(bc)
+}
+
+func (bc *bitmapContainer) getSizeInBytes() int {
+	return len(bc.bitmap) * 8 // + bcBaseBytes
+}
+
+func (bc *bitmapContainer) serializedSizeInBytes() int {
+	//return bc.Msgsize()// NOO! This breaks GetSerializedSizeInBytes
+	return len(bc.bitmap) * 8
+}
+
+const bcBaseBytes = int(unsafe.Sizeof(bitmapContainer{}))
+
+// bitmapContainer doesn't depend on card, always fully allocated
+func bitmapContainerSizeInBytes() int {
+	return bcBaseBytes + (1<<16)/8
+}
+
+func bitmapEquals(a, b []uint64) bool {
+	if len(a) != len(b) {
+		//p("bitmaps differ on length. len(a)=%v; len(b)=%v", len(a), len(b))
+		return false
+	}
+	for i, v := range a {
+		if v != b[i] {
+			//p("bitmaps differ on element i=%v", i)
+			return false
+		}
+	}
+	//p("bitmapEquals returning true")
+	return true
+}
+
+func (bc *bitmapContainer) fillLeastSignificant16bits(x []uint32, i int, mask uint32) {
+	// TODO: should be written as optimized assembly
+	pos := i
+	base := mask
+	for k := 0; k < len(bc.bitmap); k++ {
+		bitset := bc.bitmap[k]
+		for bitset != 0 {
+			t := bitset & -bitset
+			x[pos] = base + uint32(popcount(t-1))
+			pos++
+			bitset ^= t
+		}
+		base += 64
+	}
+}
+
+func (bc *bitmapContainer) equals(o container) bool {
+	srb, ok := o.(*bitmapContainer)
+	if ok {
+		//p("bitmapContainers.equals: both are bitmapContainers")
+		if srb.cardinality != bc.cardinality {
+			//p("bitmapContainers.equals: card differs: %v vs %v", srb.cardinality, bc.cardinality)
+			return false
+		}
+		return bitmapEquals(bc.bitmap, srb.bitmap)
+	}
+
+	// use generic comparison
+	if bc.getCardinality() != o.getCardinality() {
+		return false
+	}
+	ait := o.getShortIterator()
+	bit := bc.getShortIterator()
+
+	for ait.hasNext() {
+		if bit.next() != ait.next() {
+			return false
+		}
+	}
+	return true
+}
+
+func (bc *bitmapContainer) iaddReturnMinimized(i uint16) container {
+	bc.iadd(i)
+	if bc.isFull() {
+		return newRunContainer16Range(0, MaxUint16)
+	}
+	return bc
+}
+
+func (bc *bitmapContainer) iadd(i uint16) bool {
+	x := int(i)
+	previous := bc.bitmap[x/64]
+	mask := uint64(1) << (uint(x) % 64)
+	newb := previous | mask
+	bc.bitmap[x/64] = newb
+	bc.cardinality += int((previous ^ newb) >> (uint(x) % 64))
+	return newb != previous
+}
+
+func (bc *bitmapContainer) iremoveReturnMinimized(i uint16) container {
+	if bc.iremove(i) {
+		if bc.cardinality == arrayDefaultMaxSize {
+			return bc.toArrayContainer()
+		}
+	}
+	return bc
+}
+
+// iremove returns true if i was found.
+func (bc *bitmapContainer) iremove(i uint16) bool {
+	/* branchless code
+	  w := bc.bitmap[i>>6]
+		mask := uint64(1) << (i % 64)
+		neww := w &^ mask
+		bc.cardinality -= int((w ^ neww) >> (i % 64))
+		bc.bitmap[i>>6] = neww */
+	if bc.contains(i) {
+		bc.cardinality--
+		bc.bitmap[i/64] &^= (uint64(1) << (i % 64))
+		return true
+	}
+	return false
+}
+
+func (bc *bitmapContainer) isFull() bool {
+	return bc.cardinality == int(MaxUint16)+1
+}
+
+func (bc *bitmapContainer) getCardinality() int {
+	return bc.cardinality
+}
+
+func (bc *bitmapContainer) clone() container {
+	ptr := bitmapContainer{bc.cardinality, make([]uint64, len(bc.bitmap))}
+	copy(ptr.bitmap, bc.bitmap[:])
+	return &ptr
+}
+
+// add all values in range [firstOfRange,lastOfRange)
+func (bc *bitmapContainer) iaddRange(firstOfRange, lastOfRange int) container {
+	bc.cardinality += setBitmapRangeAndCardinalityChange(bc.bitmap, firstOfRange, lastOfRange)
+	return bc
+}
+
+// remove all values in range [firstOfRange,lastOfRange)
+func (bc *bitmapContainer) iremoveRange(firstOfRange, lastOfRange int) container {
+	bc.cardinality += resetBitmapRangeAndCardinalityChange(bc.bitmap, firstOfRange, lastOfRange)
+	if bc.getCardinality() <= arrayDefaultMaxSize {
+		return bc.toArrayContainer()
+	}
+	return bc
+}
+
+// flip all values in range [firstOfRange,endx)
+func (bc *bitmapContainer) inot(firstOfRange, endx int) container {
+	p("bc.inot() called with [%v, %v)", firstOfRange, endx)
+	if endx-firstOfRange == maxCapacity {
+		//p("endx-firstOfRange == maxCapacity")
+		flipBitmapRange(bc.bitmap, firstOfRange, endx)
+		bc.cardinality = maxCapacity - bc.cardinality
+		//p("bc.cardinality is now %v", bc.cardinality)
+	} else if endx-firstOfRange > maxCapacity/2 {
+		//p("endx-firstOfRange > maxCapacity/2")
+		flipBitmapRange(bc.bitmap, firstOfRange, endx)
+		bc.computeCardinality()
+	} else {
+		bc.cardinality += flipBitmapRangeAndCardinalityChange(bc.bitmap, firstOfRange, endx)
+	}
+	if bc.getCardinality() <= arrayDefaultMaxSize {
+		return bc.toArrayContainer()
+	}
+	return bc
+}
+
+// flip all values in range [firstOfRange,endx)
+func (bc *bitmapContainer) not(firstOfRange, endx int) container {
+	answer := bc.clone()
+	return answer.inot(firstOfRange, endx)
+}
+
+func (bc *bitmapContainer) or(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.orArray(x)
+	case *bitmapContainer:
+		return bc.orBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+		return x.orBitmapContainer(bc)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) orCardinality(a container) int {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.orArrayCardinality(x)
+	case *bitmapContainer:
+		return bc.orBitmapCardinality(x)
+	case *runContainer16:
+		return x.orBitmapContainerCardinality(bc)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) ior(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.iorArray(x)
+	case *bitmapContainer:
+		return bc.iorBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+		for i := range x.iv {
+			bc.iaddRange(int(x.iv[i].start), int(x.iv[i].last())+1)
+		}
+		if bc.isFull() {
+			return newRunContainer16Range(0, MaxUint16)
+		}
+		//bc.computeCardinality()
+		return bc
+	}
+	panic(fmt.Errorf("unsupported container type %T", a))
+}
+
+func (bc *bitmapContainer) lazyIOR(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.lazyIORArray(x)
+	case *bitmapContainer:
+		return bc.lazyIORBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+
+		// Manually inlined setBitmapRange function
+		bitmap := bc.bitmap
+		for _, iv := range x.iv {
+			start := int(iv.start)
+			end := int(iv.last()) + 1
+			if start >= end {
+				continue
+			}
+			firstword := start / 64
+			endword := (end - 1) / 64
+			if firstword == endword {
+				bitmap[firstword] |= (^uint64(0) << uint(start%64)) & (^uint64(0) >> (uint(-end) % 64))
+				continue
+			}
+			bitmap[firstword] |= ^uint64(0) << uint(start%64)
+			for i := firstword + 1; i < endword; i++ {
+				bitmap[i] = ^uint64(0)
+			}
+			bitmap[endword] |= ^uint64(0) >> (uint(-end) % 64)
+		}
+		bc.cardinality = invalidCardinality
+		return bc
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) lazyOR(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.lazyORArray(x)
+	case *bitmapContainer:
+		return bc.lazyORBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return x.clone()
+		}
+		// TODO: implement lazy OR
+		return x.orBitmapContainer(bc)
+
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) orArray(value2 *arrayContainer) container {
+	answer := bc.clone().(*bitmapContainer)
+	c := value2.getCardinality()
+	for k := 0; k < c; k++ {
+		v := value2.content[k]
+		i := uint(v) >> 6
+		bef := answer.bitmap[i]
+		aft := bef | (uint64(1) << (v % 64))
+		answer.bitmap[i] = aft
+		answer.cardinality += int((bef - aft) >> 63)
+	}
+	return answer
+}
+
+func (bc *bitmapContainer) orArrayCardinality(value2 *arrayContainer) int {
+	answer := 0
+	c := value2.getCardinality()
+	for k := 0; k < c; k++ {
+		// branchless:
+		v := value2.content[k]
+		i := uint(v) >> 6
+		bef := bc.bitmap[i]
+		aft := bef | (uint64(1) << (v % 64))
+		answer += int((bef - aft) >> 63)
+	}
+	return answer
+}
+
+func (bc *bitmapContainer) orBitmap(value2 *bitmapContainer) container {
+	answer := newBitmapContainer()
+	for k := 0; k < len(answer.bitmap); k++ {
+		answer.bitmap[k] = bc.bitmap[k] | value2.bitmap[k]
+	}
+	answer.computeCardinality()
+	if answer.isFull() {
+		return newRunContainer16Range(0, MaxUint16)
+	}
+	return answer
+}
+
+func (bc *bitmapContainer) orBitmapCardinality(value2 *bitmapContainer) int {
+	return int(popcntOrSlice(bc.bitmap, value2.bitmap))
+}
+
+func (bc *bitmapContainer) andBitmapCardinality(value2 *bitmapContainer) int {
+	return int(popcntAndSlice(bc.bitmap, value2.bitmap))
+}
+
+func (bc *bitmapContainer) computeCardinality() {
+	bc.cardinality = int(popcntSlice(bc.bitmap))
+}
+
+func (bc *bitmapContainer) iorArray(ac *arrayContainer) container {
+	for k := range ac.content {
+		vc := ac.content[k]
+		i := uint(vc) >> 6
+		bef := bc.bitmap[i]
+		aft := bef | (uint64(1) << (vc % 64))
+		bc.bitmap[i] = aft
+		bc.cardinality += int((bef - aft) >> 63)
+	}
+	if bc.isFull() {
+		return newRunContainer16Range(0, MaxUint16)
+	}
+	return bc
+}
+
+func (bc *bitmapContainer) iorBitmap(value2 *bitmapContainer) container {
+	answer := bc
+	answer.cardinality = 0
+	for k := 0; k < len(answer.bitmap); k++ {
+		answer.bitmap[k] = bc.bitmap[k] | value2.bitmap[k]
+	}
+	answer.computeCardinality()
+	if bc.isFull() {
+		return newRunContainer16Range(0, MaxUint16)
+	}
+	return answer
+}
+
+func (bc *bitmapContainer) lazyIORArray(value2 *arrayContainer) container {
+	answer := bc
+	c := value2.getCardinality()
+	for k := 0; k < c; k++ {
+		vc := value2.content[k]
+		i := uint(vc) >> 6
+		answer.bitmap[i] = answer.bitmap[i] | (uint64(1) << (vc % 64))
+	}
+	answer.cardinality = invalidCardinality
+	return answer
+}
+
+func (bc *bitmapContainer) lazyORArray(value2 *arrayContainer) container {
+	answer := bc.clone().(*bitmapContainer)
+	return answer.lazyIORArray(value2)
+}
+
+func (bc *bitmapContainer) lazyIORBitmap(value2 *bitmapContainer) container {
+	answer := bc
+	for k := 0; k < len(answer.bitmap); k++ {
+		answer.bitmap[k] = bc.bitmap[k] | value2.bitmap[k]
+	}
+	bc.cardinality = invalidCardinality
+	return answer
+}
+
+func (bc *bitmapContainer) lazyORBitmap(value2 *bitmapContainer) container {
+	answer := bc.clone().(*bitmapContainer)
+	return answer.lazyIORBitmap(value2)
+}
+
+func (bc *bitmapContainer) xor(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.xorArray(x)
+	case *bitmapContainer:
+		return bc.xorBitmap(x)
+	case *runContainer16:
+		return x.xorBitmap(bc)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) xorArray(value2 *arrayContainer) container {
+	answer := bc.clone().(*bitmapContainer)
+	c := value2.getCardinality()
+	for k := 0; k < c; k++ {
+		vc := value2.content[k]
+		index := uint(vc) >> 6
+		abi := answer.bitmap[index]
+		mask := uint64(1) << (vc % 64)
+		answer.cardinality += 1 - 2*int((abi&mask)>>(vc%64))
+		answer.bitmap[index] = abi ^ mask
+	}
+	if answer.cardinality <= arrayDefaultMaxSize {
+		return answer.toArrayContainer()
+	}
+	return answer
+}
+
+func (bc *bitmapContainer) rank(x uint16) int {
+	// TODO: rewrite in assembly
+	leftover := (uint(x) + 1) & 63
+	if leftover == 0 {
+		return int(popcntSlice(bc.bitmap[:(uint(x)+1)/64]))
+	}
+	return int(popcntSlice(bc.bitmap[:(uint(x)+1)/64]) + popcount(bc.bitmap[(uint(x)+1)/64]<<(64-leftover)))
+}
+
+func (bc *bitmapContainer) selectInt(x uint16) int {
+	remaining := x
+	for k := 0; k < len(bc.bitmap); k++ {
+		w := popcount(bc.bitmap[k])
+		if uint16(w) > remaining {
+			return k*64 + selectBitPosition(bc.bitmap[k], int(remaining))
+		}
+		remaining -= uint16(w)
+	}
+	return -1
+}
+
+func (bc *bitmapContainer) xorBitmap(value2 *bitmapContainer) container {
+	newCardinality := int(popcntXorSlice(bc.bitmap, value2.bitmap))
+
+	if newCardinality > arrayDefaultMaxSize {
+		answer := newBitmapContainer()
+		for k := 0; k < len(answer.bitmap); k++ {
+			answer.bitmap[k] = bc.bitmap[k] ^ value2.bitmap[k]
+		}
+		answer.cardinality = newCardinality
+		if answer.isFull() {
+			return newRunContainer16Range(0, MaxUint16)
+		}
+		return answer
+	}
+	ac := newArrayContainerSize(newCardinality)
+	fillArrayXOR(ac.content, bc.bitmap, value2.bitmap)
+	ac.content = ac.content[:newCardinality]
+	return ac
+}
+
+func (bc *bitmapContainer) and(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.andArray(x)
+	case *bitmapContainer:
+		return bc.andBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return bc.clone()
+		}
+		return x.andBitmapContainer(bc)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) andCardinality(a container) int {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.andArrayCardinality(x)
+	case *bitmapContainer:
+		return bc.andBitmapCardinality(x)
+	case *runContainer16:
+		return x.andBitmapContainerCardinality(bc)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) intersects(a container) bool {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.intersectsArray(x)
+	case *bitmapContainer:
+		return bc.intersectsBitmap(x)
+	case *runContainer16:
+		return x.intersects(bc)
+
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) iand(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.iandArray(x)
+	case *bitmapContainer:
+		return bc.iandBitmap(x)
+	case *runContainer16:
+		if x.isFull() {
+			return bc.clone()
+		}
+		return bc.iandRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) iandRun16(rc *runContainer16) container {
+	rcb := newBitmapContainerFromRun(rc)
+	return bc.iandBitmap(rcb)
+}
+
+func (bc *bitmapContainer) iandArray(ac *arrayContainer) container {
+	acb := ac.toBitmapContainer()
+	return bc.iandBitmap(acb)
+}
+
+func (bc *bitmapContainer) andArray(value2 *arrayContainer) *arrayContainer {
+	answer := newArrayContainerCapacity(len(value2.content))
+	answer.content = answer.content[:cap(answer.content)]
+	c := value2.getCardinality()
+	pos := 0
+	for k := 0; k < c; k++ {
+		v := value2.content[k]
+		answer.content[pos] = v
+		pos += int(bc.bitValue(v))
+	}
+	answer.content = answer.content[:pos]
+	return answer
+}
+
+func (bc *bitmapContainer) andArrayCardinality(value2 *arrayContainer) int {
+	c := value2.getCardinality()
+	pos := 0
+	for k := 0; k < c; k++ {
+		v := value2.content[k]
+		pos += int(bc.bitValue(v))
+	}
+	return pos
+}
+
+func (bc *bitmapContainer) getCardinalityInRange(start, end uint) int {
+	if start >= end {
+		return 0
+	}
+	firstword := start / 64
+	endword := (end - 1) / 64
+	const allones = ^uint64(0)
+	if firstword == endword {
+		return int(popcount(bc.bitmap[firstword] & ((allones << (start % 64)) & (allones >> ((64 - end) & 63)))))
+	}
+	answer := popcount(bc.bitmap[firstword] & (allones << (start % 64)))
+	answer += popcntSlice(bc.bitmap[firstword+1 : endword])
+	answer += popcount(bc.bitmap[endword] & (allones >> ((64 - end) & 63)))
+	return int(answer)
+}
+
+func (bc *bitmapContainer) andBitmap(value2 *bitmapContainer) container {
+	newcardinality := int(popcntAndSlice(bc.bitmap, value2.bitmap))
+	if newcardinality > arrayDefaultMaxSize {
+		answer := newBitmapContainer()
+		for k := 0; k < len(answer.bitmap); k++ {
+			answer.bitmap[k] = bc.bitmap[k] & value2.bitmap[k]
+		}
+		answer.cardinality = newcardinality
+		return answer
+	}
+	ac := newArrayContainerSize(newcardinality)
+	fillArrayAND(ac.content, bc.bitmap, value2.bitmap)
+	ac.content = ac.content[:newcardinality] //not sure why i need this
+	return ac
+
+}
+
+func (bc *bitmapContainer) intersectsArray(value2 *arrayContainer) bool {
+	c := value2.getCardinality()
+	for k := 0; k < c; k++ {
+		v := value2.content[k]
+		if bc.contains(v) {
+			return true
+		}
+	}
+	return false
+}
+
+func (bc *bitmapContainer) intersectsBitmap(value2 *bitmapContainer) bool {
+	for k := 0; k < len(bc.bitmap); k++ {
+		if (bc.bitmap[k] & value2.bitmap[k]) != 0 {
+			return true
+		}
+	}
+	return false
+
+}
+
+func (bc *bitmapContainer) iandBitmap(value2 *bitmapContainer) container {
+	newcardinality := int(popcntAndSlice(bc.bitmap, value2.bitmap))
+	for k := 0; k < len(bc.bitmap); k++ {
+		bc.bitmap[k] = bc.bitmap[k] & value2.bitmap[k]
+	}
+	bc.cardinality = newcardinality
+
+	if newcardinality <= arrayDefaultMaxSize {
+		return newArrayContainerFromBitmap(bc)
+	}
+	return bc
+}
+
+func (bc *bitmapContainer) andNot(a container) container {
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.andNotArray(x)
+	case *bitmapContainer:
+		return bc.andNotBitmap(x)
+	case *runContainer16:
+		return bc.andNotRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) andNotRun16(rc *runContainer16) container {
+	rcb := rc.toBitmapContainer()
+	return bc.andNotBitmap(rcb)
+}
+
+func (bc *bitmapContainer) iandNot(a container) container {
+	//p("bitmapContainer.iandNot() starting")
+
+	switch x := a.(type) {
+	case *arrayContainer:
+		return bc.iandNotArray(x)
+	case *bitmapContainer:
+		return bc.iandNotBitmapSurely(x)
+	case *runContainer16:
+		return bc.iandNotRun16(x)
+	}
+	panic("unsupported container type")
+}
+
+func (bc *bitmapContainer) iandNotArray(ac *arrayContainer) container {
+	acb := ac.toBitmapContainer()
+	return bc.iandNotBitmapSurely(acb)
+}
+
+func (bc *bitmapContainer) iandNotRun16(rc *runContainer16) container {
+	rcb := rc.toBitmapContainer()
+	return bc.iandNotBitmapSurely(rcb)
+}
+
+func (bc *bitmapContainer) andNotArray(value2 *arrayContainer) container {
+	answer := bc.clone().(*bitmapContainer)
+	c := value2.getCardinality()
+	for k := 0; k < c; k++ {
+		vc := value2.content[k]
+		i := uint(vc) >> 6
+		oldv := answer.bitmap[i]
+		newv := oldv &^ (uint64(1) << (vc % 64))
+		answer.bitmap[i] = newv
+		answer.cardinality -= int((oldv ^ newv) >> (vc % 64))
+	}
+	if answer.cardinality <= arrayDefaultMaxSize {
+		return answer.toArrayContainer()
+	}
+	return answer
+}
+
+func (bc *bitmapContainer) andNotBitmap(value2 *bitmapContainer) container {
+	newCardinality := int(popcntMaskSlice(bc.bitmap, value2.bitmap))
+	if newCardinality > arrayDefaultMaxSize {
+		answer := newBitmapContainer()
+		for k := 0; k < len(answer.bitmap); k++ {
+			answer.bitmap[k] = bc.bitmap[k] &^ value2.bitmap[k]
+		}
+		answer.cardinality = newCardinality
+		return answer
+	}
+	ac := newArrayContainerSize(newCardinality)
+	fillArrayANDNOT(ac.content, bc.bitmap, value2.bitmap)
+	return ac
+}
+
+func (bc *bitmapContainer) iandNotBitmapSurely(value2 *bitmapContainer) *bitmapContainer {
+	newCardinality := int(popcntMaskSlice(bc.bitmap, value2.bitmap))
+	for k := 0; k < len(bc.bitmap); k++ {
+		bc.bitmap[k] = bc.bitmap[k] &^ value2.bitmap[k]
+	}
+	bc.cardinality = newCardinality
+	return bc
+}
+
+func (bc *bitmapContainer) contains(i uint16) bool { //testbit
+	x := uint(i)
+	w := bc.bitmap[x>>6]
+	mask := uint64(1) << (x & 63)
+	return (w & mask) != 0
+}
+
+func (bc *bitmapContainer) bitValue(i uint16) uint64 {
+	x := uint(i)
+	w := bc.bitmap[x>>6]
+	return (w >> (x & 63)) & 1
+}
+
+func (bc *bitmapContainer) loadData(arrayContainer *arrayContainer) {
+	bc.cardinality = arrayContainer.getCardinality()
+	c := arrayContainer.getCardinality()
+	for k := 0; k < c; k++ {
+		x := arrayContainer.content[k]
+		i := int(x) / 64
+		bc.bitmap[i] |= (uint64(1) << uint(x%64))
+	}
+}
+
+func (bc *bitmapContainer) toArrayContainer() *arrayContainer {
+	ac := &arrayContainer{}
+	ac.loadData(bc)
+	return ac
+}
+
+func (bc *bitmapContainer) fillArray(container []uint16) {
+	//TODO: rewrite in assembly
+	pos := 0
+	base := 0
+	for k := 0; k < len(bc.bitmap); k++ {
+		bitset := bc.bitmap[k]
+		for bitset != 0 {
+			t := bitset & -bitset
+			container[pos] = uint16((base + int(popcount(t-1))))
+			pos = pos + 1
+			bitset ^= t
+		}
+		base += 64
+	}
+}
+
+func (bc *bitmapContainer) NextSetBit(i int) int {
+	x := i / 64
+	if x >= len(bc.bitmap) {
+		return -1
+	}
+	w := bc.bitmap[x]
+	w = w >> uint(i%64)
+	if w != 0 {
+		return i + countTrailingZeros(w)
+	}
+	x++
+	for ; x < len(bc.bitmap); x++ {
+		if bc.bitmap[x] != 0 {
+			return (x * 64) + countTrailingZeros(bc.bitmap[x])
+		}
+	}
+	return -1
+}
+
+// reference the java implementation
+// https://github.com/RoaringBitmap/RoaringBitmap/blob/master/src/main/java/org/roaringbitmap/BitmapContainer.java#L875-L892
+//
+func (bc *bitmapContainer) numberOfRuns() int {
+	if bc.cardinality == 0 {
+		return 0
+	}
+
+	var numRuns uint64
+	nextWord := bc.bitmap[0]
+
+	for i := 0; i < len(bc.bitmap)-1; i++ {
+		word := nextWord
+		nextWord = bc.bitmap[i+1]
+		numRuns += popcount((^word)&(word<<1)) + ((word >> 63) &^ nextWord)
+	}
+
+	word := nextWord
+	numRuns += popcount((^word) & (word << 1))
+	if (word & 0x8000000000000000) != 0 {
+		numRuns++
+	}
+
+	return int(numRuns)
+}
+
+// convert to run or array *if needed*
+func (bc *bitmapContainer) toEfficientContainer() container {
+
+	numRuns := bc.numberOfRuns()
+
+	sizeAsRunContainer := runContainer16SerializedSizeInBytes(numRuns)
+	sizeAsBitmapContainer := bitmapContainerSizeInBytes()
+	card := bc.getCardinality()
+	sizeAsArrayContainer := arrayContainerSizeInBytes(card)
+
+	if sizeAsRunContainer <= minOfInt(sizeAsBitmapContainer, sizeAsArrayContainer) {
+		return newRunContainer16FromBitmapContainer(bc)
+	}
+	if card <= arrayDefaultMaxSize {
+		return bc.toArrayContainer()
+	}
+	return bc
+}
+
+func newBitmapContainerFromRun(rc *runContainer16) *bitmapContainer {
+
+	if len(rc.iv) == 1 {
+		return newBitmapContainerwithRange(int(rc.iv[0].start), int(rc.iv[0].last()))
+	}
+
+	bc := newBitmapContainer()
+	for i := range rc.iv {
+		setBitmapRange(bc.bitmap, int(rc.iv[i].start), int(rc.iv[i].last())+1)
+		bc.cardinality += int(rc.iv[i].last()) + 1 - int(rc.iv[i].start)
+	}
+	//bc.computeCardinality()
+	return bc
+}
+
+func (bc *bitmapContainer) containerType() contype {
+	return bitmapContype
+}

vendor/github.com/RoaringBitmap/roaring/bitmapcontainer_gen.go

@@ -0,0 +1,415 @@
+package roaring
+
+// NOTE: THIS FILE WAS PRODUCED BY THE
+// MSGP CODE GENERATION TOOL (github.com/tinylib/msgp)
+// DO NOT EDIT
+
+import "github.com/tinylib/msgp/msgp"
+
+// DecodeMsg implements msgp.Decodable
+func (z *bitmapContainer) DecodeMsg(dc *msgp.Reader) (err error) {
+	var field []byte
+	_ = field
+	var zbzg uint32
+	zbzg, err = dc.ReadMapHeader()
+	if err != nil {
+		return
+	}
+	for zbzg > 0 {
+		zbzg--
+		field, err = dc.ReadMapKeyPtr()
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "cardinality":
+			z.cardinality, err = dc.ReadInt()
+			if err != nil {
+				return
+			}
+		case "bitmap":
+			var zbai uint32
+			zbai, err = dc.ReadArrayHeader()
+			if err != nil {
+				return
+			}
+			if cap(z.bitmap) >= int(zbai) {
+				z.bitmap = (z.bitmap)[:zbai]
+			} else {
+				z.bitmap = make([]uint64, zbai)
+			}
+			for zxvk := range z.bitmap {
+				z.bitmap[zxvk], err = dc.ReadUint64()
+				if err != nil {
+					return
+				}
+			}
+		default:
+			err = dc.Skip()
+			if err != nil {
+				return
+			}
+		}
+	}
+	return
+}
+
+// EncodeMsg implements msgp.Encodable
+func (z *bitmapContainer) EncodeMsg(en *msgp.Writer) (err error) {
+	// map header, size 2
+	// write "cardinality"
+	err = en.Append(0x82, 0xab, 0x63, 0x61, 0x72, 0x64, 0x69, 0x6e, 0x61, 0x6c, 0x69, 0x74, 0x79)
+	if err != nil {
+		return err
+	}
+	err = en.WriteInt(z.cardinality)
+	if err != nil {
+		return
+	}
+	// write "bitmap"
+	err = en.Append(0xa6, 0x62, 0x69, 0x74, 0x6d, 0x61, 0x70)
+	if err != nil {
+		return err
+	}
+	err = en.WriteArrayHeader(uint32(len(z.bitmap)))
+	if err != nil {
+		return
+	}
+	for zxvk := range z.bitmap {
+		err = en.WriteUint64(z.bitmap[zxvk])
+		if err != nil {
+			return
+		}
+	}
+	return
+}
+
+// MarshalMsg implements msgp.Marshaler
+func (z *bitmapContainer) MarshalMsg(b []byte) (o []byte, err error) {
+	o = msgp.Require(b, z.Msgsize())
+	// map header, size 2
+	// string "cardinality"
+	o = append(o, 0x82, 0xab, 0x63, 0x61, 0x72, 0x64, 0x69, 0x6e, 0x61, 0x6c, 0x69, 0x74, 0x79)
+	o = msgp.AppendInt(o, z.cardinality)
+	// string "bitmap"
+	o = append(o, 0xa6, 0x62, 0x69, 0x74, 0x6d, 0x61, 0x70)
+	o = msgp.AppendArrayHeader(o, uint32(len(z.bitmap)))
+	for zxvk := range z.bitmap {
+		o = msgp.AppendUint64(o, z.bitmap[zxvk])
+	}
+	return
+}
+
+// UnmarshalMsg implements msgp.Unmarshaler
+func (z *bitmapContainer) UnmarshalMsg(bts []byte) (o []byte, err error) {
+	var field []byte
+	_ = field
+	var zcmr uint32
+	zcmr, bts, err = msgp.ReadMapHeaderBytes(bts)
+	if err != nil {
+		return
+	}
+	for zcmr > 0 {
+		zcmr--
+		field, bts, err = msgp.ReadMapKeyZC(bts)
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "cardinality":
+			z.cardinality, bts, err = msgp.ReadIntBytes(bts)
+			if err != nil {
+				return
+			}
+		case "bitmap":
+			var zajw uint32
+			zajw, bts, err = msgp.ReadArrayHeaderBytes(bts)
+			if err != nil {
+				return
+			}
+			if cap(z.bitmap) >= int(zajw) {
+				z.bitmap = (z.bitmap)[:zajw]
+			} else {
+				z.bitmap = make([]uint64, zajw)
+			}
+			for zxvk := range z.bitmap {
+				z.bitmap[zxvk], bts, err = msgp.ReadUint64Bytes(bts)
+				if err != nil {
+					return
+				}
+			}
+		default:
+			bts, err = msgp.Skip(bts)
+			if err != nil {
+				return
+			}
+		}
+	}
+	o = bts
+	return
+}
+
+// Msgsize returns an upper bound estimate of the number of bytes occupied by the serialized message
+func (z *bitmapContainer) Msgsize() (s int) {
+	s = 1 + 12 + msgp.IntSize + 7 + msgp.ArrayHeaderSize + (len(z.bitmap) * (msgp.Uint64Size))
+	return
+}
+
+// DecodeMsg implements msgp.Decodable
+func (z *bitmapContainerShortIterator) DecodeMsg(dc *msgp.Reader) (err error) {
+	var field []byte
+	_ = field
+	var zhct uint32
+	zhct, err = dc.ReadMapHeader()
+	if err != nil {
+		return
+	}
+	for zhct > 0 {
+		zhct--
+		field, err = dc.ReadMapKeyPtr()
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "ptr":
+			if dc.IsNil() {
+				err = dc.ReadNil()
+				if err != nil {
+					return
+				}
+				z.ptr = nil
+			} else {
+				if z.ptr == nil {
+					z.ptr = new(bitmapContainer)
+				}
+				var zcua uint32
+				zcua, err = dc.ReadMapHeader()
+				if err != nil {
+					return
+				}
+				for zcua > 0 {
+					zcua--
+					field, err = dc.ReadMapKeyPtr()
+					if err != nil {
+						return
+					}
+					switch msgp.UnsafeString(field) {
+					case "cardinality":
+						z.ptr.cardinality, err = dc.ReadInt()
+						if err != nil {
+							return
+						}
+					case "bitmap":
+						var zxhx uint32
+						zxhx, err = dc.ReadArrayHeader()
+						if err != nil {
+							return
+						}
+						if cap(z.ptr.bitmap) >= int(zxhx) {
+							z.ptr.bitmap = (z.ptr.bitmap)[:zxhx]
+						} else {
+							z.ptr.bitmap = make([]uint64, zxhx)
+						}
+						for zwht := range z.ptr.bitmap {
+							z.ptr.bitmap[zwht], err = dc.ReadUint64()
+							if err != nil {
+								return
+							}
+						}
+					default:
+						err = dc.Skip()
+						if err != nil {
+							return
+						}
+					}
+				}
+			}
+		case "i":
+			z.i, err = dc.ReadInt()
+			if err != nil {
+				return
+			}
+		default:
+			err = dc.Skip()
+			if err != nil {
+				return
+			}
+		}
+	}
+	return
+}
+
+// EncodeMsg implements msgp.Encodable
+func (z *bitmapContainerShortIterator) EncodeMsg(en *msgp.Writer) (err error) {
+	// map header, size 2
+	// write "ptr"
+	err = en.Append(0x82, 0xa3, 0x70, 0x74, 0x72)
+	if err != nil {
+		return err
+	}
+	if z.ptr == nil {
+		err = en.WriteNil()
+		if err != nil {
+			return
+		}
+	} else {
+		// map header, size 2
+		// write "cardinality"
+		err = en.Append(0x82, 0xab, 0x63, 0x61, 0x72, 0x64, 0x69, 0x6e, 0x61, 0x6c, 0x69, 0x74, 0x79)
+		if err != nil {
+			return err
+		}
+		err = en.WriteInt(z.ptr.cardinality)
+		if err != nil {
+			return
+		}
+		// write "bitmap"
+		err = en.Append(0xa6, 0x62, 0x69, 0x74, 0x6d, 0x61, 0x70)
+		if err != nil {
+			return err
+		}
+		err = en.WriteArrayHeader(uint32(len(z.ptr.bitmap)))
+		if err != nil {
+			return
+		}
+		for zwht := range z.ptr.bitmap {
+			err = en.WriteUint64(z.ptr.bitmap[zwht])
+			if err != nil {
+				return
+			}
+		}
+	}
+	// write "i"
+	err = en.Append(0xa1, 0x69)
+	if err != nil {
+		return err
+	}
+	err = en.WriteInt(z.i)
+	if err != nil {
+		return
+	}
+	return
+}
+
+// MarshalMsg implements msgp.Marshaler
+func (z *bitmapContainerShortIterator) MarshalMsg(b []byte) (o []byte, err error) {
+	o = msgp.Require(b, z.Msgsize())
+	// map header, size 2
+	// string "ptr"
+	o = append(o, 0x82, 0xa3, 0x70, 0x74, 0x72)
+	if z.ptr == nil {
+		o = msgp.AppendNil(o)
+	} else {
+		// map header, size 2
+		// string "cardinality"
+		o = append(o, 0x82, 0xab, 0x63, 0x61, 0x72, 0x64, 0x69, 0x6e, 0x61, 0x6c, 0x69, 0x74, 0x79)
+		o = msgp.AppendInt(o, z.ptr.cardinality)
+		// string "bitmap"
+		o = append(o, 0xa6, 0x62, 0x69, 0x74, 0x6d, 0x61, 0x70)
+		o = msgp.AppendArrayHeader(o, uint32(len(z.ptr.bitmap)))
+		for zwht := range z.ptr.bitmap {
+			o = msgp.AppendUint64(o, z.ptr.bitmap[zwht])
+		}
+	}
+	// string "i"
+	o = append(o, 0xa1, 0x69)
+	o = msgp.AppendInt(o, z.i)
+	return
+}
+
+// UnmarshalMsg implements msgp.Unmarshaler
+func (z *bitmapContainerShortIterator) UnmarshalMsg(bts []byte) (o []byte, err error) {
+	var field []byte
+	_ = field
+	var zlqf uint32
+	zlqf, bts, err = msgp.ReadMapHeaderBytes(bts)
+	if err != nil {
+		return
+	}
+	for zlqf > 0 {
+		zlqf--
+		field, bts, err = msgp.ReadMapKeyZC(bts)
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "ptr":
+			if msgp.IsNil(bts) {
+				bts, err = msgp.ReadNilBytes(bts)
+				if err != nil {
+					return
+				}
+				z.ptr = nil
+			} else {
+				if z.ptr == nil {
+					z.ptr = new(bitmapContainer)
+				}
+				var zdaf uint32
+				zdaf, bts, err = msgp.ReadMapHeaderBytes(bts)
+				if err != nil {
+					return
+				}
+				for zdaf > 0 {
+					zdaf--
+					field, bts, err = msgp.ReadMapKeyZC(bts)
+					if err != nil {
+						return
+					}
+					switch msgp.UnsafeString(field) {
+					case "cardinality":
+						z.ptr.cardinality, bts, err = msgp.ReadIntBytes(bts)
+						if err != nil {
+							return
+						}
+					case "bitmap":
+						var zpks uint32
+						zpks, bts, err = msgp.ReadArrayHeaderBytes(bts)
+						if err != nil {
+							return
+						}
+						if cap(z.ptr.bitmap) >= int(zpks) {
+							z.ptr.bitmap = (z.ptr.bitmap)[:zpks]
+						} else {
+							z.ptr.bitmap = make([]uint64, zpks)
+						}
+						for zwht := range z.ptr.bitmap {
+							z.ptr.bitmap[zwht], bts, err = msgp.ReadUint64Bytes(bts)
+							if err != nil {
+								return
+							}
+						}
+					default:
+						bts, err = msgp.Skip(bts)
+						if err != nil {
+							return
+						}
+					}
+				}
+			}
+		case "i":
+			z.i, bts, err = msgp.ReadIntBytes(bts)
+			if err != nil {
+				return
+			}
+		default:
+			bts, err = msgp.Skip(bts)
+			if err != nil {
+				return
+			}
+		}
+	}
+	o = bts
+	return
+}
+
+// Msgsize returns an upper bound estimate of the number of bytes occupied by the serialized message
+func (z *bitmapContainerShortIterator) Msgsize() (s int) {
+	s = 1 + 4
+	if z.ptr == nil {
+		s += msgp.NilSize
+	} else {
+		s += 1 + 12 + msgp.IntSize + 7 + msgp.ArrayHeaderSize + (len(z.ptr.bitmap) * (msgp.Uint64Size))
+	}
+	s += 2 + msgp.IntSize
+	return
+}

vendor/github.com/RoaringBitmap/roaring/ctz.go

@@ -0,0 +1,11 @@
+// +build go1.9
+// "go1.9", from Go version 1.9 onward
+// See https://golang.org/pkg/go/build/#hdr-Build_Constraints
+
+package roaring
+
+import "math/bits"
+
+func countTrailingZeros(x uint64) int {
+	return bits.TrailingZeros64(x)
+}

vendor/github.com/RoaringBitmap/roaring/ctz_compat.go

@@ -0,0 +1,71 @@
+// +build !go1.9
+
+package roaring
+
+// Reuse of portions of go/src/math/big standard lib code
+// under this license:
+/*
+Copyright (c) 2009 The Go Authors. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+   * Redistributions of source code must retain the above copyright
+notice, this list of conditions and the following disclaimer.
+   * Redistributions in binary form must reproduce the above
+copyright notice, this list of conditions and the following disclaimer
+in the documentation and/or other materials provided with the
+distribution.
+   * Neither the name of Google Inc. nor the names of its
+contributors may be used to endorse or promote products derived from
+this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+const deBruijn32 = 0x077CB531
+
+var deBruijn32Lookup = []byte{
+	0, 1, 28, 2, 29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4, 8,
+	31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6, 11, 5, 10, 9,
+}
+
+const deBruijn64 = 0x03f79d71b4ca8b09
+
+var deBruijn64Lookup = []byte{
+	0, 1, 56, 2, 57, 49, 28, 3, 61, 58, 42, 50, 38, 29, 17, 4,
+	62, 47, 59, 36, 45, 43, 51, 22, 53, 39, 33, 30, 24, 18, 12, 5,
+	63, 55, 48, 27, 60, 41, 37, 16, 46, 35, 44, 21, 52, 32, 23, 11,
+	54, 26, 40, 15, 34, 20, 31, 10, 25, 14, 19, 9, 13, 8, 7, 6,
+}
+
+// trailingZeroBits returns the number of consecutive least significant zero
+// bits of x.
+func countTrailingZeros(x uint64) int {
+	// x & -x leaves only the right-most bit set in the word. Let k be the
+	// index of that bit. Since only a single bit is set, the value is two
+	// to the power of k. Multiplying by a power of two is equivalent to
+	// left shifting, in this case by k bits. The de Bruijn constant is
+	// such that all six bit, consecutive substrings are distinct.
+	// Therefore, if we have a left shifted version of this constant we can
+	// find by how many bits it was shifted by looking at which six bit
+	// substring ended up at the top of the word.
+	// (Knuth, volume 4, section 7.3.1)
+	if x == 0 {
+		// We have to special case 0; the fomula
+		// below doesn't work for 0.
+		return 64
+	}
+	return int(deBruijn64Lookup[((x&-x)*(deBruijn64))>>58])
+}

vendor/github.com/RoaringBitmap/roaring/fastaggregation.go

@@ -0,0 +1,215 @@
+package roaring
+
+import (
+	"container/heap"
+)
+
+// Or function that requires repairAfterLazy
+func lazyOR(x1, x2 *Bitmap) *Bitmap {
+	answer := NewBitmap()
+	pos1 := 0
+	pos2 := 0
+	length1 := x1.highlowcontainer.size()
+	length2 := x2.highlowcontainer.size()
+main:
+	for (pos1 < length1) && (pos2 < length2) {
+		s1 := x1.highlowcontainer.getKeyAtIndex(pos1)
+		s2 := x2.highlowcontainer.getKeyAtIndex(pos2)
+
+		for {
+			if s1 < s2 {
+				answer.highlowcontainer.appendCopy(x1.highlowcontainer, pos1)
+				pos1++
+				if pos1 == length1 {
+					break main
+				}
+				s1 = x1.highlowcontainer.getKeyAtIndex(pos1)
+			} else if s1 > s2 {
+				answer.highlowcontainer.appendCopy(x2.highlowcontainer, pos2)
+				pos2++
+				if pos2 == length2 {
+					break main
+				}
+				s2 = x2.highlowcontainer.getKeyAtIndex(pos2)
+			} else {
+				c1 := x1.highlowcontainer.getContainerAtIndex(pos1)
+				switch t := c1.(type) {
+				case *arrayContainer:
+					c1 = t.toBitmapContainer()
+				case *runContainer16:
+					if !t.isFull() {
+						c1 = t.toBitmapContainer()
+					}
+				}
+
+				answer.highlowcontainer.appendContainer(s1, c1.lazyOR(x2.highlowcontainer.getContainerAtIndex(pos2)), false)
+				pos1++
+				pos2++
+				if (pos1 == length1) || (pos2 == length2) {
+					break main
+				}
+				s1 = x1.highlowcontainer.getKeyAtIndex(pos1)
+				s2 = x2.highlowcontainer.getKeyAtIndex(pos2)
+			}
+		}
+	}
+	if pos1 == length1 {
+		answer.highlowcontainer.appendCopyMany(x2.highlowcontainer, pos2, length2)
+	} else if pos2 == length2 {
+		answer.highlowcontainer.appendCopyMany(x1.highlowcontainer, pos1, length1)
+	}
+	return answer
+}
+
+// In-place Or function that requires repairAfterLazy
+func (x1 *Bitmap) lazyOR(x2 *Bitmap) *Bitmap {
+	pos1 := 0
+	pos2 := 0
+	length1 := x1.highlowcontainer.size()
+	length2 := x2.highlowcontainer.size()
+main:
+	for (pos1 < length1) && (pos2 < length2) {
+		s1 := x1.highlowcontainer.getKeyAtIndex(pos1)
+		s2 := x2.highlowcontainer.getKeyAtIndex(pos2)
+
+		for {
+			if s1 < s2 {
+				pos1++
+				if pos1 == length1 {
+					break main
+				}
+				s1 = x1.highlowcontainer.getKeyAtIndex(pos1)
+			} else if s1 > s2 {
+				x1.highlowcontainer.insertNewKeyValueAt(pos1, s2, x2.highlowcontainer.getContainerAtIndex(pos2).clone())
+				pos2++
+				pos1++
+				length1++
+				if pos2 == length2 {
+					break main
+				}
+				s2 = x2.highlowcontainer.getKeyAtIndex(pos2)
+			} else {
+				c1 := x1.highlowcontainer.getContainerAtIndex(pos1)
+				switch t := c1.(type) {
+				case *arrayContainer:
+					c1 = t.toBitmapContainer()
+				case *runContainer16:
+					if !t.isFull() {
+						c1 = t.toBitmapContainer()
+					}
+				case *bitmapContainer:
+					c1 = x1.highlowcontainer.getWritableContainerAtIndex(pos1)
+				}
+
+				x1.highlowcontainer.containers[pos1] = c1.lazyIOR(x2.highlowcontainer.getContainerAtIndex(pos2))
+				x1.highlowcontainer.needCopyOnWrite[pos1] = false
+				pos1++
+				pos2++
+				if (pos1 == length1) || (pos2 == length2) {
+					break main
+				}
+				s1 = x1.highlowcontainer.getKeyAtIndex(pos1)
+				s2 = x2.highlowcontainer.getKeyAtIndex(pos2)
+			}
+		}
+	}
+	if pos1 == length1 {
+		x1.highlowcontainer.appendCopyMany(x2.highlowcontainer, pos2, length2)
+	}
+	return x1
+}
+
+// to be called after lazy aggregates
+func (x1 *Bitmap) repairAfterLazy() {
+	for pos := 0; pos < x1.highlowcontainer.size(); pos++ {
+		c := x1.highlowcontainer.getContainerAtIndex(pos)
+		switch c.(type) {
+		case *bitmapContainer:
+			if c.(*bitmapContainer).cardinality == invalidCardinality {
+				c = x1.highlowcontainer.getWritableContainerAtIndex(pos)
+				c.(*bitmapContainer).computeCardinality()
+				if c.(*bitmapContainer).getCardinality() <= arrayDefaultMaxSize {
+					x1.highlowcontainer.setContainerAtIndex(pos, c.(*bitmapContainer).toArrayContainer())
+				} else if c.(*bitmapContainer).isFull() {
+					x1.highlowcontainer.setContainerAtIndex(pos, newRunContainer16Range(0, MaxUint16))
+				}
+			}
+		}
+	}
+}
+
+// FastAnd computes the intersection between many bitmaps quickly
+// Compared to the And function, it can take many bitmaps as input, thus saving the trouble
+// of manually calling "And" many times.
+func FastAnd(bitmaps ...*Bitmap) *Bitmap {
+	if len(bitmaps) == 0 {
+		return NewBitmap()
+	} else if len(bitmaps) == 1 {
+		return bitmaps[0].Clone()
+	}
+	answer := And(bitmaps[0], bitmaps[1])
+	for _, bm := range bitmaps[2:] {
+		answer.And(bm)
+	}
+	return answer
+}
+
+// FastOr computes the union between many bitmaps quickly, as opposed to having to call Or repeatedly.
+// It might also be faster than calling Or repeatedly.
+func FastOr(bitmaps ...*Bitmap) *Bitmap {
+	if len(bitmaps) == 0 {
+		return NewBitmap()
+	} else if len(bitmaps) == 1 {
+		return bitmaps[0].Clone()
+	}
+	answer := lazyOR(bitmaps[0], bitmaps[1])
+	for _, bm := range bitmaps[2:] {
+		answer = answer.lazyOR(bm)
+	}
+	// here is where repairAfterLazy is called.
+	answer.repairAfterLazy()
+	return answer
+}
+
+// HeapOr computes the union between many bitmaps quickly using a heap.
+// It might be faster than calling Or repeatedly.
+func HeapOr(bitmaps ...*Bitmap) *Bitmap {
+	if len(bitmaps) == 0 {
+		return NewBitmap()
+	}
+	// TODO:  for better speed, we could do the operation lazily, see Java implementation
+	pq := make(priorityQueue, len(bitmaps))
+	for i, bm := range bitmaps {
+		pq[i] = &item{bm, i}
+	}
+	heap.Init(&pq)
+
+	for pq.Len() > 1 {
+		x1 := heap.Pop(&pq).(*item)
+		x2 := heap.Pop(&pq).(*item)
+		heap.Push(&pq, &item{Or(x1.value, x2.value), 0})
+	}
+	return heap.Pop(&pq).(*item).value
+}
+
+// HeapXor computes the symmetric difference between many bitmaps quickly (as opposed to calling Xor repeated).
+// Internally, this function uses a heap.
+// It might be faster than calling Xor repeatedly.
+func HeapXor(bitmaps ...*Bitmap) *Bitmap {
+	if len(bitmaps) == 0 {
+		return NewBitmap()
+	}
+
+	pq := make(priorityQueue, len(bitmaps))
+	for i, bm := range bitmaps {
+		pq[i] = &item{bm, i}
+	}
+	heap.Init(&pq)
+
+	for pq.Len() > 1 {
+		x1 := heap.Pop(&pq).(*item)
+		x2 := heap.Pop(&pq).(*item)
+		heap.Push(&pq, &item{Xor(x1.value, x2.value), 0})
+	}
+	return heap.Pop(&pq).(*item).value
+}

vendor/github.com/RoaringBitmap/roaring/manyiterator.go

@@ -0,0 +1,23 @@
+package roaring
+
+type manyIterable interface {
+	nextMany(hs uint32, buf []uint32) int
+}
+
+type manyIterator struct {
+	slice []uint16
+	loc   int
+}
+
+func (si *manyIterator) nextMany(hs uint32, buf []uint32) int {
+	n := 0
+	l := si.loc
+	s := si.slice
+	for n < len(buf) && l < len(s) {
+		buf[n] = uint32(s[l]) | hs
+		l++
+		n++
+	}
+	si.loc = l
+	return n
+}

vendor/github.com/RoaringBitmap/roaring/parallel.go

@@ -0,0 +1,613 @@
+package roaring
+
+import (
+	"container/heap"
+	"fmt"
+	"runtime"
+	"sync"
+)
+
+var defaultWorkerCount = runtime.NumCPU()
+
+type bitmapContainerKey struct {
+	key    uint16
+	idx    int
+	bitmap *Bitmap
+}
+
+type multipleContainers struct {
+	key        uint16
+	containers []container
+	idx        int
+}
+
+type keyedContainer struct {
+	key       uint16
+	container container
+	idx       int
+}
+
+type bitmapContainerHeap []bitmapContainerKey
+
+func (h bitmapContainerHeap) Len() int           { return len(h) }
+func (h bitmapContainerHeap) Less(i, j int) bool { return h[i].key < h[j].key }
+func (h bitmapContainerHeap) Swap(i, j int)      { h[i], h[j] = h[j], h[i] }
+
+func (h *bitmapContainerHeap) Push(x interface{}) {
+	// Push and Pop use pointer receivers because they modify the slice's length,
+	// not just its contents.
+	*h = append(*h, x.(bitmapContainerKey))
+}
+
+func (h *bitmapContainerHeap) Pop() interface{} {
+	old := *h
+	n := len(old)
+	x := old[n-1]
+	*h = old[0 : n-1]
+	return x
+}
+
+func (h bitmapContainerHeap) Peek() bitmapContainerKey {
+	return h[0]
+}
+
+func (h *bitmapContainerHeap) popIncrementing() (key uint16, container container) {
+	k := h.Peek()
+	key = k.key
+	container = k.bitmap.highlowcontainer.containers[k.idx]
+
+	newIdx := k.idx + 1
+	if newIdx < k.bitmap.highlowcontainer.size() {
+		k = bitmapContainerKey{
+			k.bitmap.highlowcontainer.keys[newIdx],
+			newIdx,
+			k.bitmap,
+		}
+		(*h)[0] = k
+		heap.Fix(h, 0)
+	} else {
+		heap.Pop(h)
+	}
+
+	return
+}
+
+func (h *bitmapContainerHeap) Next(containers []container) multipleContainers {
+	if h.Len() == 0 {
+		return multipleContainers{}
+	}
+
+	key, container := h.popIncrementing()
+	containers = append(containers, container)
+
+	for h.Len() > 0 && key == h.Peek().key {
+		_, container = h.popIncrementing()
+		containers = append(containers, container)
+	}
+
+	return multipleContainers{
+		key,
+		containers,
+		-1,
+	}
+}
+
+func newBitmapContainerHeap(bitmaps ...*Bitmap) bitmapContainerHeap {
+	// Initialize heap
+	var h bitmapContainerHeap = make([]bitmapContainerKey, 0, len(bitmaps))
+	for _, bitmap := range bitmaps {
+		if !bitmap.IsEmpty() {
+			key := bitmapContainerKey{
+				bitmap.highlowcontainer.keys[0],
+				0,
+				bitmap,
+			}
+			h = append(h, key)
+		}
+	}
+
+	heap.Init(&h)
+
+	return h
+}
+
+func repairAfterLazy(c container) container {
+	switch t := c.(type) {
+	case *bitmapContainer:
+		if t.cardinality == invalidCardinality {
+			t.computeCardinality()
+		}
+
+		if t.getCardinality() <= arrayDefaultMaxSize {
+			return t.toArrayContainer()
+		} else if c.(*bitmapContainer).isFull() {
+			return newRunContainer16Range(0, MaxUint16)
+		}
+	}
+
+	return c
+}
+
+func toBitmapContainer(c container) container {
+	switch t := c.(type) {
+	case *arrayContainer:
+		return t.toBitmapContainer()
+	case *runContainer16:
+		if !t.isFull() {
+			return t.toBitmapContainer()
+		}
+	}
+	return c
+}
+
+func appenderRoutine(bitmapChan chan<- *Bitmap, resultChan <-chan keyedContainer, expectedKeysChan <-chan int) {
+	expectedKeys := -1
+	appendedKeys := 0
+	keys := make([]uint16, 0)
+	containers := make([]container, 0)
+	for appendedKeys != expectedKeys {
+		select {
+		case item := <-resultChan:
+			if len(keys) <= item.idx {
+				keys = append(keys, make([]uint16, item.idx-len(keys)+1)...)
+				containers = append(containers, make([]container, item.idx-len(containers)+1)...)
+			}
+			keys[item.idx] = item.key
+			containers[item.idx] = item.container
+
+			appendedKeys++
+		case msg := <-expectedKeysChan:
+			expectedKeys = msg
+		}
+	}
+	answer := &Bitmap{
+		roaringArray{
+			make([]uint16, 0, expectedKeys),
+			make([]container, 0, expectedKeys),
+			make([]bool, 0, expectedKeys),
+			false,
+			nil,
+		},
+	}
+	for i := range keys {
+		if containers[i] != nil { // in case a resulting container was empty, see ParAnd function
+			answer.highlowcontainer.appendContainer(keys[i], containers[i], false)
+		}
+	}
+
+	bitmapChan <- answer
+}
+
+// ParHeapOr computes the union (OR) of all provided bitmaps in parallel,
+// where the parameter "parallelism" determines how many workers are to be used
+// (if it is set to 0, a default number of workers is chosen)
+// ParHeapOr uses a heap to compute the union. For rare cases it might be faster than ParOr
+func ParHeapOr(parallelism int, bitmaps ...*Bitmap) *Bitmap {
+
+	bitmapCount := len(bitmaps)
+	if bitmapCount == 0 {
+		return NewBitmap()
+	} else if bitmapCount == 1 {
+		return bitmaps[0].Clone()
+	}
+
+	if parallelism == 0 {
+		parallelism = defaultWorkerCount
+	}
+
+	h := newBitmapContainerHeap(bitmaps...)
+
+	bitmapChan := make(chan *Bitmap)
+	inputChan := make(chan multipleContainers, 128)
+	resultChan := make(chan keyedContainer, 32)
+	expectedKeysChan := make(chan int)
+
+	pool := sync.Pool{
+		New: func() interface{} {
+			return make([]container, 0, len(bitmaps))
+		},
+	}
+
+	orFunc := func() {
+		// Assumes only structs with >=2 containers are passed
+		for input := range inputChan {
+			c := toBitmapContainer(input.containers[0]).lazyOR(input.containers[1])
+			for _, next := range input.containers[2:] {
+				c = c.lazyIOR(next)
+			}
+			c = repairAfterLazy(c)
+			kx := keyedContainer{
+				input.key,
+				c,
+				input.idx,
+			}
+			resultChan <- kx
+			pool.Put(input.containers[:0])
+		}
+	}
+
+	go appenderRoutine(bitmapChan, resultChan, expectedKeysChan)
+
+	for i := 0; i < parallelism; i++ {
+		go orFunc()
+	}
+
+	idx := 0
+	for h.Len() > 0 {
+		ck := h.Next(pool.Get().([]container))
+		if len(ck.containers) == 1 {
+			resultChan <- keyedContainer{
+				ck.key,
+				ck.containers[0],
+				idx,
+			}
+			pool.Put(ck.containers[:0])
+		} else {
+			ck.idx = idx
+			inputChan <- ck
+		}
+		idx++
+	}
+	expectedKeysChan <- idx
+
+	bitmap := <-bitmapChan
+
+	close(inputChan)
+	close(resultChan)
+	close(expectedKeysChan)
+
+	return bitmap
+}
+
+// ParAnd computes the intersection (AND) of all provided bitmaps in parallel,
+// where the parameter "parallelism" determines how many workers are to be used
+// (if it is set to 0, a default number of workers is chosen)
+func ParAnd(parallelism int, bitmaps ...*Bitmap) *Bitmap {
+	bitmapCount := len(bitmaps)
+	if bitmapCount == 0 {
+		return NewBitmap()
+	} else if bitmapCount == 1 {
+		return bitmaps[0].Clone()
+	}
+
+	if parallelism == 0 {
+		parallelism = defaultWorkerCount
+	}
+
+	h := newBitmapContainerHeap(bitmaps...)
+
+	bitmapChan := make(chan *Bitmap)
+	inputChan := make(chan multipleContainers, 128)
+	resultChan := make(chan keyedContainer, 32)
+	expectedKeysChan := make(chan int)
+
+	andFunc := func() {
+		// Assumes only structs with >=2 containers are passed
+		for input := range inputChan {
+			c := input.containers[0].and(input.containers[1])
+			for _, next := range input.containers[2:] {
+				if c.getCardinality() == 0 {
+					break
+				}
+				c = c.iand(next)
+			}
+
+			// Send a nil explicitly if the result of the intersection is an empty container
+			if c.getCardinality() == 0 {
+				c = nil
+			}
+
+			kx := keyedContainer{
+				input.key,
+				c,
+				input.idx,
+			}
+			resultChan <- kx
+		}
+	}
+
+	go appenderRoutine(bitmapChan, resultChan, expectedKeysChan)
+
+	for i := 0; i < parallelism; i++ {
+		go andFunc()
+	}
+
+	idx := 0
+	for h.Len() > 0 {
+		ck := h.Next(make([]container, 0, 4))
+		if len(ck.containers) == bitmapCount {
+			ck.idx = idx
+			inputChan <- ck
+			idx++
+		}
+	}
+	expectedKeysChan <- idx
+
+	bitmap := <-bitmapChan
+
+	close(inputChan)
+	close(resultChan)
+	close(expectedKeysChan)
+
+	return bitmap
+}
+
+// ParOr computes the union (OR) of all provided bitmaps in parallel,
+// where the parameter "parallelism" determines how many workers are to be used
+// (if it is set to 0, a default number of workers is chosen)
+func ParOr(parallelism int, bitmaps ...*Bitmap) *Bitmap {
+	var lKey uint16 = MaxUint16
+	var hKey uint16 = 0
+
+	bitmapsFiltered := bitmaps[:0]
+	for _, b := range bitmaps {
+		if !b.IsEmpty() {
+			bitmapsFiltered = append(bitmapsFiltered, b)
+		}
+	}
+	bitmaps = bitmapsFiltered
+
+	for _, b := range bitmaps {
+		lKey = minOfUint16(lKey, b.highlowcontainer.keys[0])
+		hKey = maxOfUint16(hKey, b.highlowcontainer.keys[b.highlowcontainer.size()-1])
+	}
+
+	if lKey == MaxUint16 && hKey == 0 {
+		return New()
+	} else if len(bitmaps) == 1 {
+		return bitmaps[0]
+	}
+
+	keyRange := hKey - lKey + 1
+	if keyRange == 1 {
+		// revert to FastOr. Since the key range is 0
+		// no container-level aggregation parallelism is achievable
+		return FastOr(bitmaps...)
+	}
+
+	if parallelism == 0 {
+		parallelism = defaultWorkerCount
+	}
+
+	var chunkSize int
+	var chunkCount int
+	if parallelism*4 > int(keyRange) {
+		chunkSize = 1
+		chunkCount = int(keyRange)
+	} else {
+		chunkCount = parallelism * 4
+		chunkSize = (int(keyRange) + chunkCount - 1) / chunkCount
+	}
+
+	if chunkCount*chunkSize < int(keyRange) {
+		// it's fine to panic to indicate an implementation error
+		panic(fmt.Sprintf("invariant check failed: chunkCount * chunkSize < keyRange, %d * %d < %d", chunkCount, chunkSize, keyRange))
+	}
+
+	chunks := make([]*roaringArray, chunkCount)
+
+	chunkSpecChan := make(chan parChunkSpec, minOfInt(maxOfInt(64, 2*parallelism), int(chunkCount)))
+	chunkChan := make(chan parChunk, minOfInt(32, int(chunkCount)))
+
+	orFunc := func() {
+		for spec := range chunkSpecChan {
+			ra := lazyOrOnRange(&bitmaps[0].highlowcontainer, &bitmaps[1].highlowcontainer, spec.start, spec.end)
+			for _, b := range bitmaps[2:] {
+				ra = lazyIOrOnRange(ra, &b.highlowcontainer, spec.start, spec.end)
+			}
+
+			for i, c := range ra.containers {
+				ra.containers[i] = repairAfterLazy(c)
+			}
+
+			chunkChan <- parChunk{ra, spec.idx}
+		}
+	}
+
+	for i := 0; i < parallelism; i++ {
+		go orFunc()
+	}
+
+	go func() {
+		for i := 0; i < chunkCount; i++ {
+			spec := parChunkSpec{
+				start: uint16(int(lKey) + i*chunkSize),
+				end:   uint16(minOfInt(int(lKey)+(i+1)*chunkSize-1, int(hKey))),
+				idx:   int(i),
+			}
+			chunkSpecChan <- spec
+		}
+	}()
+
+	chunksRemaining := chunkCount
+	for chunk := range chunkChan {
+		chunks[chunk.idx] = chunk.ra
+		chunksRemaining--
+		if chunksRemaining == 0 {
+			break
+		}
+	}
+	close(chunkChan)
+	close(chunkSpecChan)
+
+	containerCount := 0
+	for _, chunk := range chunks {
+		containerCount += chunk.size()
+	}
+
+	result := Bitmap{
+		roaringArray{
+			containers:      make([]container, containerCount),
+			keys:            make([]uint16, containerCount),
+			needCopyOnWrite: make([]bool, containerCount),
+		},
+	}
+
+	resultOffset := 0
+	for _, chunk := range chunks {
+		copy(result.highlowcontainer.containers[resultOffset:], chunk.containers)
+		copy(result.highlowcontainer.keys[resultOffset:], chunk.keys)
+		copy(result.highlowcontainer.needCopyOnWrite[resultOffset:], chunk.needCopyOnWrite)
+		resultOffset += chunk.size()
+	}
+
+	return &result
+}
+
+type parChunkSpec struct {
+	start uint16
+	end   uint16
+	idx   int
+}
+
+type parChunk struct {
+	ra  *roaringArray
+	idx int
+}
+
+func (c parChunk) size() int {
+	return c.ra.size()
+}
+
+func parNaiveStartAt(ra *roaringArray, start uint16, last uint16) int {
+	for idx, key := range ra.keys {
+		if key >= start && key <= last {
+			return idx
+		} else if key > last {
+			break
+		}
+	}
+	return ra.size()
+}
+
+func lazyOrOnRange(ra1, ra2 *roaringArray, start, last uint16) *roaringArray {
+	answer := newRoaringArray()
+	length1 := ra1.size()
+	length2 := ra2.size()
+
+	idx1 := parNaiveStartAt(ra1, start, last)
+	idx2 := parNaiveStartAt(ra2, start, last)
+
+	var key1 uint16
+	var key2 uint16
+	if idx1 < length1 && idx2 < length2 {
+		key1 = ra1.getKeyAtIndex(idx1)
+		key2 = ra2.getKeyAtIndex(idx2)
+
+		for key1 <= last && key2 <= last {
+
+			if key1 < key2 {
+				answer.appendCopy(*ra1, idx1)
+				idx1++
+				if idx1 == length1 {
+					break
+				}
+				key1 = ra1.getKeyAtIndex(idx1)
+			} else if key1 > key2 {
+				answer.appendCopy(*ra2, idx2)
+				idx2++
+				if idx2 == length2 {
+					break
+				}
+				key2 = ra2.getKeyAtIndex(idx2)
+			} else {
+				c1 := ra1.getFastContainerAtIndex(idx1, false)
+
+				answer.appendContainer(key1, c1.lazyOR(ra2.getContainerAtIndex(idx2)), false)
+				idx1++
+				idx2++
+				if idx1 == length1 || idx2 == length2 {
+					break
+				}
+
+				key1 = ra1.getKeyAtIndex(idx1)
+				key2 = ra2.getKeyAtIndex(idx2)
+			}
+		}
+	}
+
+	if idx2 < length2 {
+		key2 = ra2.getKeyAtIndex(idx2)
+		for key2 <= last {
+			answer.appendCopy(*ra2, idx2)
+			idx2++
+			if idx2 == length2 {
+				break
+			}
+			key2 = ra2.getKeyAtIndex(idx2)
+		}
+	}
+
+	if idx1 < length1 {
+		key1 = ra1.getKeyAtIndex(idx1)
+		for key1 <= last {
+			answer.appendCopy(*ra1, idx1)
+			idx1++
+			if idx1 == length1 {
+				break
+			}
+			key1 = ra1.getKeyAtIndex(idx1)
+		}
+	}
+	return answer
+}
+
+func lazyIOrOnRange(ra1, ra2 *roaringArray, start, last uint16) *roaringArray {
+	length1 := ra1.size()
+	length2 := ra2.size()
+
+	idx1 := 0
+	idx2 := parNaiveStartAt(ra2, start, last)
+
+	var key1 uint16
+	var key2 uint16
+	if idx1 < length1 && idx2 < length2 {
+		key1 = ra1.getKeyAtIndex(idx1)
+		key2 = ra2.getKeyAtIndex(idx2)
+
+		for key1 <= last && key2 <= last {
+			if key1 < key2 {
+				idx1++
+				if idx1 >= length1 {
+					break
+				}
+				key1 = ra1.getKeyAtIndex(idx1)
+			} else if key1 > key2 {
+				ra1.insertNewKeyValueAt(idx1, key2, ra2.getContainerAtIndex(idx2))
+				ra1.needCopyOnWrite[idx1] = true
+				idx2++
+				idx1++
+				length1++
+				if idx2 >= length2 {
+					break
+				}
+				key2 = ra2.getKeyAtIndex(idx2)
+			} else {
+				c1 := ra1.getFastContainerAtIndex(idx1, true)
+
+				ra1.containers[idx1] = c1.lazyIOR(ra2.getContainerAtIndex(idx2))
+				ra1.needCopyOnWrite[idx1] = false
+				idx1++
+				idx2++
+				if idx1 >= length1 || idx2 >= length2 {
+					break
+				}
+
+				key1 = ra1.getKeyAtIndex(idx1)
+				key2 = ra2.getKeyAtIndex(idx2)
+			}
+		}
+	}
+	if idx2 < length2 {
+		key2 = ra2.getKeyAtIndex(idx2)
+		for key2 <= last {
+			ra1.appendCopy(*ra2, idx2)
+			idx2++
+			if idx2 >= length2 {
+				break
+			}
+			key2 = ra2.getKeyAtIndex(idx2)
+		}
+	}
+	return ra1
+}

vendor/github.com/RoaringBitmap/roaring/popcnt.go

@@ -0,0 +1,11 @@
+// +build go1.9
+// "go1.9", from Go version 1.9 onward
+// See https://golang.org/pkg/go/build/#hdr-Build_Constraints
+
+package roaring
+
+import "math/bits"
+
+func popcount(x uint64) uint64 {
+	return uint64(bits.OnesCount64(x))
+}

vendor/github.com/RoaringBitmap/roaring/popcnt_amd64.s

@@ -0,0 +1,103 @@
+// +build amd64,!appengine,!go1.9
+
+TEXT ·hasAsm(SB),4,$0-1
+MOVQ $1, AX
+CPUID
+SHRQ $23, CX
+ANDQ $1, CX
+MOVB CX, ret+0(FP)
+RET
+
+#define POPCNTQ_DX_DX BYTE $0xf3; BYTE $0x48; BYTE $0x0f; BYTE $0xb8; BYTE $0xd2
+
+TEXT ·popcntSliceAsm(SB),4,$0-32
+XORQ	AX, AX
+MOVQ	s+0(FP), SI
+MOVQ	s_len+8(FP), CX
+TESTQ	CX, CX
+JZ		popcntSliceEnd
+popcntSliceLoop:
+BYTE $0xf3; BYTE $0x48; BYTE $0x0f; BYTE $0xb8; BYTE $0x16 // POPCNTQ (SI), DX
+ADDQ	DX, AX
+ADDQ	$8, SI
+LOOP	popcntSliceLoop
+popcntSliceEnd:
+MOVQ	AX, ret+24(FP)
+RET
+
+TEXT ·popcntMaskSliceAsm(SB),4,$0-56
+XORQ	AX, AX
+MOVQ	s+0(FP), SI
+MOVQ	s_len+8(FP), CX
+TESTQ	CX, CX
+JZ		popcntMaskSliceEnd
+MOVQ	m+24(FP), DI
+popcntMaskSliceLoop:
+MOVQ	(DI), DX
+NOTQ	DX
+ANDQ	(SI), DX
+POPCNTQ_DX_DX
+ADDQ	DX, AX
+ADDQ	$8, SI
+ADDQ	$8, DI
+LOOP	popcntMaskSliceLoop
+popcntMaskSliceEnd:
+MOVQ	AX, ret+48(FP)
+RET
+
+TEXT ·popcntAndSliceAsm(SB),4,$0-56
+XORQ	AX, AX
+MOVQ	s+0(FP), SI
+MOVQ	s_len+8(FP), CX
+TESTQ	CX, CX
+JZ		popcntAndSliceEnd
+MOVQ	m+24(FP), DI
+popcntAndSliceLoop:
+MOVQ	(DI), DX
+ANDQ	(SI), DX
+POPCNTQ_DX_DX
+ADDQ	DX, AX
+ADDQ	$8, SI
+ADDQ	$8, DI
+LOOP	popcntAndSliceLoop
+popcntAndSliceEnd:
+MOVQ	AX, ret+48(FP)
+RET
+
+TEXT ·popcntOrSliceAsm(SB),4,$0-56
+XORQ	AX, AX
+MOVQ	s+0(FP), SI
+MOVQ	s_len+8(FP), CX
+TESTQ	CX, CX
+JZ		popcntOrSliceEnd
+MOVQ	m+24(FP), DI
+popcntOrSliceLoop:
+MOVQ	(DI), DX
+ORQ		(SI), DX
+POPCNTQ_DX_DX
+ADDQ	DX, AX
+ADDQ	$8, SI
+ADDQ	$8, DI
+LOOP	popcntOrSliceLoop
+popcntOrSliceEnd:
+MOVQ	AX, ret+48(FP)
+RET
+
+TEXT ·popcntXorSliceAsm(SB),4,$0-56
+XORQ	AX, AX
+MOVQ	s+0(FP), SI
+MOVQ	s_len+8(FP), CX
+TESTQ	CX, CX
+JZ		popcntXorSliceEnd
+MOVQ	m+24(FP), DI
+popcntXorSliceLoop:
+MOVQ	(DI), DX
+XORQ	(SI), DX
+POPCNTQ_DX_DX
+ADDQ	DX, AX
+ADDQ	$8, SI
+ADDQ	$8, DI
+LOOP	popcntXorSliceLoop
+popcntXorSliceEnd:
+MOVQ	AX, ret+48(FP)
+RET

vendor/github.com/RoaringBitmap/roaring/popcnt_asm.go

@@ -0,0 +1,67 @@
+// +build amd64,!appengine,!go1.9
+
+package roaring
+
+// *** the following functions are defined in popcnt_amd64.s
+
+//go:noescape
+
+func hasAsm() bool
+
+// useAsm is a flag used to select the GO or ASM implementation of the popcnt function
+var useAsm = hasAsm()
+
+//go:noescape
+
+func popcntSliceAsm(s []uint64) uint64
+
+//go:noescape
+
+func popcntMaskSliceAsm(s, m []uint64) uint64
+
+//go:noescape
+
+func popcntAndSliceAsm(s, m []uint64) uint64
+
+//go:noescape
+
+func popcntOrSliceAsm(s, m []uint64) uint64
+
+//go:noescape
+
+func popcntXorSliceAsm(s, m []uint64) uint64
+
+func popcntSlice(s []uint64) uint64 {
+	if useAsm {
+		return popcntSliceAsm(s)
+	}
+	return popcntSliceGo(s)
+}
+
+func popcntMaskSlice(s, m []uint64) uint64 {
+	if useAsm {
+		return popcntMaskSliceAsm(s, m)
+	}
+	return popcntMaskSliceGo(s, m)
+}
+
+func popcntAndSlice(s, m []uint64) uint64 {
+	if useAsm {
+		return popcntAndSliceAsm(s, m)
+	}
+	return popcntAndSliceGo(s, m)
+}
+
+func popcntOrSlice(s, m []uint64) uint64 {
+	if useAsm {
+		return popcntOrSliceAsm(s, m)
+	}
+	return popcntOrSliceGo(s, m)
+}
+
+func popcntXorSlice(s, m []uint64) uint64 {
+	if useAsm {
+		return popcntXorSliceAsm(s, m)
+	}
+	return popcntXorSliceGo(s, m)
+}

vendor/github.com/RoaringBitmap/roaring/popcnt_compat.go

@@ -0,0 +1,17 @@
+// +build !go1.9
+
+package roaring
+
+// bit population count, take from
+// https://code.google.com/p/go/issues/detail?id=4988#c11
+// credit: https://code.google.com/u/arnehormann/
+// credit: https://play.golang.org/p/U7SogJ7psJ
+// credit: http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel
+func popcount(x uint64) uint64 {
+	x -= (x >> 1) & 0x5555555555555555
+	x = (x>>2)&0x3333333333333333 + x&0x3333333333333333
+	x += x >> 4
+	x &= 0x0f0f0f0f0f0f0f0f
+	x *= 0x0101010101010101
+	return x >> 56
+}

vendor/github.com/RoaringBitmap/roaring/popcnt_generic.go

@@ -0,0 +1,23 @@
+// +build !amd64 appengine go1.9
+
+package roaring
+
+func popcntSlice(s []uint64) uint64 {
+	return popcntSliceGo(s)
+}
+
+func popcntMaskSlice(s, m []uint64) uint64 {
+	return popcntMaskSliceGo(s, m)
+}
+
+func popcntAndSlice(s, m []uint64) uint64 {
+	return popcntAndSliceGo(s, m)
+}
+
+func popcntOrSlice(s, m []uint64) uint64 {
+	return popcntOrSliceGo(s, m)
+}
+
+func popcntXorSlice(s, m []uint64) uint64 {
+	return popcntXorSliceGo(s, m)
+}

vendor/github.com/RoaringBitmap/roaring/popcnt_slices.go

@@ -0,0 +1,41 @@
+package roaring
+
+func popcntSliceGo(s []uint64) uint64 {
+	cnt := uint64(0)
+	for _, x := range s {
+		cnt += popcount(x)
+	}
+	return cnt
+}
+
+func popcntMaskSliceGo(s, m []uint64) uint64 {
+	cnt := uint64(0)
+	for i := range s {
+		cnt += popcount(s[i] &^ m[i])
+	}
+	return cnt
+}
+
+func popcntAndSliceGo(s, m []uint64) uint64 {
+	cnt := uint64(0)
+	for i := range s {
+		cnt += popcount(s[i] & m[i])
+	}
+	return cnt
+}
+
+func popcntOrSliceGo(s, m []uint64) uint64 {
+	cnt := uint64(0)
+	for i := range s {
+		cnt += popcount(s[i] | m[i])
+	}
+	return cnt
+}
+
+func popcntXorSliceGo(s, m []uint64) uint64 {
+	cnt := uint64(0)
+	for i := range s {
+		cnt += popcount(s[i] ^ m[i])
+	}
+	return cnt
+}

vendor/github.com/RoaringBitmap/roaring/priorityqueue.go

@@ -0,0 +1,101 @@
+package roaring
+
+import "container/heap"
+
+/////////////
+// The priorityQueue is used to keep Bitmaps sorted.
+////////////
+
+type item struct {
+	value *Bitmap
+	index int
+}
+
+type priorityQueue []*item
+
+func (pq priorityQueue) Len() int { return len(pq) }
+
+func (pq priorityQueue) Less(i, j int) bool {
+	return pq[i].value.GetSizeInBytes() < pq[j].value.GetSizeInBytes()
+}
+
+func (pq priorityQueue) Swap(i, j int) {
+	pq[i], pq[j] = pq[j], pq[i]
+	pq[i].index = i
+	pq[j].index = j
+}
+
+func (pq *priorityQueue) Push(x interface{}) {
+	n := len(*pq)
+	item := x.(*item)
+	item.index = n
+	*pq = append(*pq, item)
+}
+
+func (pq *priorityQueue) Pop() interface{} {
+	old := *pq
+	n := len(old)
+	item := old[n-1]
+	item.index = -1 // for safety
+	*pq = old[0 : n-1]
+	return item
+}
+
+func (pq *priorityQueue) update(item *item, value *Bitmap) {
+	item.value = value
+	heap.Fix(pq, item.index)
+}
+
+/////////////
+// The containerPriorityQueue is used to keep the containers of various Bitmaps sorted.
+////////////
+
+type containeritem struct {
+	value    *Bitmap
+	keyindex int
+	index    int
+}
+
+type containerPriorityQueue []*containeritem
+
+func (pq containerPriorityQueue) Len() int { return len(pq) }
+
+func (pq containerPriorityQueue) Less(i, j int) bool {
+	k1 := pq[i].value.highlowcontainer.getKeyAtIndex(pq[i].keyindex)
+	k2 := pq[j].value.highlowcontainer.getKeyAtIndex(pq[j].keyindex)
+	if k1 != k2 {
+		return k1 < k2
+	}
+	c1 := pq[i].value.highlowcontainer.getContainerAtIndex(pq[i].keyindex)
+	c2 := pq[j].value.highlowcontainer.getContainerAtIndex(pq[j].keyindex)
+
+	return c1.getCardinality() > c2.getCardinality()
+}
+
+func (pq containerPriorityQueue) Swap(i, j int) {
+	pq[i], pq[j] = pq[j], pq[i]
+	pq[i].index = i
+	pq[j].index = j
+}
+
+func (pq *containerPriorityQueue) Push(x interface{}) {
+	n := len(*pq)
+	item := x.(*containeritem)
+	item.index = n
+	*pq = append(*pq, item)
+}
+
+func (pq *containerPriorityQueue) Pop() interface{} {
+	old := *pq
+	n := len(old)
+	item := old[n-1]
+	item.index = -1 // for safety
+	*pq = old[0 : n-1]
+	return item
+}
+
+//func (pq *containerPriorityQueue) update(item *containeritem, value *Bitmap, keyindex int) {
+//	item.value = value
+//	item.keyindex = keyindex
+//	heap.Fix(pq, item.index)
+//}

vendor/github.com/RoaringBitmap/roaring/rlecommon.go

@@ -0,0 +1,163 @@
+package roaring
+
+import (
+	"fmt"
+)
+
+// common to rle32.go and rle16.go
+
+// rleVerbose controls whether p() prints show up.
+// The testing package sets this based on
+// testing.Verbose().
+var rleVerbose bool
+
+// p is a shorthand for fmt.Printf with beginning and
+// trailing newlines. p() makes it easy
+// to add diagnostic print statements.
+func p(format string, args ...interface{}) {
+	if rleVerbose {
+		fmt.Printf("\n"+format+"\n", args...)
+	}
+}
+
+// MaxUint32 is the largest uint32 value.
+const MaxUint32 = 4294967295
+
+// MaxUint16 is the largest 16 bit unsigned int.
+// This is the largest value an interval16 can store.
+const MaxUint16 = 65535
+
+// searchOptions allows us to accelerate runContainer32.search with
+// prior knowledge of (mostly lower) bounds. This is used by Union
+// and Intersect.
+type searchOptions struct {
+	// start here instead of at 0
+	startIndex int64
+
+	// upper bound instead of len(rc.iv);
+	// endxIndex == 0 means ignore the bound and use
+	// endxIndex == n ==len(rc.iv) which is also
+	// naturally the default for search()
+	// when opt = nil.
+	endxIndex int64
+}
+
+// And finds the intersection of rc and b.
+func (rc *runContainer32) And(b *Bitmap) *Bitmap {
+	out := NewBitmap()
+	for _, p := range rc.iv {
+		for i := p.start; i <= p.last; i++ {
+			if b.Contains(i) {
+				out.Add(i)
+			}
+		}
+	}
+	return out
+}
+
+// Xor returns the exclusive-or of rc and b.
+func (rc *runContainer32) Xor(b *Bitmap) *Bitmap {
+	out := b.Clone()
+	for _, p := range rc.iv {
+		for v := p.start; v <= p.last; v++ {
+			if out.Contains(v) {
+				out.RemoveRange(uint64(v), uint64(v+1))
+			} else {
+				out.Add(v)
+			}
+		}
+	}
+	return out
+}
+
+// Or returns the union of rc and b.
+func (rc *runContainer32) Or(b *Bitmap) *Bitmap {
+	out := b.Clone()
+	for _, p := range rc.iv {
+		for v := p.start; v <= p.last; v++ {
+			out.Add(v)
+		}
+	}
+	return out
+}
+
+// trial is used in the randomized testing of runContainers
+type trial struct {
+	n           int
+	percentFill float64
+	ntrial      int
+
+	// only in the union test
+	// only subtract test
+	percentDelete float64
+
+	// only in 067 randomized operations
+	// we do this + 1 passes
+	numRandomOpsPass int
+
+	// allow sampling range control
+	// only recent tests respect this.
+	srang *interval16
+}
+
+// And finds the intersection of rc and b.
+func (rc *runContainer16) And(b *Bitmap) *Bitmap {
+	out := NewBitmap()
+	for _, p := range rc.iv {
+		plast := p.last()
+		for i := p.start; i <= plast; i++ {
+			if b.Contains(uint32(i)) {
+				out.Add(uint32(i))
+			}
+		}
+	}
+	return out
+}
+
+// Xor returns the exclusive-or of rc and b.
+func (rc *runContainer16) Xor(b *Bitmap) *Bitmap {
+	out := b.Clone()
+	for _, p := range rc.iv {
+		plast := p.last()
+		for v := p.start; v <= plast; v++ {
+			w := uint32(v)
+			if out.Contains(w) {
+				out.RemoveRange(uint64(w), uint64(w+1))
+			} else {
+				out.Add(w)
+			}
+		}
+	}
+	return out
+}
+
+// Or returns the union of rc and b.
+func (rc *runContainer16) Or(b *Bitmap) *Bitmap {
+	out := b.Clone()
+	for _, p := range rc.iv {
+		plast := p.last()
+		for v := p.start; v <= plast; v++ {
+			out.Add(uint32(v))
+		}
+	}
+	return out
+}
+
+//func (rc *runContainer32) and(container) container {
+//	panic("TODO. not yet implemented")
+//}
+
+// serializedSizeInBytes returns the number of bytes of memory
+// required by this runContainer16. This is for the
+// Roaring format, as specified https://github.com/RoaringBitmap/RoaringFormatSpec/
+func (rc *runContainer16) serializedSizeInBytes() int {
+	// number of runs in one uint16, then each run
+	// needs two more uint16
+	return 2 + len(rc.iv)*4
+}
+
+// serializedSizeInBytes returns the number of bytes of memory
+// required by this runContainer32.
+func (rc *runContainer32) serializedSizeInBytes() int {
+	return 4 + len(rc.iv)*8
+}

vendor/github.com/RoaringBitmap/roaring/rlei.go

@@ -0,0 +1,695 @@
+package roaring
+
+///////////////////////////////////////////////////
+//
+// container interface methods for runContainer16
+//
+///////////////////////////////////////////////////
+
+import (
+	"fmt"
+)
+
+// compile time verify we meet interface requirements
+var _ container = &runContainer16{}
+
+func (rc *runContainer16) clone() container {
+	return newRunContainer16CopyIv(rc.iv)
+}
+
+func (rc *runContainer16) minimum() uint16 {
+	return rc.iv[0].start // assume not empty
+}
+
+func (rc *runContainer16) maximum() uint16 {
+	return rc.iv[len(rc.iv)-1].last() // assume not empty
+}
+
+func (rc *runContainer16) isFull() bool {
+	return (len(rc.iv) == 1) && ((rc.iv[0].start == 0) && (rc.iv[0].last() == MaxUint16))
+}
+
+func (rc *runContainer16) and(a container) container {
+	if rc.isFull() {
+		return a.clone()
+	}
+	switch c := a.(type) {
+	case *runContainer16:
+		return rc.intersect(c)
+	case *arrayContainer:
+		return rc.andArray(c)
+	case *bitmapContainer:
+		return rc.andBitmapContainer(c)
+	}
+	panic("unsupported container type")
+}
+
+func (rc *runContainer16) andCardinality(a container) int {
+	switch c := a.(type) {
+	case *runContainer16:
+		return int(rc.intersectCardinality(c))
+	case *arrayContainer:
+		return rc.andArrayCardinality(c)
+	case *bitmapContainer:
+		return rc.andBitmapContainerCardinality(c)
+	}
+	panic("unsupported container type")
+}
+
+// andBitmapContainer finds the intersection of rc and b.
+func (rc *runContainer16) andBitmapContainer(bc *bitmapContainer) container {
+	bc2 := newBitmapContainerFromRun(rc)
+	return bc2.andBitmap(bc)
+}
+
+func (rc *runContainer16) andArrayCardinality(ac *arrayContainer) int {
+	pos := 0
+	answer := 0
+	maxpos := ac.getCardinality()
+	if maxpos == 0 {
+		return 0 // won't happen in actual code
+	}
+	v := ac.content[pos]
+mainloop:
+	for _, p := range rc.iv {
+		for v < p.start {
+			pos++
+			if pos == maxpos {
+				break mainloop
+			}
+			v = ac.content[pos]
+		}
+		for v <= p.last() {
+			answer++
+			pos++
+			if pos == maxpos {
+				break mainloop
+			}
+			v = ac.content[pos]
+		}
+	}
+	return answer
+}
+
+func (rc *runContainer16) iand(a container) container {
+	if rc.isFull() {
+		return a.clone()
+	}
+	switch c := a.(type) {
+	case *runContainer16:
+		return rc.inplaceIntersect(c)
+	case *arrayContainer:
+		return rc.andArray(c)
+	case *bitmapContainer:
+		return rc.iandBitmapContainer(c)
+	}
+	panic("unsupported container type")
+}
+
+func (rc *runContainer16) inplaceIntersect(rc2 *runContainer16) container {
+	// TODO: optimize by doing less allocation, possibly?
+
+	// sect will be new
+	sect := rc.intersect(rc2)
+	*rc = *sect
+	return rc
+}
+
+func (rc *runContainer16) iandBitmapContainer(bc *bitmapContainer) container {
+	isect := rc.andBitmapContainer(bc)
+	*rc = *newRunContainer16FromContainer(isect)
+	return rc
+}
+
+func (rc *runContainer16) andArray(ac *arrayContainer) container {
+	if len(rc.iv) == 0 {
+		return newArrayContainer()
+	}
+
+	acCardinality := ac.getCardinality()
+	c := newArrayContainerCapacity(acCardinality)
+
+	for rlePos, arrayPos := 0, 0; arrayPos < acCardinality; {
+		iv := rc.iv[rlePos]
+		arrayVal := ac.content[arrayPos]
+
+		for iv.last() < arrayVal {
+			rlePos++
+			if rlePos == len(rc.iv) {
+				return c
+			}
+			iv = rc.iv[rlePos]
+		}
+
+		if iv.start > arrayVal {
+			arrayPos = advanceUntil(ac.content, arrayPos, len(ac.content), iv.start)
+		} else {
+			c.content = append(c.content, arrayVal)
+			arrayPos++
+		}
+	}
+	return c
+}
+
+func (rc *runContainer16) andNot(a container) container {
+	switch c := a.(type) {
+	case *arrayContainer:
+		return rc.andNotArray(c)
+	case *bitmapContainer:
+		return rc.andNotBitmap(c)
+	case *runContainer16:
+		return rc.andNotRunContainer16(c)
+	}
+	panic("unsupported container type")
+}
+
+func (rc *runContainer16) fillLeastSignificant16bits(x []uint32, i int, mask uint32) {
+	k := 0
+	var val int64
+	for _, p := range rc.iv {
+		n := p.runlen()
+		for j := int64(0); j < n; j++ {
+			val = int64(p.start) + j
+			x[k+i] = uint32(val) | mask
+			k++
+		}
+	}
+}
+
+func (rc *runContainer16) getShortIterator() shortIterable {
+	return rc.newRunIterator16()
+}
+
+func (rc *runContainer16) getManyIterator() manyIterable {
+	return rc.newManyRunIterator16()
+}
+
+// add the values in the range [firstOfRange, endx). endx
+// is still abe to express 2^16 because it is an int not an uint16.
+func (rc *runContainer16) iaddRange(firstOfRange, endx int) container {
+
+	if firstOfRange >= endx {
+		panic(fmt.Sprintf("invalid %v = endx >= firstOfRange", endx))
+	}
+	addme := newRunContainer16TakeOwnership([]interval16{
+		{
+			start:  uint16(firstOfRange),
+			length: uint16(endx - 1 - firstOfRange),
+		},
+	})
+	*rc = *rc.union(addme)
+	return rc
+}
+
+// remove the values in the range [firstOfRange,endx)
+func (rc *runContainer16) iremoveRange(firstOfRange, endx int) container {
+	if firstOfRange >= endx {
+		panic(fmt.Sprintf("request to iremove empty set [%v, %v),"+
+			" nothing to do.", firstOfRange, endx))
+		//return rc
+	}
+	x := newInterval16Range(uint16(firstOfRange), uint16(endx-1))
+	rc.isubtract(x)
+	return rc
+}
+
+// not flip the values in the range [firstOfRange,endx)
+func (rc *runContainer16) not(firstOfRange, endx int) container {
+	if firstOfRange >= endx {
+		panic(fmt.Sprintf("invalid %v = endx >= firstOfRange = %v", endx, firstOfRange))
+	}
+
+	return rc.Not(firstOfRange, endx)
+}
+
+// Not flips the values in the range [firstOfRange,endx).
+// This is not inplace. Only the returned value has the flipped bits.
+//
+// Currently implemented as (!A intersect B) union (A minus B),
+// where A is rc, and B is the supplied [firstOfRange, endx) interval.
+//
+// TODO(time optimization): convert this to a single pass
+// algorithm by copying AndNotRunContainer16() and modifying it.
+// Current routine is correct but
+// makes 2 more passes through the arrays than should be
+// strictly necessary. Measure both ways though--this may not matter.
+//
+func (rc *runContainer16) Not(firstOfRange, endx int) *runContainer16 {
+
+	if firstOfRange >= endx {
+		panic(fmt.Sprintf("invalid %v = endx >= firstOfRange == %v", endx, firstOfRange))
+	}
+
+	if firstOfRange >= endx {
+		return rc.Clone()
+	}
+
+	a := rc
+	// algo:
+	// (!A intersect B) union (A minus B)
+
+	nota := a.invert()
+
+	bs := []interval16{newInterval16Range(uint16(firstOfRange), uint16(endx-1))}
+	b := newRunContainer16TakeOwnership(bs)
+
+	notAintersectB := nota.intersect(b)
+
+	aMinusB := a.AndNotRunContainer16(b)
+
+	rc2 := notAintersectB.union(aMinusB)
+	return rc2
+}
+
+// equals is now logical equals; it does not require the
+// same underlying container type.
+func (rc *runContainer16) equals(o container) bool {
+	srb, ok := o.(*runContainer16)
+
+	if !ok {
+		// maybe value instead of pointer
+		val, valok := o.(*runContainer16)
+		if valok {
+			srb = val
+			ok = true
+		}
+	}
+	if ok {
+		// Check if the containers are the same object.
+		if rc == srb {
+			return true
+		}
+
+		if len(srb.iv) != len(rc.iv) {
+			return false
+		}
+
+		for i, v := range rc.iv {
+			if v != srb.iv[i] {
+				return false
+			}
+		}
+		return true
+	}
+
+	// use generic comparison
+	if o.getCardinality() != rc.getCardinality() {
+		return false
+	}
+	rit := rc.getShortIterator()
+	bit := o.getShortIterator()
+
+	//k := 0
+	for rit.hasNext() {
+		if bit.next() != rit.next() {
+			return false
+		}
+		//k++
+	}
+	return true
+}
+
+func (rc *runContainer16) iaddReturnMinimized(x uint16) container {
+	rc.Add(x)
+	return rc
+}
+
+func (rc *runContainer16) iadd(x uint16) (wasNew bool) {
+	return rc.Add(x)
+}
+
+func (rc *runContainer16) iremoveReturnMinimized(x uint16) container {
+	rc.removeKey(x)
+	return rc
+}
+
+func (rc *runContainer16) iremove(x uint16) bool {
+	return rc.removeKey(x)
+}
+
+func (rc *runContainer16) or(a container) container {
+	if rc.isFull() {
+		return rc.clone()
+	}
+	switch c := a.(type) {
+	case *runContainer16:
+		return rc.union(c)
+	case *arrayContainer:
+		return rc.orArray(c)
+	case *bitmapContainer:
+		return rc.orBitmapContainer(c)
+	}
+	panic("unsupported container type")
+}
+
+func (rc *runContainer16) orCardinality(a container) int {
+	switch c := a.(type) {
+	case *runContainer16:
+		return int(rc.unionCardinality(c))
+	case *arrayContainer:
+		return rc.orArrayCardinality(c)
+	case *bitmapContainer:
+		return rc.orBitmapContainerCardinality(c)
+	}
+	panic("unsupported container type")
+}
+
+// orBitmapContainer finds the union of rc and bc.
+func (rc *runContainer16) orBitmapContainer(bc *bitmapContainer) container {
+	bc2 := newBitmapContainerFromRun(rc)
+	return bc2.iorBitmap(bc)
+}
+
+func (rc *runContainer16) andBitmapContainerCardinality(bc *bitmapContainer) int {
+	answer := 0
+	for i := range rc.iv {
+		answer += bc.getCardinalityInRange(uint(rc.iv[i].start), uint(rc.iv[i].last())+1)
+	}
+	//bc.computeCardinality()
+	return answer
+}
+
+func (rc *runContainer16) orBitmapContainerCardinality(bc *bitmapContainer) int {
+	return rc.getCardinality() + bc.getCardinality() - rc.andBitmapContainerCardinality(bc)
+}
+
+// orArray finds the union of rc and ac.
+func (rc *runContainer16) orArray(ac *arrayContainer) container {
+	bc1 := newBitmapContainerFromRun(rc)
+	bc2 := ac.toBitmapContainer()
+	return bc1.orBitmap(bc2)
+}
+
+// orArray finds the union of rc and ac.
+func (rc *runContainer16) orArrayCardinality(ac *arrayContainer) int {
+	return ac.getCardinality() + rc.getCardinality() - rc.andArrayCardinality(ac)
+}
+
+func (rc *runContainer16) ior(a container) container {
+	if rc.isFull() {
+		return rc
+	}
+	switch c := a.(type) {
+	case *runContainer16:
+		return rc.inplaceUnion(c)
+	case *arrayContainer:
+		return rc.iorArray(c)
+	case *bitmapContainer:
+		return rc.iorBitmapContainer(c)
+	}
+	panic("unsupported container type")
+}
+
+func (rc *runContainer16) inplaceUnion(rc2 *runContainer16) container {
+	p("rc.inplaceUnion with len(rc2.iv)=%v", len(rc2.iv))
+	for _, p := range rc2.iv {
+		last := int64(p.last())
+		for i := int64(p.start); i <= last; i++ {
+			rc.Add(uint16(i))
+		}
+	}
+	return rc
+}
+
+func (rc *runContainer16) iorBitmapContainer(bc *bitmapContainer) container {
+
+	it := bc.getShortIterator()
+	for it.hasNext() {
+		rc.Add(it.next())
+	}
+	return rc
+}
+
+func (rc *runContainer16) iorArray(ac *arrayContainer) container {
+	it := ac.getShortIterator()
+	for it.hasNext() {
+		rc.Add(it.next())
+	}
+	return rc
+}
+
+// lazyIOR is described (not yet implemented) in
+// this nice note from @lemire on
+// https://github.com/RoaringBitmap/roaring/pull/70#issuecomment-263613737
+//
+// Description of lazyOR and lazyIOR from @lemire:
+//
+// Lazy functions are optional and can be simply
+// wrapper around non-lazy functions.
+//
+// The idea of "laziness" is as follows. It is
+// inspired by the concept of lazy evaluation
+// you might be familiar with (functional programming
+// and all that). So a roaring bitmap is
+// such that all its containers are, in some
+// sense, chosen to use as little memory as
+// possible. This is nice. Also, all bitsets
+// are "cardinality aware" so that you can do
+// fast rank/select queries, or query the
+// cardinality of the whole bitmap... very fast,
+// without latency.
+//
+// However, imagine that you are aggregating 100
+// bitmaps together. So you OR the first two, then OR
+// that with the third one and so forth. Clearly,
+// intermediate bitmaps don't need to be as
+// compressed as possible, right? They can be
+// in a "dirty state". You only need the end
+// result to be in a nice state... which you
+// can achieve by calling repairAfterLazy at the end.
+//
+// The Java/C code does something special for
+// the in-place lazy OR runs. The idea is that
+// instead of taking two run containers and
+// generating a new one, we actually try to
+// do the computation in-place through a
+// technique invented by @gssiyankai (pinging him!).
+// What you do is you check whether the host
+// run container has lots of extra capacity.
+// If it does, you move its data at the end of
+// the backing array, and then you write
+// the answer at the beginning. What this
+// trick does is minimize memory allocations.
+//
+func (rc *runContainer16) lazyIOR(a container) container {
+	// not lazy at the moment
+	// TODO: make it lazy
+	return rc.ior(a)
+
+	/*
+		switch c := a.(type) {
+		case *arrayContainer:
+			return rc.lazyIorArray(c)
+		case *bitmapContainer:
+			return rc.lazyIorBitmap(c)
+		case *runContainer16:
+			return rc.lazyIorRun16(c)
+		}
+		panic("unsupported container type")
+	*/
+}
+
+// lazyOR is described above in lazyIOR.
+func (rc *runContainer16) lazyOR(a container) container {
+
+	// not lazy at the moment
+	// TODO: make it lazy
+	return rc.or(a)
+
+	/*
+		switch c := a.(type) {
+		case *arrayContainer:
+			return rc.lazyOrArray(c)
+		case *bitmapContainer:
+			return rc.lazyOrBitmap(c)
+		case *runContainer16:
+			return rc.lazyOrRunContainer16(c)
+		}
+		panic("unsupported container type")
+	*/
+}
+
+func (rc *runContainer16) intersects(a container) bool {
+	// TODO: optimize by doing inplace/less allocation, possibly?
+	isect := rc.and(a)
+	return isect.getCardinality() > 0
+}
+
+func (rc *runContainer16) xor(a container) container {
+	switch c := a.(type) {
+	case *arrayContainer:
+		return rc.xorArray(c)
+	case *bitmapContainer:
+		return rc.xorBitmap(c)
+	case *runContainer16:
+		return rc.xorRunContainer16(c)
+	}
+	panic("unsupported container type")
+}
+
+func (rc *runContainer16) iandNot(a container) container {
+	switch c := a.(type) {
+	case *arrayContainer:
+		return rc.iandNotArray(c)
+	case *bitmapContainer:
+		return rc.iandNotBitmap(c)
+	case *runContainer16:
+		return rc.iandNotRunContainer16(c)
+	}
+	panic("unsupported container type")
+}
+
+// flip the values in the range [firstOfRange,endx)
+func (rc *runContainer16) inot(firstOfRange, endx int) container {
+	if firstOfRange >= endx {
+		panic(fmt.Sprintf("invalid %v = endx >= firstOfRange = %v", endx, firstOfRange))
+	}
+	// TODO: minimize copies, do it all inplace; not() makes a copy.
+	rc = rc.Not(firstOfRange, endx)
+	return rc
+}
+
+func (rc *runContainer16) getCardinality() int {
+	return int(rc.cardinality())
+}
+
+func (rc *runContainer16) rank(x uint16) int {
+	n := int64(len(rc.iv))
+	xx := int64(x)
+	w, already, _ := rc.search(xx, nil)
+	if w < 0 {
+		return 0
+	}
+	if !already && w == n-1 {
+		return rc.getCardinality()
+	}
+	var rnk int64
+	if !already {
+		for i := int64(0); i <= w; i++ {
+			rnk += rc.iv[i].runlen()
+		}
+		return int(rnk)
+	}
+	for i := int64(0); i < w; i++ {
+		rnk += rc.iv[i].runlen()
+	}
+	rnk += int64(x-rc.iv[w].start) + 1
+	return int(rnk)
+}
+
+func (rc *runContainer16) selectInt(x uint16) int {
+	return rc.selectInt16(x)
+}
+
+func (rc *runContainer16) andNotRunContainer16(b *runContainer16) container {
+	return rc.AndNotRunContainer16(b)
+}
+
+func (rc *runContainer16) andNotArray(ac *arrayContainer) container {
+	rcb := rc.toBitmapContainer()
+	acb := ac.toBitmapContainer()
+	return rcb.andNotBitmap(acb)
+}
+
+func (rc *runContainer16) andNotBitmap(bc *bitmapContainer) container {
+	rcb := rc.toBitmapContainer()
+	return rcb.andNotBitmap(bc)
+}
+
+func (rc *runContainer16) toBitmapContainer() *bitmapContainer {
+	p("run16 toBitmap starting; rc has %v ranges", len(rc.iv))
+	bc := newBitmapContainer()
+	for i := range rc.iv {
+		bc.iaddRange(int(rc.iv[i].start), int(rc.iv[i].last())+1)
+	}
+	bc.computeCardinality()
+	return bc
+}
+
+func (rc *runContainer16) iandNotRunContainer16(x2 *runContainer16) container {
+	rcb := rc.toBitmapContainer()
+	x2b := x2.toBitmapContainer()
+	rcb.iandNotBitmapSurely(x2b)
+	// TODO: check size and optimize the return value
+	// TODO: is inplace modification really required? If not, elide the copy.
+	rc2 := newRunContainer16FromBitmapContainer(rcb)
+	*rc = *rc2
+	return rc
+}
+
+func (rc *runContainer16) iandNotArray(ac *arrayContainer) container {
+	rcb := rc.toBitmapContainer()
+	acb := ac.toBitmapContainer()
+	rcb.iandNotBitmapSurely(acb)
+	// TODO: check size and optimize the return value
+	// TODO: is inplace modification really required? If not, elide the copy.
+	rc2 := newRunContainer16FromBitmapContainer(rcb)
+	*rc = *rc2
+	return rc
+}
+
+func (rc *runContainer16) iandNotBitmap(bc *bitmapContainer) container {
+	rcb := rc.toBitmapContainer()
+	rcb.iandNotBitmapSurely(bc)
+	// TODO: check size and optimize the return value
+	// TODO: is inplace modification really required? If not, elide the copy.
+	rc2 := newRunContainer16FromBitmapContainer(rcb)
+	*rc = *rc2
+	return rc
+}
+
+func (rc *runContainer16) xorRunContainer16(x2 *runContainer16) container {
+	rcb := rc.toBitmapContainer()
+	x2b := x2.toBitmapContainer()
+	return rcb.xorBitmap(x2b)
+}
+
+func (rc *runContainer16) xorArray(ac *arrayContainer) container {
+	rcb := rc.toBitmapContainer()
+	acb := ac.toBitmapContainer()
+	return rcb.xorBitmap(acb)
+}
+
+func (rc *runContainer16) xorBitmap(bc *bitmapContainer) container {
+	rcb := rc.toBitmapContainer()
+	return rcb.xorBitmap(bc)
+}
+
+// convert to bitmap or array *if needed*
+func (rc *runContainer16) toEfficientContainer() container {
+
+	// runContainer16SerializedSizeInBytes(numRuns)
+	sizeAsRunContainer := rc.getSizeInBytes()
+	sizeAsBitmapContainer := bitmapContainerSizeInBytes()
+	card := int(rc.cardinality())
+	sizeAsArrayContainer := arrayContainerSizeInBytes(card)
+	if sizeAsRunContainer <= minOfInt(sizeAsBitmapContainer, sizeAsArrayContainer) {
+		return rc
+	}
+	if card <= arrayDefaultMaxSize {
+		return rc.toArrayContainer()
+	}
+	bc := newBitmapContainerFromRun(rc)
+	return bc
+}
+
+func (rc *runContainer16) toArrayContainer() *arrayContainer {
+	ac := newArrayContainer()
+	for i := range rc.iv {
+		ac.iaddRange(int(rc.iv[i].start), int(rc.iv[i].last())+1)
+	}
+	return ac
+}
+
+func newRunContainer16FromContainer(c container) *runContainer16 {
+
+	switch x := c.(type) {
+	case *runContainer16:
+		return x.Clone()
+	case *arrayContainer:
+		return newRunContainer16FromArray(x)
+	case *bitmapContainer:
+		return newRunContainer16FromBitmapContainer(x)
+	}
+	panic("unsupported container type")
+}

vendor/github.com/RoaringBitmap/roaring/roaringarray.go

@@ -0,0 +1,893 @@
+package roaring
+
+import (
+	"bytes"
+	"encoding/binary"
+	"fmt"
+	"io"
+	"io/ioutil"
+
+	snappy "github.com/glycerine/go-unsnap-stream"
+	"github.com/tinylib/msgp/msgp"
+)
+
+//go:generate msgp -unexported
+
+type container interface {
+	clone() container
+	and(container) container
+	andCardinality(container) int
+	iand(container) container // i stands for inplace
+	andNot(container) container
+	iandNot(container) container // i stands for inplace
+	getCardinality() int
+	// rank returns the number of integers that are
+	// smaller or equal to x. rank(infinity) would be getCardinality().
+	rank(uint16) int
+
+	iadd(x uint16) bool                   // inplace, returns true if x was new.
+	iaddReturnMinimized(uint16) container // may change return type to minimize storage.
+
+	//addRange(start, final int) container  // range is [firstOfRange,lastOfRange) (unused)
+	iaddRange(start, endx int) container // i stands for inplace, range is [firstOfRange,endx)
+
+	iremove(x uint16) bool                   // inplace, returns true if x was present.
+	iremoveReturnMinimized(uint16) container // may change return type to minimize storage.
+
+	not(start, final int) container        // range is [firstOfRange,lastOfRange)
+	inot(firstOfRange, endx int) container // i stands for inplace, range is [firstOfRange,endx)
+	xor(r container) container
+	getShortIterator() shortIterable
+	getManyIterator() manyIterable
+	contains(i uint16) bool
+	maximum() uint16
+	minimum() uint16
+
+	// equals is now logical equals; it does not require the
+	// same underlying container types, but compares across
+	// any of the implementations.
+	equals(r container) bool
+
+	fillLeastSignificant16bits(array []uint32, i int, mask uint32)
+	or(r container) container
+	orCardinality(r container) int
+	isFull() bool
+	ior(r container) container   // i stands for inplace
+	intersects(r container) bool // whether the two containers intersect
+	lazyOR(r container) container
+	lazyIOR(r container) container
+	getSizeInBytes() int
+	//removeRange(start, final int) container  // range is [firstOfRange,lastOfRange) (unused)
+	iremoveRange(start, final int) container // i stands for inplace, range is [firstOfRange,lastOfRange)
+	selectInt(x uint16) int                  // selectInt returns the xth integer in the container
+	serializedSizeInBytes() int
+	readFrom(io.Reader) (int, error)
+	writeTo(io.Writer) (int, error)
+
+	numberOfRuns() int
+	toEfficientContainer() container
+	String() string
+	containerType() contype
+}
+
+type contype uint8
+
+const (
+	bitmapContype contype = iota
+	arrayContype
+	run16Contype
+	run32Contype
+)
+
+// careful: range is [firstOfRange,lastOfRange]
+func rangeOfOnes(start, last int) container {
+	if start > MaxUint16 {
+		panic("rangeOfOnes called with start > MaxUint16")
+	}
+	if last > MaxUint16 {
+		panic("rangeOfOnes called with last > MaxUint16")
+	}
+	if start < 0 {
+		panic("rangeOfOnes called with start < 0")
+	}
+	if last < 0 {
+		panic("rangeOfOnes called with last < 0")
+	}
+	return newRunContainer16Range(uint16(start), uint16(last))
+}
+
+type roaringArray struct {
+	keys            []uint16
+	containers      []container `msg:"-"` // don't try to serialize directly.
+	needCopyOnWrite []bool
+	copyOnWrite     bool
+
+	// conserz is used at serialization time
+	// to serialize containers. Otherwise empty.
+	conserz []containerSerz
+}
+
+// containerSerz facilitates serializing container (tricky to
+// serialize because it is an interface) by providing a
+// light wrapper with a type identifier.
+type containerSerz struct {
+	t contype  `msg:"t"` // type
+	r msgp.Raw `msg:"r"` // Raw msgpack of the actual container type
+}
+
+func newRoaringArray() *roaringArray {
+	return &roaringArray{}
+}
+
+// runOptimize compresses the element containers to minimize space consumed.
+// Q: how does this interact with copyOnWrite and needCopyOnWrite?
+// A: since we aren't changing the logical content, just the representation,
+//    we don't bother to check the needCopyOnWrite bits. We replace
+//    (possibly all) elements of ra.containers in-place with space
+//    optimized versions.
+func (ra *roaringArray) runOptimize() {
+	for i := range ra.containers {
+		ra.containers[i] = ra.containers[i].toEfficientContainer()
+	}
+}
+
+func (ra *roaringArray) appendContainer(key uint16, value container, mustCopyOnWrite bool) {
+	ra.keys = append(ra.keys, key)
+	ra.containers = append(ra.containers, value)
+	ra.needCopyOnWrite = append(ra.needCopyOnWrite, mustCopyOnWrite)
+}
+
+func (ra *roaringArray) appendWithoutCopy(sa roaringArray, startingindex int) {
+	mustCopyOnWrite := sa.needCopyOnWrite[startingindex]
+	ra.appendContainer(sa.keys[startingindex], sa.containers[startingindex], mustCopyOnWrite)
+}
+
+func (ra *roaringArray) appendCopy(sa roaringArray, startingindex int) {
+	// cow only if the two request it, or if we already have a lightweight copy
+	copyonwrite := (ra.copyOnWrite && sa.copyOnWrite) || sa.needsCopyOnWrite(startingindex)
+	if !copyonwrite {
+		// since there is no copy-on-write, we need to clone the container (this is important)
+		ra.appendContainer(sa.keys[startingindex], sa.containers[startingindex].clone(), copyonwrite)
+	} else {
+		ra.appendContainer(sa.keys[startingindex], sa.containers[startingindex], copyonwrite)
+		if !sa.needsCopyOnWrite(startingindex) {
+			sa.setNeedsCopyOnWrite(startingindex)
+		}
+	}
+}
+
+func (ra *roaringArray) appendWithoutCopyMany(sa roaringArray, startingindex, end int) {
+	for i := startingindex; i < end; i++ {
+		ra.appendWithoutCopy(sa, i)
+	}
+}
+
+func (ra *roaringArray) appendCopyMany(sa roaringArray, startingindex, end int) {
+	for i := startingindex; i < end; i++ {
+		ra.appendCopy(sa, i)
+	}
+}
+
+func (ra *roaringArray) appendCopiesUntil(sa roaringArray, stoppingKey uint16) {
+	// cow only if the two request it, or if we already have a lightweight copy
+	copyonwrite := ra.copyOnWrite && sa.copyOnWrite
+
+	for i := 0; i < sa.size(); i++ {
+		if sa.keys[i] >= stoppingKey {
+			break
+		}
+		thiscopyonewrite := copyonwrite || sa.needsCopyOnWrite(i)
+		if thiscopyonewrite {
+			ra.appendContainer(sa.keys[i], sa.containers[i], thiscopyonewrite)
+			if !sa.needsCopyOnWrite(i) {
+				sa.setNeedsCopyOnWrite(i)
+			}
+
+		} else {
+			// since there is no copy-on-write, we need to clone the container (this is important)
+			ra.appendContainer(sa.keys[i], sa.containers[i].clone(), thiscopyonewrite)
+
+		}
+	}
+}
+
+func (ra *roaringArray) appendCopiesAfter(sa roaringArray, beforeStart uint16) {
+	// cow only if the two request it, or if we already have a lightweight copy
+	copyonwrite := ra.copyOnWrite && sa.copyOnWrite
+
+	startLocation := sa.getIndex(beforeStart)
+	if startLocation >= 0 {
+		startLocation++
+	} else {
+		startLocation = -startLocation - 1
+	}
+
+	for i := startLocation; i < sa.size(); i++ {
+		thiscopyonewrite := copyonwrite || sa.needsCopyOnWrite(i)
+		if thiscopyonewrite {
+			ra.appendContainer(sa.keys[i], sa.containers[i], thiscopyonewrite)
+			if !sa.needsCopyOnWrite(i) {
+				sa.setNeedsCopyOnWrite(i)
+			}
+		} else {
+			// since there is no copy-on-write, we need to clone the container (this is important)
+			ra.appendContainer(sa.keys[i], sa.containers[i].clone(), thiscopyonewrite)
+
+		}
+	}
+}
+
+func (ra *roaringArray) removeIndexRange(begin, end int) {
+	if end <= begin {
+		return
+	}
+
+	r := end - begin
+
+	copy(ra.keys[begin:], ra.keys[end:])
+	copy(ra.containers[begin:], ra.containers[end:])
+	copy(ra.needCopyOnWrite[begin:], ra.needCopyOnWrite[end:])
+
+	ra.resize(len(ra.keys) - r)
+}
+
+func (ra *roaringArray) resize(newsize int) {
+	for k := newsize; k < len(ra.containers); k++ {
+		ra.containers[k] = nil
+	}
+
+	ra.keys = ra.keys[:newsize]
+	ra.containers = ra.containers[:newsize]
+	ra.needCopyOnWrite = ra.needCopyOnWrite[:newsize]
+}
+
+func (ra *roaringArray) clear() {
+	ra.resize(0)
+	ra.copyOnWrite = false
+	ra.conserz = nil
+}
+
+func (ra *roaringArray) clone() *roaringArray {
+
+	sa := roaringArray{}
+	sa.copyOnWrite = ra.copyOnWrite
+
+	// this is where copyOnWrite is used.
+	if ra.copyOnWrite {
+		sa.keys = make([]uint16, len(ra.keys))
+		copy(sa.keys, ra.keys)
+		sa.containers = make([]container, len(ra.containers))
+		copy(sa.containers, ra.containers)
+		sa.needCopyOnWrite = make([]bool, len(ra.needCopyOnWrite))
+
+		ra.markAllAsNeedingCopyOnWrite()
+		sa.markAllAsNeedingCopyOnWrite()
+
+		// sa.needCopyOnWrite is shared
+	} else {
+		// make a full copy
+
+		sa.keys = make([]uint16, len(ra.keys))
+		copy(sa.keys, ra.keys)
+
+		sa.containers = make([]container, len(ra.containers))
+		for i := range sa.containers {
+			sa.containers[i] = ra.containers[i].clone()
+		}
+
+		sa.needCopyOnWrite = make([]bool, len(ra.needCopyOnWrite))
+	}
+	return &sa
+}
+
+// unused function:
+//func (ra *roaringArray) containsKey(x uint16) bool {
+//	return (ra.binarySearch(0, int64(len(ra.keys)), x) >= 0)
+//}
+
+func (ra *roaringArray) getContainer(x uint16) container {
+	i := ra.binarySearch(0, int64(len(ra.keys)), x)
+	if i < 0 {
+		return nil
+	}
+	return ra.containers[i]
+}
+
+func (ra *roaringArray) getContainerAtIndex(i int) container {
+	return ra.containers[i]
+}
+
+func (ra *roaringArray) getFastContainerAtIndex(i int, needsWriteable bool) container {
+	c := ra.getContainerAtIndex(i)
+	switch t := c.(type) {
+	case *arrayContainer:
+		c = t.toBitmapContainer()
+	case *runContainer16:
+		if !t.isFull() {
+			c = t.toBitmapContainer()
+		}
+	case *bitmapContainer:
+		if needsWriteable && ra.needCopyOnWrite[i] {
+			c = ra.containers[i].clone()
+		}
+	}
+	return c
+}
+
+func (ra *roaringArray) getWritableContainerAtIndex(i int) container {
+	if ra.needCopyOnWrite[i] {
+		ra.containers[i] = ra.containers[i].clone()
+		ra.needCopyOnWrite[i] = false
+	}
+	return ra.containers[i]
+}
+
+func (ra *roaringArray) getIndex(x uint16) int {
+	// before the binary search, we optimize for frequent cases
+	size := len(ra.keys)
+	if (size == 0) || (ra.keys[size-1] == x) {
+		return size - 1
+	}
+	return ra.binarySearch(0, int64(size), x)
+}
+
+func (ra *roaringArray) getKeyAtIndex(i int) uint16 {
+	return ra.keys[i]
+}
+
+func (ra *roaringArray) insertNewKeyValueAt(i int, key uint16, value container) {
+	ra.keys = append(ra.keys, 0)
+	ra.containers = append(ra.containers, nil)
+
+	copy(ra.keys[i+1:], ra.keys[i:])
+	copy(ra.containers[i+1:], ra.containers[i:])
+
+	ra.keys[i] = key
+	ra.containers[i] = value
+
+	ra.needCopyOnWrite = append(ra.needCopyOnWrite, false)
+	copy(ra.needCopyOnWrite[i+1:], ra.needCopyOnWrite[i:])
+	ra.needCopyOnWrite[i] = false
+}
+
+func (ra *roaringArray) remove(key uint16) bool {
+	i := ra.binarySearch(0, int64(len(ra.keys)), key)
+	if i >= 0 { // if a new key
+		ra.removeAtIndex(i)
+		return true
+	}
+	return false
+}
+
+func (ra *roaringArray) removeAtIndex(i int) {
+	copy(ra.keys[i:], ra.keys[i+1:])
+	copy(ra.containers[i:], ra.containers[i+1:])
+
+	copy(ra.needCopyOnWrite[i:], ra.needCopyOnWrite[i+1:])
+
+	ra.resize(len(ra.keys) - 1)
+}
+
+func (ra *roaringArray) setContainerAtIndex(i int, c container) {
+	ra.containers[i] = c
+}
+
+func (ra *roaringArray) replaceKeyAndContainerAtIndex(i int, key uint16, c container, mustCopyOnWrite bool) {
+	ra.keys[i] = key
+	ra.containers[i] = c
+	ra.needCopyOnWrite[i] = mustCopyOnWrite
+}
+
+func (ra *roaringArray) size() int {
+	return len(ra.keys)
+}
+
+func (ra *roaringArray) binarySearch(begin, end int64, ikey uint16) int {
+	low := begin
+	high := end - 1
+	for low+16 <= high {
+		middleIndex := low + (high-low)/2 // avoid overflow
+		middleValue := ra.keys[middleIndex]
+
+		if middleValue < ikey {
+			low = middleIndex + 1
+		} else if middleValue > ikey {
+			high = middleIndex - 1
+		} else {
+			return int(middleIndex)
+		}
+	}
+	for ; low <= high; low++ {
+		val := ra.keys[low]
+		if val >= ikey {
+			if val == ikey {
+				return int(low)
+			}
+			break
+		}
+	}
+	return -int(low + 1)
+}
+
+func (ra *roaringArray) equals(o interface{}) bool {
+	srb, ok := o.(roaringArray)
+	if ok {
+
+		if srb.size() != ra.size() {
+			return false
+		}
+		for i, k := range ra.keys {
+			if k != srb.keys[i] {
+				return false
+			}
+		}
+
+		for i, c := range ra.containers {
+			if !c.equals(srb.containers[i]) {
+				return false
+			}
+		}
+		return true
+	}
+	return false
+}
+
+func (ra *roaringArray) headerSize() uint64 {
+	size := uint64(len(ra.keys))
+	if ra.hasRunCompression() {
+		if size < noOffsetThreshold { // for small bitmaps, we omit the offsets
+			return 4 + (size+7)/8 + 4*size
+		}
+		return 4 + (size+7)/8 + 8*size // - 4 because we pack the size with the cookie
+	}
+	return 4 + 4 + 8*size
+
+}
+
+// should be dirt cheap
+func (ra *roaringArray) serializedSizeInBytes() uint64 {
+	answer := ra.headerSize()
+	for _, c := range ra.containers {
+		answer += uint64(c.serializedSizeInBytes())
+	}
+	return answer
+}
+
+//
+// spec: https://github.com/RoaringBitmap/RoaringFormatSpec
+//
+func (ra *roaringArray) toBytes() ([]byte, error) {
+	stream := &bytes.Buffer{}
+	hasRun := ra.hasRunCompression()
+	isRunSizeInBytes := 0
+	cookieSize := 8
+	if hasRun {
+		cookieSize = 4
+		isRunSizeInBytes = (len(ra.keys) + 7) / 8
+	}
+	descriptiveHeaderSize := 4 * len(ra.keys)
+	preambleSize := cookieSize + isRunSizeInBytes + descriptiveHeaderSize
+
+	buf := make([]byte, preambleSize+4*len(ra.keys))
+
+	nw := 0
+
+	if hasRun {
+		binary.LittleEndian.PutUint16(buf[0:], uint16(serialCookie))
+		nw += 2
+		binary.LittleEndian.PutUint16(buf[2:], uint16(len(ra.keys)-1))
+		nw += 2
+
+		// compute isRun bitmap
+		var ir []byte
+
+		isRun := newBitmapContainer()
+		for i, c := range ra.containers {
+			switch c.(type) {
+			case *runContainer16:
+				isRun.iadd(uint16(i))
+			}
+		}
+		// convert to little endian
+		ir = isRun.asLittleEndianByteSlice()[:isRunSizeInBytes]
+		nw += copy(buf[nw:], ir)
+	} else {
+		binary.LittleEndian.PutUint32(buf[0:], uint32(serialCookieNoRunContainer))
+		nw += 4
+		binary.LittleEndian.PutUint32(buf[4:], uint32(len(ra.keys)))
+		nw += 4
+	}
+
+	// descriptive header
+	for i, key := range ra.keys {
+		binary.LittleEndian.PutUint16(buf[nw:], key)
+		nw += 2
+		c := ra.containers[i]
+		binary.LittleEndian.PutUint16(buf[nw:], uint16(c.getCardinality()-1))
+		nw += 2
+	}
+
+	startOffset := int64(preambleSize + 4*len(ra.keys))
+	if !hasRun || (len(ra.keys) >= noOffsetThreshold) {
+		// offset header
+		for _, c := range ra.containers {
+			binary.LittleEndian.PutUint32(buf[nw:], uint32(startOffset))
+			nw += 4
+			switch rc := c.(type) {
+			case *runContainer16:
+				startOffset += 2 + int64(len(rc.iv))*4
+			default:
+				startOffset += int64(getSizeInBytesFromCardinality(c.getCardinality()))
+			}
+		}
+	}
+
+	_, err := stream.Write(buf[:nw])
+	if err != nil {
+		return nil, err
+	}
+	for i, c := range ra.containers {
+		_ = i
+		_, err := c.writeTo(stream)
+		if err != nil {
+			return nil, err
+		}
+	}
+	return stream.Bytes(), nil
+}
+
+//
+// spec: https://github.com/RoaringBitmap/RoaringFormatSpec
+//
+func (ra *roaringArray) writeTo(out io.Writer) (int64, error) {
+	by, err := ra.toBytes()
+	if err != nil {
+		return 0, err
+	}
+	n, err := out.Write(by)
+	if err == nil && n < len(by) {
+		err = io.ErrShortWrite
+	}
+	return int64(n), err
+}
+
+func (ra *roaringArray) fromBuffer(buf []byte) (int64, error) {
+	pos := 0
+	if len(buf) < 8 {
+		return 0, fmt.Errorf("buffer too small, expecting at least 8 bytes, was %d", len(buf))
+	}
+
+	cookie := binary.LittleEndian.Uint32(buf)
+	pos += 4
+	var size uint32 // number of containers
+	haveRunContainers := false
+	var isRunBitmap []byte
+
+	// cookie header
+	if cookie&0x0000FFFF == serialCookie {
+		haveRunContainers = true
+		size = uint32(uint16(cookie>>16) + 1) // number of containers
+
+		// create is-run-container bitmap
+		isRunBitmapSize := (int(size) + 7) / 8
+		if pos+isRunBitmapSize > len(buf) {
+			return 0, fmt.Errorf("malformed bitmap, is-run bitmap overruns buffer at %d", pos+isRunBitmapSize)
+		}
+
+		isRunBitmap = buf[pos : pos+isRunBitmapSize]
+		pos += isRunBitmapSize
+	} else if cookie == serialCookieNoRunContainer {
+		size = binary.LittleEndian.Uint32(buf[pos:])
+		pos += 4
+	} else {
+		return 0, fmt.Errorf("error in roaringArray.readFrom: did not find expected serialCookie in header")
+	}
+	if size > (1 << 16) {
+		return 0, fmt.Errorf("It is logically impossible to have more than (1<<16) containers.")
+	}
+	// descriptive header
+	// keycard - is {key, cardinality} tuple slice
+	if pos+2*2*int(size) > len(buf) {
+		return 0, fmt.Errorf("malfomred bitmap, key-cardinality slice overruns buffer at %d", pos+2*2*int(size))
+	}
+	keycard := byteSliceAsUint16Slice(buf[pos : pos+2*2*int(size)])
+	pos += 2 * 2 * int(size)
+
+	if !haveRunContainers || size >= noOffsetThreshold {
+		pos += 4 * int(size)
+	}
+
+	// Allocate slices upfront as number of containers is known
+	if cap(ra.containers) >= int(size) {
+		ra.containers = ra.containers[:size]
+	} else {
+		ra.containers = make([]container, size)
+	}
+	if cap(ra.keys) >= int(size) {
+		ra.keys = ra.keys[:size]
+	} else {
+		ra.keys = make([]uint16, size)
+	}
+	if cap(ra.needCopyOnWrite) >= int(size) {
+		ra.needCopyOnWrite = ra.needCopyOnWrite[:size]
+	} else {
+		ra.needCopyOnWrite = make([]bool, size)
+	}
+
+	for i := uint32(0); i < size; i++ {
+		key := uint16(keycard[2*i])
+		card := int(keycard[2*i+1]) + 1
+		ra.keys[i] = key
+		ra.needCopyOnWrite[i] = true
+
+		if haveRunContainers && isRunBitmap[i/8]&(1<<(i%8)) != 0 {
+			// run container
+			nr := binary.LittleEndian.Uint16(buf[pos:])
+			pos += 2
+			if pos+int(nr)*4 > len(buf) {
+				return 0, fmt.Errorf("malformed bitmap, a run container overruns buffer at %d:%d", pos, pos+int(nr)*4)
+			}
+			nb := runContainer16{
+				iv:   byteSliceAsInterval16Slice(buf[pos : pos+int(nr)*4]),
+				card: int64(card),
+			}
+			pos += int(nr) * 4
+			ra.containers[i] = &nb
+		} else if card > arrayDefaultMaxSize {
+			// bitmap container
+			nb := bitmapContainer{
+				cardinality: card,
+				bitmap:      byteSliceAsUint64Slice(buf[pos : pos+arrayDefaultMaxSize*2]),
+			}
+			pos += arrayDefaultMaxSize * 2
+			ra.containers[i] = &nb
+		} else {
+			// array container
+			nb := arrayContainer{
+				byteSliceAsUint16Slice(buf[pos : pos+card*2]),
+			}
+			pos += card * 2
+			ra.containers[i] = &nb
+		}
+	}
+
+	return int64(pos), nil
+}
+
+func (ra *roaringArray) readFrom(stream io.Reader) (int64, error) {
+	pos := 0
+	var cookie uint32
+	err := binary.Read(stream, binary.LittleEndian, &cookie)
+	if err != nil {
+		return 0, fmt.Errorf("error in roaringArray.readFrom: could not read initial cookie: %s", err)
+	}
+	pos += 4
+	var size uint32
+	haveRunContainers := false
+	var isRun *bitmapContainer
+	if cookie&0x0000FFFF == serialCookie {
+		haveRunContainers = true
+		size = uint32(uint16(cookie>>16) + 1)
+		bytesToRead := (int(size) + 7) / 8
+		numwords := (bytesToRead + 7) / 8
+		by := make([]byte, bytesToRead, numwords*8)
+		nr, err := io.ReadFull(stream, by)
+		if err != nil {
+			return 8 + int64(nr), fmt.Errorf("error in readFrom: could not read the "+
+				"runContainer bit flags of length %v bytes: %v", bytesToRead, err)
+		}
+		pos += bytesToRead
+		by = by[:cap(by)]
+		isRun = newBitmapContainer()
+		for i := 0; i < numwords; i++ {
+			isRun.bitmap[i] = binary.LittleEndian.Uint64(by)
+			by = by[8:]
+		}
+	} else if cookie == serialCookieNoRunContainer {
+		err = binary.Read(stream, binary.LittleEndian, &size)
+		if err != nil {
+			return 0, fmt.Errorf("error in roaringArray.readFrom: when reading size, got: %s", err)
+		}
+		pos += 4
+	} else {
+		return 0, fmt.Errorf("error in roaringArray.readFrom: did not find expected serialCookie in header")
+	}
+	if size > (1 << 16) {
+		return 0, fmt.Errorf("It is logically impossible to have more than (1<<16) containers.")
+	}
+	// descriptive header
+	keycard := make([]uint16, 2*size, 2*size)
+	err = binary.Read(stream, binary.LittleEndian, keycard)
+	if err != nil {
+		return 0, err
+	}
+	pos += 2 * 2 * int(size)
+	// offset header
+	if !haveRunContainers || size >= noOffsetThreshold {
+		io.CopyN(ioutil.Discard, stream, 4*int64(size)) // we never skip ahead so this data can be ignored
+		pos += 4 * int(size)
+	}
+	for i := uint32(0); i < size; i++ {
+		key := int(keycard[2*i])
+		card := int(keycard[2*i+1]) + 1
+		if haveRunContainers && isRun.contains(uint16(i)) {
+			nb := newRunContainer16()
+			nr, err := nb.readFrom(stream)
+			if err != nil {
+				return 0, err
+			}
+			pos += nr
+			ra.appendContainer(uint16(key), nb, false)
+		} else if card > arrayDefaultMaxSize {
+			nb := newBitmapContainer()
+			nr, err := nb.readFrom(stream)
+			if err != nil {
+				return 0, err
+			}
+			nb.cardinality = card
+			pos += nr
+			ra.appendContainer(keycard[2*i], nb, false)
+		} else {
+			nb := newArrayContainerSize(card)
+			nr, err := nb.readFrom(stream)
+			if err != nil {
+				return 0, err
+			}
+			pos += nr
+			ra.appendContainer(keycard[2*i], nb, false)
+		}
+	}
+	return int64(pos), nil
+}
+
+func (ra *roaringArray) hasRunCompression() bool {
+	for _, c := range ra.containers {
+		switch c.(type) {
+		case *runContainer16:
+			return true
+		}
+	}
+	return false
+}
+
+func (ra *roaringArray) writeToMsgpack(stream io.Writer) error {
+
+	ra.conserz = make([]containerSerz, len(ra.containers))
+	for i, v := range ra.containers {
+		switch cn := v.(type) {
+		case *bitmapContainer:
+			bts, err := cn.MarshalMsg(nil)
+			if err != nil {
+				return err
+			}
+			ra.conserz[i].t = bitmapContype
+			ra.conserz[i].r = bts
+		case *arrayContainer:
+			bts, err := cn.MarshalMsg(nil)
+			if err != nil {
+				return err
+			}
+			ra.conserz[i].t = arrayContype
+			ra.conserz[i].r = bts
+		case *runContainer16:
+			bts, err := cn.MarshalMsg(nil)
+			if err != nil {
+				return err
+			}
+			ra.conserz[i].t = run16Contype
+			ra.conserz[i].r = bts
+		default:
+			panic(fmt.Errorf("Unrecognized container implementation: %T", cn))
+		}
+	}
+	w := snappy.NewWriter(stream)
+	err := msgp.Encode(w, ra)
+	ra.conserz = nil
+	return err
+}
+
+func (ra *roaringArray) readFromMsgpack(stream io.Reader) error {
+	r := snappy.NewReader(stream)
+	err := msgp.Decode(r, ra)
+	if err != nil {
+		return err
+	}
+
+	if len(ra.containers) != len(ra.keys) {
+		ra.containers = make([]container, len(ra.keys))
+	}
+
+	for i, v := range ra.conserz {
+		switch v.t {
+		case bitmapContype:
+			c := &bitmapContainer{}
+			_, err = c.UnmarshalMsg(v.r)
+			if err != nil {
+				return err
+			}
+			ra.containers[i] = c
+		case arrayContype:
+			c := &arrayContainer{}
+			_, err = c.UnmarshalMsg(v.r)
+			if err != nil {
+				return err
+			}
+			ra.containers[i] = c
+		case run16Contype:
+			c := &runContainer16{}
+			_, err = c.UnmarshalMsg(v.r)
+			if err != nil {
+				return err
+			}
+			ra.containers[i] = c
+		default:
+			return fmt.Errorf("unrecognized contype serialization code: '%v'", v.t)
+		}
+	}
+	ra.conserz = nil
+	return nil
+}
+
+func (ra *roaringArray) advanceUntil(min uint16, pos int) int {
+	lower := pos + 1
+
+	if lower >= len(ra.keys) || ra.keys[lower] >= min {
+		return lower
+	}
+
+	spansize := 1
+
+	for lower+spansize < len(ra.keys) && ra.keys[lower+spansize] < min {
+		spansize *= 2
+	}
+	var upper int
+	if lower+spansize < len(ra.keys) {
+		upper = lower + spansize
+	} else {
+		upper = len(ra.keys) - 1
+	}
+
+	if ra.keys[upper] == min {
+		return upper
+	}
+
+	if ra.keys[upper] < min {
+		// means
+		// array
+		// has no
+		// item
+		// >= min
+		// pos = array.length;
+		return len(ra.keys)
+	}
+
+	// we know that the next-smallest span was too small
+	lower += (spansize >> 1)
+
+	mid := 0
+	for lower+1 != upper {
+		mid = (lower + upper) >> 1
+		if ra.keys[mid] == min {
+			return mid
+		} else if ra.keys[mid] < min {
+			lower = mid
+		} else {
+			upper = mid
+		}
+	}
+	return upper
+}
+
+func (ra *roaringArray) markAllAsNeedingCopyOnWrite() {
+	for i := range ra.needCopyOnWrite {
+		ra.needCopyOnWrite[i] = true
+	}
+}
+
+func (ra *roaringArray) needsCopyOnWrite(i int) bool {
+	return ra.needCopyOnWrite[i]
+}
+
+func (ra *roaringArray) setNeedsCopyOnWrite(i int) {
+	ra.needCopyOnWrite[i] = true
+}

vendor/github.com/RoaringBitmap/roaring/roaringarray_gen.go

@@ -0,0 +1,529 @@
+package roaring
+
+// NOTE: THIS FILE WAS PRODUCED BY THE
+// MSGP CODE GENERATION TOOL (github.com/tinylib/msgp)
+// DO NOT EDIT
+
+import (
+	"github.com/tinylib/msgp/msgp"
+)
+
+// DecodeMsg implements msgp.Decodable
+func (z *containerSerz) DecodeMsg(dc *msgp.Reader) (err error) {
+	var field []byte
+	_ = field
+	var zxvk uint32
+	zxvk, err = dc.ReadMapHeader()
+	if err != nil {
+		return
+	}
+	for zxvk > 0 {
+		zxvk--
+		field, err = dc.ReadMapKeyPtr()
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "t":
+			{
+				var zbzg uint8
+				zbzg, err = dc.ReadUint8()
+				z.t = contype(zbzg)
+			}
+			if err != nil {
+				return
+			}
+		case "r":
+			err = z.r.DecodeMsg(dc)
+			if err != nil {
+				return
+			}
+		default:
+			err = dc.Skip()
+			if err != nil {
+				return
+			}
+		}
+	}
+	return
+}
+
+// EncodeMsg implements msgp.Encodable
+func (z *containerSerz) EncodeMsg(en *msgp.Writer) (err error) {
+	// map header, size 2
+	// write "t"
+	err = en.Append(0x82, 0xa1, 0x74)
+	if err != nil {
+		return err
+	}
+	err = en.WriteUint8(uint8(z.t))
+	if err != nil {
+		return
+	}
+	// write "r"
+	err = en.Append(0xa1, 0x72)
+	if err != nil {
+		return err
+	}
+	err = z.r.EncodeMsg(en)
+	if err != nil {
+		return
+	}
+	return
+}
+
+// MarshalMsg implements msgp.Marshaler
+func (z *containerSerz) MarshalMsg(b []byte) (o []byte, err error) {
+	o = msgp.Require(b, z.Msgsize())
+	// map header, size 2
+	// string "t"
+	o = append(o, 0x82, 0xa1, 0x74)
+	o = msgp.AppendUint8(o, uint8(z.t))
+	// string "r"
+	o = append(o, 0xa1, 0x72)
+	o, err = z.r.MarshalMsg(o)
+	if err != nil {
+		return
+	}
+	return
+}
+
+// UnmarshalMsg implements msgp.Unmarshaler
+func (z *containerSerz) UnmarshalMsg(bts []byte) (o []byte, err error) {
+	var field []byte
+	_ = field
+	var zbai uint32
+	zbai, bts, err = msgp.ReadMapHeaderBytes(bts)
+	if err != nil {
+		return
+	}
+	for zbai > 0 {
+		zbai--
+		field, bts, err = msgp.ReadMapKeyZC(bts)
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "t":
+			{
+				var zcmr uint8
+				zcmr, bts, err = msgp.ReadUint8Bytes(bts)
+				z.t = contype(zcmr)
+			}
+			if err != nil {
+				return
+			}
+		case "r":
+			bts, err = z.r.UnmarshalMsg(bts)
+			if err != nil {
+				return
+			}
+		default:
+			bts, err = msgp.Skip(bts)
+			if err != nil {
+				return
+			}
+		}
+	}
+	o = bts
+	return
+}
+
+// Msgsize returns an upper bound estimate of the number of bytes occupied by the serialized message
+func (z *containerSerz) Msgsize() (s int) {
+	s = 1 + 2 + msgp.Uint8Size + 2 + z.r.Msgsize()
+	return
+}
+
+// DecodeMsg implements msgp.Decodable
+func (z *contype) DecodeMsg(dc *msgp.Reader) (err error) {
+	{
+		var zajw uint8
+		zajw, err = dc.ReadUint8()
+		(*z) = contype(zajw)
+	}
+	if err != nil {
+		return
+	}
+	return
+}
+
+// EncodeMsg implements msgp.Encodable
+func (z contype) EncodeMsg(en *msgp.Writer) (err error) {
+	err = en.WriteUint8(uint8(z))
+	if err != nil {
+		return
+	}
+	return
+}
+
+// MarshalMsg implements msgp.Marshaler
+func (z contype) MarshalMsg(b []byte) (o []byte, err error) {
+	o = msgp.Require(b, z.Msgsize())
+	o = msgp.AppendUint8(o, uint8(z))
+	return
+}
+
+// UnmarshalMsg implements msgp.Unmarshaler
+func (z *contype) UnmarshalMsg(bts []byte) (o []byte, err error) {
+	{
+		var zwht uint8
+		zwht, bts, err = msgp.ReadUint8Bytes(bts)
+		(*z) = contype(zwht)
+	}
+	if err != nil {
+		return
+	}
+	o = bts
+	return
+}
+
+// Msgsize returns an upper bound estimate of the number of bytes occupied by the serialized message
+func (z contype) Msgsize() (s int) {
+	s = msgp.Uint8Size
+	return
+}
+
+// DecodeMsg implements msgp.Decodable
+func (z *roaringArray) DecodeMsg(dc *msgp.Reader) (err error) {
+	var field []byte
+	_ = field
+	var zlqf uint32
+	zlqf, err = dc.ReadMapHeader()
+	if err != nil {
+		return
+	}
+	for zlqf > 0 {
+		zlqf--
+		field, err = dc.ReadMapKeyPtr()
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "keys":
+			var zdaf uint32
+			zdaf, err = dc.ReadArrayHeader()
+			if err != nil {
+				return
+			}
+			if cap(z.keys) >= int(zdaf) {
+				z.keys = (z.keys)[:zdaf]
+			} else {
+				z.keys = make([]uint16, zdaf)
+			}
+			for zhct := range z.keys {
+				z.keys[zhct], err = dc.ReadUint16()
+				if err != nil {
+					return
+				}
+			}
+		case "needCopyOnWrite":
+			var zpks uint32
+			zpks, err = dc.ReadArrayHeader()
+			if err != nil {
+				return
+			}
+			if cap(z.needCopyOnWrite) >= int(zpks) {
+				z.needCopyOnWrite = (z.needCopyOnWrite)[:zpks]
+			} else {
+				z.needCopyOnWrite = make([]bool, zpks)
+			}
+			for zcua := range z.needCopyOnWrite {
+				z.needCopyOnWrite[zcua], err = dc.ReadBool()
+				if err != nil {
+					return
+				}
+			}
+		case "copyOnWrite":
+			z.copyOnWrite, err = dc.ReadBool()
+			if err != nil {
+				return
+			}
+		case "conserz":
+			var zjfb uint32
+			zjfb, err = dc.ReadArrayHeader()
+			if err != nil {
+				return
+			}
+			if cap(z.conserz) >= int(zjfb) {
+				z.conserz = (z.conserz)[:zjfb]
+			} else {
+				z.conserz = make([]containerSerz, zjfb)
+			}
+			for zxhx := range z.conserz {
+				var zcxo uint32
+				zcxo, err = dc.ReadMapHeader()
+				if err != nil {
+					return
+				}
+				for zcxo > 0 {
+					zcxo--
+					field, err = dc.ReadMapKeyPtr()
+					if err != nil {
+						return
+					}
+					switch msgp.UnsafeString(field) {
+					case "t":
+						{
+							var zeff uint8
+							zeff, err = dc.ReadUint8()
+							z.conserz[zxhx].t = contype(zeff)
+						}
+						if err != nil {
+							return
+						}
+					case "r":
+						err = z.conserz[zxhx].r.DecodeMsg(dc)
+						if err != nil {
+							return
+						}
+					default:
+						err = dc.Skip()
+						if err != nil {
+							return
+						}
+					}
+				}
+			}
+		default:
+			err = dc.Skip()
+			if err != nil {
+				return
+			}
+		}
+	}
+	return
+}
+
+// EncodeMsg implements msgp.Encodable
+func (z *roaringArray) EncodeMsg(en *msgp.Writer) (err error) {
+	// map header, size 4
+	// write "keys"
+	err = en.Append(0x84, 0xa4, 0x6b, 0x65, 0x79, 0x73)
+	if err != nil {
+		return err
+	}
+	err = en.WriteArrayHeader(uint32(len(z.keys)))
+	if err != nil {
+		return
+	}
+	for zhct := range z.keys {
+		err = en.WriteUint16(z.keys[zhct])
+		if err != nil {
+			return
+		}
+	}
+	// write "needCopyOnWrite"
+	err = en.Append(0xaf, 0x6e, 0x65, 0x65, 0x64, 0x43, 0x6f, 0x70, 0x79, 0x4f, 0x6e, 0x57, 0x72, 0x69, 0x74, 0x65)
+	if err != nil {
+		return err
+	}
+	err = en.WriteArrayHeader(uint32(len(z.needCopyOnWrite)))
+	if err != nil {
+		return
+	}
+	for zcua := range z.needCopyOnWrite {
+		err = en.WriteBool(z.needCopyOnWrite[zcua])
+		if err != nil {
+			return
+		}
+	}
+	// write "copyOnWrite"
+	err = en.Append(0xab, 0x63, 0x6f, 0x70, 0x79, 0x4f, 0x6e, 0x57, 0x72, 0x69, 0x74, 0x65)
+	if err != nil {
+		return err
+	}
+	err = en.WriteBool(z.copyOnWrite)
+	if err != nil {
+		return
+	}
+	// write "conserz"
+	err = en.Append(0xa7, 0x63, 0x6f, 0x6e, 0x73, 0x65, 0x72, 0x7a)
+	if err != nil {
+		return err
+	}
+	err = en.WriteArrayHeader(uint32(len(z.conserz)))
+	if err != nil {
+		return
+	}
+	for zxhx := range z.conserz {
+		// map header, size 2
+		// write "t"
+		err = en.Append(0x82, 0xa1, 0x74)
+		if err != nil {
+			return err
+		}
+		err = en.WriteUint8(uint8(z.conserz[zxhx].t))
+		if err != nil {
+			return
+		}
+		// write "r"
+		err = en.Append(0xa1, 0x72)
+		if err != nil {
+			return err
+		}
+		err = z.conserz[zxhx].r.EncodeMsg(en)
+		if err != nil {
+			return
+		}
+	}
+	return
+}
+
+// MarshalMsg implements msgp.Marshaler
+func (z *roaringArray) MarshalMsg(b []byte) (o []byte, err error) {
+	o = msgp.Require(b, z.Msgsize())
+	// map header, size 4
+	// string "keys"
+	o = append(o, 0x84, 0xa4, 0x6b, 0x65, 0x79, 0x73)
+	o = msgp.AppendArrayHeader(o, uint32(len(z.keys)))
+	for zhct := range z.keys {
+		o = msgp.AppendUint16(o, z.keys[zhct])
+	}
+	// string "needCopyOnWrite"
+	o = append(o, 0xaf, 0x6e, 0x65, 0x65, 0x64, 0x43, 0x6f, 0x70, 0x79, 0x4f, 0x6e, 0x57, 0x72, 0x69, 0x74, 0x65)
+	o = msgp.AppendArrayHeader(o, uint32(len(z.needCopyOnWrite)))
+	for zcua := range z.needCopyOnWrite {
+		o = msgp.AppendBool(o, z.needCopyOnWrite[zcua])
+	}
+	// string "copyOnWrite"
+	o = append(o, 0xab, 0x63, 0x6f, 0x70, 0x79, 0x4f, 0x6e, 0x57, 0x72, 0x69, 0x74, 0x65)
+	o = msgp.AppendBool(o, z.copyOnWrite)
+	// string "conserz"
+	o = append(o, 0xa7, 0x63, 0x6f, 0x6e, 0x73, 0x65, 0x72, 0x7a)
+	o = msgp.AppendArrayHeader(o, uint32(len(z.conserz)))
+	for zxhx := range z.conserz {
+		// map header, size 2
+		// string "t"
+		o = append(o, 0x82, 0xa1, 0x74)
+		o = msgp.AppendUint8(o, uint8(z.conserz[zxhx].t))
+		// string "r"
+		o = append(o, 0xa1, 0x72)
+		o, err = z.conserz[zxhx].r.MarshalMsg(o)
+		if err != nil {
+			return
+		}
+	}
+	return
+}
+
+// UnmarshalMsg implements msgp.Unmarshaler
+func (z *roaringArray) UnmarshalMsg(bts []byte) (o []byte, err error) {
+	var field []byte
+	_ = field
+	var zrsw uint32
+	zrsw, bts, err = msgp.ReadMapHeaderBytes(bts)
+	if err != nil {
+		return
+	}
+	for zrsw > 0 {
+		zrsw--
+		field, bts, err = msgp.ReadMapKeyZC(bts)
+		if err != nil {
+			return
+		}
+		switch msgp.UnsafeString(field) {
+		case "keys":
+			var zxpk uint32
+			zxpk, bts, err = msgp.ReadArrayHeaderBytes(bts)
+			if err != nil {
+				return
+			}
+			if cap(z.keys) >= int(zxpk) {
+				z.keys = (z.keys)[:zxpk]
+			} else {
+				z.keys = make([]uint16, zxpk)
+			}
+			for zhct := range z.keys {
+				z.keys[zhct], bts, err = msgp.ReadUint16Bytes(bts)
+				if err != nil {
+					return
+				}
+			}
+		case "needCopyOnWrite":
+			var zdnj uint32
+			zdnj, bts, err = msgp.ReadArrayHeaderBytes(bts)
+			if err != nil {
+				return
+			}
+			if cap(z.needCopyOnWrite) >= int(zdnj) {
+				z.needCopyOnWrite = (z.needCopyOnWrite)[:zdnj]
+			} else {
+				z.needCopyOnWrite = make([]bool, zdnj)
+			}
+			for zcua := range z.needCopyOnWrite {
+				z.needCopyOnWrite[zcua], bts, err = msgp.ReadBoolBytes(bts)
+				if err != nil {
+					return
+				}
+			}
+		case "copyOnWrite":
+			z.copyOnWrite, bts, err = msgp.ReadBoolBytes(bts)
+			if err != nil {
+				return
+			}
+		case "conserz":
+			var zobc uint32
+			zobc, bts, err = msgp.ReadArrayHeaderBytes(bts)
+			if err != nil {
+				return
+			}
+			if cap(z.conserz) >= int(zobc) {
+				z.conserz = (z.conserz)[:zobc]
+			} else {
+				z.conserz = make([]containerSerz, zobc)
+			}
+			for zxhx := range z.conserz {
+				var zsnv uint32
+				zsnv, bts, err = msgp.ReadMapHeaderBytes(bts)
+				if err != nil {
+					return
+				}
+				for zsnv > 0 {
+					zsnv--
+					field, bts, err = msgp.ReadMapKeyZC(bts)
+					if err != nil {
+						return
+					}
+					switch msgp.UnsafeString(field) {
+					case "t":
+						{
+							var zkgt uint8
+							zkgt, bts, err = msgp.ReadUint8Bytes(bts)
+							z.conserz[zxhx].t = contype(zkgt)
+						}
+						if err != nil {
+							return
+						}
+					case "r":
+						bts, err = z.conserz[zxhx].r.UnmarshalMsg(bts)
+						if err != nil {
+							return
+						}
+					default:
+						bts, err = msgp.Skip(bts)
+						if err != nil {
+							return
+						}
+					}
+				}
+			}
+		default:
+			bts, err = msgp.Skip(bts)
+			if err != nil {
+				return
+			}
+		}
+	}
+	o = bts
+	return
+}
+
+// Msgsize returns an upper bound estimate of the number of bytes occupied by the serialized message
+func (z *roaringArray) Msgsize() (s int) {
+	s = 1 + 5 + msgp.ArrayHeaderSize + (len(z.keys) * (msgp.Uint16Size)) + 16 + msgp.ArrayHeaderSize + (len(z.needCopyOnWrite) * (msgp.BoolSize)) + 12 + msgp.BoolSize + 8 + msgp.ArrayHeaderSize
+	for zxhx := range z.conserz {
+		s += 1 + 2 + msgp.Uint8Size + 2 + z.conserz[zxhx].r.Msgsize()
+	}
+	return
+}

vendor/github.com/RoaringBitmap/roaring/serialization.go

@@ -0,0 +1,83 @@
+package roaring
+
+import (
+	"encoding/binary"
+	"errors"
+	"fmt"
+	"io"
+
+	"github.com/tinylib/msgp/msgp"
+)
+
+// writeTo for runContainer16 follows this
+// spec: https://github.com/RoaringBitmap/RoaringFormatSpec
+//
+func (b *runContainer16) writeTo(stream io.Writer) (int, error) {
+	buf := make([]byte, 2+4*len(b.iv))
+	binary.LittleEndian.PutUint16(buf[0:], uint16(len(b.iv)))
+	for i, v := range b.iv {
+		binary.LittleEndian.PutUint16(buf[2+i*4:], v.start)
+		binary.LittleEndian.PutUint16(buf[2+2+i*4:], v.length)
+	}
+	return stream.Write(buf)
+}
+
+func (b *runContainer32) writeToMsgpack(stream io.Writer) (int, error) {
+	bts, err := b.MarshalMsg(nil)
+	if err != nil {
+		return 0, err
+	}
+	return stream.Write(bts)
+}
+
+func (b *runContainer16) writeToMsgpack(stream io.Writer) (int, error) {
+	bts, err := b.MarshalMsg(nil)
+	if err != nil {
+		return 0, err
+	}
+	return stream.Write(bts)
+}
+
+func (b *runContainer32) readFromMsgpack(stream io.Reader) (int, error) {
+	err := msgp.Decode(stream, b)
+	return 0, err
+}
+
+func (b *runContainer16) readFromMsgpack(stream io.Reader) (int, error) {
+	err := msgp.Decode(stream, b)
+	return 0, err
+}
+
+var errCorruptedStream = errors.New("insufficient/odd number of stored bytes, corrupted stream detected")
+
+func (b *runContainer16) readFrom(stream io.Reader) (int, error) {
+	b.iv = b.iv[:0]
+	b.card = 0
+	var numRuns uint16
+	err := binary.Read(stream, binary.LittleEndian, &numRuns)
+	if err != nil {
+		return 0, err
+	}
+	nr := int(numRuns)
+	encRun := make([]uint16, 2*nr)
+	by := make([]byte, 4*nr)
+	err = binary.Read(stream, binary.LittleEndian, &by)
+	if err != nil {
+		return 0, err
+	}
+	for i := range encRun {
+		if len(by) < 2 {
+			return 0, errCorruptedStream
+		}
+		encRun[i] = binary.LittleEndian.Uint16(by)
+		by = by[2:]
+	}
+	for i := 0; i < nr; i++ {
+		if i > 0 && b.iv[i-1].last() >= encRun[i*2] {
+			return 0, fmt.Errorf("error: stored runContainer had runs that were not in sorted order!! (b.iv[i-1=%v].last = %v >= encRun[i=%v] = %v)", i-1, b.iv[i-1].last(), i, encRun[i*2])
+		}
+		b.iv = append(b.iv, interval16{start: encRun[i*2], length: encRun[i*2+1]})
+		b.card += int64(encRun[i*2+1]) + 1
+	}
+	return 0, err
+}

vendor/github.com/RoaringBitmap/roaring/serialization_generic.go

@@ -0,0 +1,118 @@
+// +build !amd64,!386 appengine
+
+package roaring
+
+import (
+	"encoding/binary"
+	"io"
+)
+
+func (b *arrayContainer) writeTo(stream io.Writer) (int, error) {
+	buf := make([]byte, 2*len(b.content))
+	for i, v := range b.content {
+		base := i * 2
+		buf[base] = byte(v)
+		buf[base+1] = byte(v >> 8)
+	}
+	return stream.Write(buf)
+}
+
+func (b *arrayContainer) readFrom(stream io.Reader) (int, error) {
+	err := binary.Read(stream, binary.LittleEndian, b.content)
+	if err != nil {
+		return 0, err
+	}
+	return 2 * len(b.content), nil
+}
+
+func (b *bitmapContainer) writeTo(stream io.Writer) (int, error) {
+	// Write set
+	buf := make([]byte, 8*len(b.bitmap))
+	for i, v := range b.bitmap {
+		base := i * 8
+		buf[base] = byte(v)
+		buf[base+1] = byte(v >> 8)
+		buf[base+2] = byte(v >> 16)
+		buf[base+3] = byte(v >> 24)
+		buf[base+4] = byte(v >> 32)
+		buf[base+5] = byte(v >> 40)
+		buf[base+6] = byte(v >> 48)
+		buf[base+7] = byte(v >> 56)
+	}
+	return stream.Write(buf)
+}
+
+func (b *bitmapContainer) readFrom(stream io.Reader) (int, error) {
+	err := binary.Read(stream, binary.LittleEndian, b.bitmap)
+	if err != nil {
+		return 0, err
+	}
+	b.computeCardinality()
+	return 8 * len(b.bitmap), nil
+}
+
+func (bc *bitmapContainer) asLittleEndianByteSlice() []byte {
+	by := make([]byte, len(bc.bitmap)*8)
+	for i := range bc.bitmap {
+		binary.LittleEndian.PutUint64(by[i*8:], bc.bitmap[i])
+	}
+	return by
+}
+
+func uint64SliceAsByteSlice(slice []uint64) []byte {
+	by := make([]byte, len(slice)*8)
+
+	for i, v := range slice {
+		binary.LittleEndian.PutUint64(by[i*8:], v)
+	}
+
+	return by
+}
+
+func byteSliceAsUint16Slice(slice []byte) []uint16 {
+	if len(slice)%2 != 0 {
+		panic("Slice size should be divisible by 2")
+	}
+
+	b := make([]uint16, len(slice)/2)
+
+	for i := range b {
+		b[i] = binary.LittleEndian.Uint16(slice[2*i:])
+	}
+
+	return b
+}
+
+func byteSliceAsUint64Slice(slice []byte) []uint64 {
+	if len(slice)%8 != 0 {
+		panic("Slice size should be divisible by 8")
+	}
+
+	b := make([]uint64, len(slice)/8)
+
+	for i := range b {
+		b[i] = binary.LittleEndian.Uint64(slice[8*i:])
+	}
+
+	return b
+}
+
+// Converts a byte slice to a interval16 slice.
+// The function assumes that the slice byte buffer is run container data
+// encoded according to Roaring Format Spec
+func byteSliceAsInterval16Slice(byteSlice []byte) []interval16 {
+	if len(byteSlice)%4 != 0 {
+		panic("Slice size should be divisible by 4")
+	}
+
+	intervalSlice := make([]interval16, len(byteSlice)/4)
+
+	for i := range intervalSlice {
+		intervalSlice[i] = interval16{
+			start:  binary.LittleEndian.Uint16(byteSlice[i*4:]),
+			length: binary.LittleEndian.Uint16(byteSlice[i*4+2:]),
+		}
+	}
+
+	return intervalSlice
+}

vendor/github.com/RoaringBitmap/roaring/serialization_littleendian.go

@@ -0,0 +1,113 @@
+// +build 386 amd64,!appengine
+
+package roaring
+
+import (
+	"io"
+	"reflect"
+	"unsafe"
+)
+
+func (ac *arrayContainer) writeTo(stream io.Writer) (int, error) {
+	buf := uint16SliceAsByteSlice(ac.content)
+	return stream.Write(buf)
+}
+
+func (bc *bitmapContainer) writeTo(stream io.Writer) (int, error) {
+	buf := uint64SliceAsByteSlice(bc.bitmap)
+	return stream.Write(buf)
+}
+
+// readFrom reads an arrayContainer from stream.
+// PRE-REQUISITE: you must size the arrayContainer correctly (allocate b.content)
+// *before* you call readFrom. We can't guess the size in the stream
+// by this point.
+func (ac *arrayContainer) readFrom(stream io.Reader) (int, error) {
+	buf := uint16SliceAsByteSlice(ac.content)
+	return io.ReadFull(stream, buf)
+}
+
+func (bc *bitmapContainer) readFrom(stream io.Reader) (int, error) {
+	buf := uint64SliceAsByteSlice(bc.bitmap)
+	n, err := io.ReadFull(stream, buf)
+	bc.computeCardinality()
+	return n, err
+}
+
+func uint64SliceAsByteSlice(slice []uint64) []byte {
+	// make a new slice header
+	header := *(*reflect.SliceHeader)(unsafe.Pointer(&slice))
+
+	// update its capacity and length
+	header.Len *= 8
+	header.Cap *= 8
+
+	// return it
+	return *(*[]byte)(unsafe.Pointer(&header))
+}
+
+func uint16SliceAsByteSlice(slice []uint16) []byte {
+	// make a new slice header
+	header := *(*reflect.SliceHeader)(unsafe.Pointer(&slice))
+
+	// update its capacity and length
+	header.Len *= 2
+	header.Cap *= 2
+
+	// return it
+	return *(*[]byte)(unsafe.Pointer(&header))
+}
+
+func (bc *bitmapContainer) asLittleEndianByteSlice() []byte {
+	return uint64SliceAsByteSlice(bc.bitmap)
+}
+
+// Deserialization code follows
+
+func byteSliceAsUint16Slice(slice []byte) []uint16 {
+	if len(slice)%2 != 0 {
+		panic("Slice size should be divisible by 2")
+	}
+
+	// make a new slice header
+	header := *(*reflect.SliceHeader)(unsafe.Pointer(&slice))
+
+	// update its capacity and length
+	header.Len /= 2
+	header.Cap /= 2
+
+	// return it
+	return *(*[]uint16)(unsafe.Pointer(&header))
+}
+
+func byteSliceAsUint64Slice(slice []byte) []uint64 {
+	if len(slice)%8 != 0 {
+		panic("Slice size should be divisible by 8")
+	}
+
+	// make a new slice header
+	header := *(*reflect.SliceHeader)(unsafe.Pointer(&slice))
+
+	// update its capacity and length
+	header.Len /= 8
+	header.Cap /= 8
+
+	// return it
+	return *(*[]uint64)(unsafe.Pointer(&header))
+}
+
+func byteSliceAsInterval16Slice(slice []byte) []interval16 {
+	if len(slice)%4 != 0 {
+		panic("Slice size should be divisible by 4")
+	}
+
+	// make a new slice header
+	header := *(*reflect.SliceHeader)(unsafe.Pointer(&slice))
+
+	// update its capacity and length
+	header.Len /= 4
+	header.Cap /= 4
+
+	// return it
+	return *(*[]interval16)(unsafe.Pointer(&header))
+}

vendor/github.com/RoaringBitmap/roaring/serializationfuzz.go

@@ -0,0 +1,21 @@
+// +build gofuzz
+
+package roaring
+
+import "bytes"
+
+func FuzzSerializationStream(data []byte) int {
+	newrb := NewBitmap()
+	if _, err := newrb.ReadFrom(bytes.NewReader(data)); err != nil {
+		return 0
+	}
+	return 1
+}
+
+func FuzzSerializationBuffer(data []byte) int {
+	newrb := NewBitmap()
+	if _, err := newrb.FromBuffer(data); err != nil {
+		return 0
+	}
+	return 1
+}

vendor/github.com/RoaringBitmap/roaring/setutil.go

@@ -0,0 +1,609 @@
+package roaring
+
+func equal(a, b []uint16) bool {
+	if len(a) != len(b) {
+		return false
+	}
+	for i := range a {
+		if a[i] != b[i] {
+			return false
+		}
+	}
+	return true
+}
+
+func difference(set1 []uint16, set2 []uint16, buffer []uint16) int {
+	if 0 == len(set2) {
+		for k := 0; k < len(set1); k++ {
+			buffer[k] = set1[k]
+		}
+		return len(set1)
+	}
+	if 0 == len(set1) {
+		return 0
+	}
+	pos := 0
+	k1 := 0
+	k2 := 0
+	buffer = buffer[:cap(buffer)]
+	s1 := set1[k1]
+	s2 := set2[k2]
+	for {
+		if s1 < s2 {
+			buffer[pos] = s1
+			pos++
+			k1++
+			if k1 >= len(set1) {
+				break
+			}
+			s1 = set1[k1]
+		} else if s1 == s2 {
+			k1++
+			k2++
+			if k1 >= len(set1) {
+				break
+			}
+			s1 = set1[k1]
+			if k2 >= len(set2) {
+				for ; k1 < len(set1); k1++ {
+					buffer[pos] = set1[k1]
+					pos++
+				}
+				break
+			}
+			s2 = set2[k2]
+		} else { // if (val1>val2)
+			k2++
+			if k2 >= len(set2) {
+				for ; k1 < len(set1); k1++ {
+					buffer[pos] = set1[k1]
+					pos++
+				}
+				break
+			}
+			s2 = set2[k2]
+		}
+	}
+	return pos
+
+}
+
+func exclusiveUnion2by2(set1 []uint16, set2 []uint16, buffer []uint16) int {
+	if 0 == len(set2) {
+		buffer = buffer[:len(set1)]
+		copy(buffer, set1[:])
+		return len(set1)
+	}
+	if 0 == len(set1) {
+		buffer = buffer[:len(set2)]
+		copy(buffer, set2[:])
+		return len(set2)
+	}
+	pos := 0
+	k1 := 0
+	k2 := 0
+	s1 := set1[k1]
+	s2 := set2[k2]
+	buffer = buffer[:cap(buffer)]
+	for {
+		if s1 < s2 {
+			buffer[pos] = s1
+			pos++
+			k1++
+			if k1 >= len(set1) {
+				for ; k2 < len(set2); k2++ {
+					buffer[pos] = set2[k2]
+					pos++
+				}
+				break
+			}
+			s1 = set1[k1]
+		} else if s1 == s2 {
+			k1++
+			k2++
+			if k1 >= len(set1) {
+				for ; k2 < len(set2); k2++ {
+					buffer[pos] = set2[k2]
+					pos++
+				}
+				break
+			}
+			if k2 >= len(set2) {
+				for ; k1 < len(set1); k1++ {
+					buffer[pos] = set1[k1]
+					pos++
+				}
+				break
+			}
+			s1 = set1[k1]
+			s2 = set2[k2]
+		} else { // if (val1>val2)
+			buffer[pos] = s2
+			pos++
+			k2++
+			if k2 >= len(set2) {
+				for ; k1 < len(set1); k1++ {
+					buffer[pos] = set1[k1]
+					pos++
+				}
+				break
+			}
+			s2 = set2[k2]
+		}
+	}
+	return pos
+}
+
+func union2by2(set1 []uint16, set2 []uint16, buffer []uint16) int {
+	pos := 0
+	k1 := 0
+	k2 := 0
+	if 0 == len(set2) {
+		buffer = buffer[:len(set1)]
+		copy(buffer, set1[:])
+		return len(set1)
+	}
+	if 0 == len(set1) {
+		buffer = buffer[:len(set2)]
+		copy(buffer, set2[:])
+		return len(set2)
+	}
+	s1 := set1[k1]
+	s2 := set2[k2]
+	buffer = buffer[:cap(buffer)]
+	for {
+		if s1 < s2 {
+			buffer[pos] = s1
+			pos++
+			k1++
+			if k1 >= len(set1) {
+				copy(buffer[pos:], set2[k2:])
+				pos += len(set2) - k2
+				break
+			}
+			s1 = set1[k1]
+		} else if s1 == s2 {
+			buffer[pos] = s1
+			pos++
+			k1++
+			k2++
+			if k1 >= len(set1) {
+				copy(buffer[pos:], set2[k2:])
+				pos += len(set2) - k2
+				break
+			}
+			if k2 >= len(set2) {
+				copy(buffer[pos:], set1[k1:])
+				pos += len(set1) - k1
+				break
+			}
+			s1 = set1[k1]
+			s2 = set2[k2]
+		} else { // if (set1[k1]>set2[k2])
+			buffer[pos] = s2
+			pos++
+			k2++
+			if k2 >= len(set2) {
+				copy(buffer[pos:], set1[k1:])
+				pos += len(set1) - k1
+				break
+			}
+			s2 = set2[k2]
+		}
+	}
+	return pos
+}
+
+func union2by2Cardinality(set1 []uint16, set2 []uint16) int {
+	pos := 0
+	k1 := 0
+	k2 := 0
+	if 0 == len(set2) {
+		return len(set1)
+	}
+	if 0 == len(set1) {
+		return len(set2)
+	}
+	s1 := set1[k1]
+	s2 := set2[k2]
+	for {
+		if s1 < s2 {
+			pos++
+			k1++
+			if k1 >= len(set1) {
+				pos += len(set2) - k2
+				break
+			}
+			s1 = set1[k1]
+		} else if s1 == s2 {
+			pos++
+			k1++
+			k2++
+			if k1 >= len(set1) {
+				pos += len(set2) - k2
+				break
+			}
+			if k2 >= len(set2) {
+				pos += len(set1) - k1
+				break
+			}
+			s1 = set1[k1]
+			s2 = set2[k2]
+		} else { // if (set1[k1]>set2[k2])
+			pos++
+			k2++
+			if k2 >= len(set2) {
+				pos += len(set1) - k1
+				break
+			}
+			s2 = set2[k2]
+		}
+	}
+	return pos
+}
+
+func intersection2by2(
+	set1 []uint16,
+	set2 []uint16,
+	buffer []uint16) int {
+
+	if len(set1)*64 < len(set2) {
+		return onesidedgallopingintersect2by2(set1, set2, buffer)
+	} else if len(set2)*64 < len(set1) {
+		return onesidedgallopingintersect2by2(set2, set1, buffer)
+	} else {
+		return localintersect2by2(set1, set2, buffer)
+	}
+}
+
+func intersection2by2Cardinality(
+	set1 []uint16,
+	set2 []uint16) int {
+
+	if len(set1)*64 < len(set2) {
+		return onesidedgallopingintersect2by2Cardinality(set1, set2)
+	} else if len(set2)*64 < len(set1) {
+		return onesidedgallopingintersect2by2Cardinality(set2, set1)
+	} else {
+		return localintersect2by2Cardinality(set1, set2)
+	}
+}
+
+func intersects2by2(
+	set1 []uint16,
+	set2 []uint16) bool {
+	// could be optimized if one set is much larger than the other one
+	if (0 == len(set1)) || (0 == len(set2)) {
+		return false
+	}
+	k1 := 0
+	k2 := 0
+	s1 := set1[k1]
+	s2 := set2[k2]
+mainwhile:
+	for {
+
+		if s2 < s1 {
+			for {
+				k2++
+				if k2 == len(set2) {
+					break mainwhile
+				}
+				s2 = set2[k2]
+				if s2 >= s1 {
+					break
+				}
+			}
+		}
+		if s1 < s2 {
+			for {
+				k1++
+				if k1 == len(set1) {
+					break mainwhile
+				}
+				s1 = set1[k1]
+				if s1 >= s2 {
+					break
+				}
+			}
+
+		} else {
+			// (set2[k2] == set1[k1])
+			return true
+		}
+	}
+	return false
+}
+
+func localintersect2by2(
+	set1 []uint16,
+	set2 []uint16,
+	buffer []uint16) int {
+
+	if (0 == len(set1)) || (0 == len(set2)) {
+		return 0
+	}
+	k1 := 0
+	k2 := 0
+	pos := 0
+	buffer = buffer[:cap(buffer)]
+	s1 := set1[k1]
+	s2 := set2[k2]
+mainwhile:
+	for {
+		if s2 < s1 {
+			for {
+				k2++
+				if k2 == len(set2) {
+					break mainwhile
+				}
+				s2 = set2[k2]
+				if s2 >= s1 {
+					break
+				}
+			}
+		}
+		if s1 < s2 {
+			for {
+				k1++
+				if k1 == len(set1) {
+					break mainwhile
+				}
+				s1 = set1[k1]
+				if s1 >= s2 {
+					break
+				}
+			}
+
+		} else {
+			// (set2[k2] == set1[k1])
+			buffer[pos] = s1
+			pos++
+			k1++
+			if k1 == len(set1) {
+				break
+			}
+			s1 = set1[k1]
+			k2++
+			if k2 == len(set2) {
+				break
+			}
+			s2 = set2[k2]
+		}
+	}
+	return pos
+}
+
+func localintersect2by2Cardinality(
+	set1 []uint16,
+	set2 []uint16) int {
+
+	if (0 == len(set1)) || (0 == len(set2)) {
+		return 0
+	}
+	k1 := 0
+	k2 := 0
+	pos := 0
+	s1 := set1[k1]
+	s2 := set2[k2]
+mainwhile:
+	for {
+		if s2 < s1 {
+			for {
+				k2++
+				if k2 == len(set2) {
+					break mainwhile
+				}
+				s2 = set2[k2]
+				if s2 >= s1 {
+					break
+				}
+			}
+		}
+		if s1 < s2 {
+			for {
+				k1++
+				if k1 == len(set1) {
+					break mainwhile
+				}
+				s1 = set1[k1]
+				if s1 >= s2 {
+					break
+				}
+			}
+
+		} else {
+			// (set2[k2] == set1[k1])
+			pos++
+			k1++
+			if k1 == len(set1) {
+				break
+			}
+			s1 = set1[k1]
+			k2++
+			if k2 == len(set2) {
+				break
+			}
+			s2 = set2[k2]
+		}
+	}
+	return pos
+}
+
+func advanceUntil(
+	array []uint16,
+	pos int,
+	length int,
+	min uint16) int {
+	lower := pos + 1
+
+	if lower >= length || array[lower] >= min {
+		return lower
+	}
+
+	spansize := 1
+
+	for lower+spansize < length && array[lower+spansize] < min {
+		spansize *= 2
+	}
+	var upper int
+	if lower+spansize < length {
+		upper = lower + spansize
+	} else {
+		upper = length - 1
+	}
+
+	if array[upper] == min {
+		return upper
+	}
+
+	if array[upper] < min {
+		// means
+		// array
+		// has no
+		// item
+		// >= min
+		// pos = array.length;
+		return length
+	}
+
+	// we know that the next-smallest span was too small
+	lower += (spansize >> 1)
+
+	mid := 0
+	for lower+1 != upper {
+		mid = (lower + upper) >> 1
+		if array[mid] == min {
+			return mid
+		} else if array[mid] < min {
+			lower = mid
+		} else {
+			upper = mid
+		}
+	}
+	return upper
+
+}
+
+func onesidedgallopingintersect2by2(
+	smallset []uint16,
+	largeset []uint16,
+	buffer []uint16) int {
+
+	if 0 == len(smallset) {
+		return 0
+	}
+	buffer = buffer[:cap(buffer)]
+	k1 := 0
+	k2 := 0
+	pos := 0
+	s1 := largeset[k1]
+	s2 := smallset[k2]
+mainwhile:
+
+	for {
+		if s1 < s2 {
+			k1 = advanceUntil(largeset, k1, len(largeset), s2)
+			if k1 == len(largeset) {
+				break mainwhile
+			}
+			s1 = largeset[k1]
+		}
+		if s2 < s1 {
+			k2++
+			if k2 == len(smallset) {
+				break mainwhile
+			}
+			s2 = smallset[k2]
+		} else {
+
+			buffer[pos] = s2
+			pos++
+			k2++
+			if k2 == len(smallset) {
+				break
+			}
+			s2 = smallset[k2]
+			k1 = advanceUntil(largeset, k1, len(largeset), s2)
+			if k1 == len(largeset) {
+				break mainwhile
+			}
+			s1 = largeset[k1]
+		}
+
+	}
+	return pos
+}
+
+func onesidedgallopingintersect2by2Cardinality(
+	smallset []uint16,
+	largeset []uint16) int {
+
+	if 0 == len(smallset) {
+		return 0
+	}
+	k1 := 0
+	k2 := 0
+	pos := 0
+	s1 := largeset[k1]
+	s2 := smallset[k2]
+mainwhile:
+
+	for {
+		if s1 < s2 {
+			k1 = advanceUntil(largeset, k1, len(largeset), s2)
+			if k1 == len(largeset) {
+				break mainwhile
+			}
+			s1 = largeset[k1]
+		}
+		if s2 < s1 {
+			k2++
+			if k2 == len(smallset) {
+				break mainwhile
+			}
+			s2 = smallset[k2]
+		} else {
+
+			pos++
+			k2++
+			if k2 == len(smallset) {
+				break
+			}
+			s2 = smallset[k2]
+			k1 = advanceUntil(largeset, k1, len(largeset), s2)
+			if k1 == len(largeset) {
+				break mainwhile
+			}
+			s1 = largeset[k1]
+		}
+
+	}
+	return pos
+}
+
+func binarySearch(array []uint16, ikey uint16) int {
+	low := 0
+	high := len(array) - 1
+	for low+16 <= high {
+		middleIndex := int(uint32(low+high) >> 1)
+		middleValue := array[middleIndex]
+		if middleValue < ikey {
+			low = middleIndex + 1
+		} else if middleValue > ikey {
+			high = middleIndex - 1
+		} else {
+			return middleIndex
+		}
+	}
+	for ; low <= high; low++ {
+		val := array[low]
+		if val >= ikey {
+			if val == ikey {
+				return low
+			}
+			break
+		}
+	}
+	return -(low + 1)
+}

vendor/github.com/RoaringBitmap/roaring/shortiterator.go

@@ -0,0 +1,21 @@
+package roaring
+
+type shortIterable interface {
+	hasNext() bool
+	next() uint16
+}
+
+type shortIterator struct {
+	slice []uint16
+	loc   int
+}
+
+func (si *shortIterator) hasNext() bool {
+	return si.loc < len(si.slice)
+}
+
+func (si *shortIterator) next() uint16 {
+	a := si.slice[si.loc]
+	si.loc++
+	return a
+}

vendor/github.com/RoaringBitmap/roaring/smat.go

@@ -0,0 +1,383 @@
+// +build gofuzz
+
+/*
+# Instructions for smat testing for roaring
+
+[smat](https://github.com/mschoch/smat) is a framework that provides
+state machine assisted fuzz testing.
+
+To run the smat tests for roaring...
+
+## Prerequisites
+
+    $ go get github.com/dvyukov/go-fuzz/go-fuzz
+    $ go get github.com/dvyukov/go-fuzz/go-fuzz-build
+
+## Steps
+
+1.  Generate initial smat corpus:
+```
+    go test -tags=gofuzz -run=TestGenerateSmatCorpus
+```
+
+2.  Build go-fuzz test program with instrumentation:
+```
+    go-fuzz-build -func FuzzSmat github.com/RoaringBitmap/roaring
+```
+
+3.  Run go-fuzz:
+```
+    go-fuzz -bin=./roaring-fuzz.zip -workdir=workdir/ -timeout=200
+```
+
+You should see output like...
+```
+2016/09/16 13:58:35 slaves: 8, corpus: 1 (3s ago), crashers: 0, restarts: 1/0, execs: 0 (0/sec), cover: 0, uptime: 3s
+2016/09/16 13:58:38 slaves: 8, corpus: 1 (6s ago), crashers: 0, restarts: 1/0, execs: 0 (0/sec), cover: 0, uptime: 6s
+2016/09/16 13:58:41 slaves: 8, corpus: 1 (9s ago), crashers: 0, restarts: 1/44, execs: 44 (5/sec), cover: 0, uptime: 9s
+2016/09/16 13:58:44 slaves: 8, corpus: 1 (12s ago), crashers: 0, restarts: 1/45, execs: 45 (4/sec), cover: 0, uptime: 12s
+2016/09/16 13:58:47 slaves: 8, corpus: 1 (15s ago), crashers: 0, restarts: 1/46, execs: 46 (3/sec), cover: 0, uptime: 15s
+2016/09/16 13:58:50 slaves: 8, corpus: 1 (18s ago), crashers: 0, restarts: 1/47, execs: 47 (3/sec), cover: 0, uptime: 18s
+2016/09/16 13:58:53 slaves: 8, corpus: 1 (21s ago), crashers: 0, restarts: 1/63, execs: 63 (3/sec), cover: 0, uptime: 21s
+2016/09/16 13:58:56 slaves: 8, corpus: 1 (24s ago), crashers: 0, restarts: 1/65, execs: 65 (3/sec), cover: 0, uptime: 24s
+2016/09/16 13:58:59 slaves: 8, corpus: 1 (27s ago), crashers: 0, restarts: 1/66, execs: 66 (2/sec), cover: 0, uptime: 27s
+2016/09/16 13:59:02 slaves: 8, corpus: 1 (30s ago), crashers: 0, restarts: 1/67, execs: 67 (2/sec), cover: 0, uptime: 30s
+2016/09/16 13:59:05 slaves: 8, corpus: 1 (33s ago), crashers: 0, restarts: 1/83, execs: 83 (3/sec), cover: 0, uptime: 33s
+2016/09/16 13:59:08 slaves: 8, corpus: 1 (36s ago), crashers: 0, restarts: 1/84, execs: 84 (2/sec), cover: 0, uptime: 36s
+2016/09/16 13:59:11 slaves: 8, corpus: 2 (0s ago), crashers: 0, restarts: 1/85, execs: 85 (2/sec), cover: 0, uptime: 39s
+2016/09/16 13:59:14 slaves: 8, corpus: 17 (2s ago), crashers: 0, restarts: 1/86, execs: 86 (2/sec), cover: 480, uptime: 42s
+2016/09/16 13:59:17 slaves: 8, corpus: 17 (5s ago), crashers: 0, restarts: 1/66, execs: 132 (3/sec), cover: 487, uptime: 45s
+2016/09/16 13:59:20 slaves: 8, corpus: 17 (8s ago), crashers: 0, restarts: 1/440, execs: 2645 (55/sec), cover: 487, uptime: 48s
+
+```
+
+Let it run, and if the # of crashers is > 0, check out the reports in
+the workdir where you should be able to find the panic goroutine stack
+traces.
+*/
+
+package roaring
+
+import (
+	"fmt"
+	"sort"
+
+	"github.com/mschoch/smat"
+	"github.com/willf/bitset"
+)
+
+// fuzz test using state machine driven by byte stream.
+func FuzzSmat(data []byte) int {
+	return smat.Fuzz(&smatContext{}, smat.ActionID('S'), smat.ActionID('T'),
+		smatActionMap, data)
+}
+
+var smatDebug = false
+
+func smatLog(prefix, format string, args ...interface{}) {
+	if smatDebug {
+		fmt.Print(prefix)
+		fmt.Printf(format, args...)
+	}
+}
+
+type smatContext struct {
+	pairs []*smatPair
+
+	// Two registers, x & y.
+	x int
+	y int
+
+	actions int
+}
+
+type smatPair struct {
+	bm *Bitmap
+	bs *bitset.BitSet
+}
+
+// ------------------------------------------------------------------
+
+var smatActionMap = smat.ActionMap{
+	smat.ActionID('X'): smatAction("x++", smatWrap(func(c *smatContext) { c.x++ })),
+	smat.ActionID('x'): smatAction("x--", smatWrap(func(c *smatContext) { c.x-- })),
+	smat.ActionID('Y'): smatAction("y++", smatWrap(func(c *smatContext) { c.y++ })),
+	smat.ActionID('y'): smatAction("y--", smatWrap(func(c *smatContext) { c.y-- })),
+	smat.ActionID('*'): smatAction("x*y", smatWrap(func(c *smatContext) { c.x = c.x * c.y })),
+	smat.ActionID('<'): smatAction("x<<", smatWrap(func(c *smatContext) { c.x = c.x << 1 })),
+
+	smat.ActionID('^'): smatAction("swap", smatWrap(func(c *smatContext) { c.x, c.y = c.y, c.x })),
+
+	smat.ActionID('['): smatAction(" pushPair", smatWrap(smatPushPair)),
+	smat.ActionID(']'): smatAction(" popPair", smatWrap(smatPopPair)),
+
+	smat.ActionID('B'): smatAction(" setBit", smatWrap(smatSetBit)),
+	smat.ActionID('b'): smatAction(" removeBit", smatWrap(smatRemoveBit)),
+
+	smat.ActionID('o'): smatAction(" or", smatWrap(smatOr)),
+	smat.ActionID('a'): smatAction(" and", smatWrap(smatAnd)),
+
+	smat.ActionID('#'): smatAction(" cardinality", smatWrap(smatCardinality)),
+
+	smat.ActionID('O'): smatAction(" orCardinality", smatWrap(smatOrCardinality)),
+	smat.ActionID('A'): smatAction(" andCardinality", smatWrap(smatAndCardinality)),
+
+	smat.ActionID('c'): smatAction(" clear", smatWrap(smatClear)),
+	smat.ActionID('r'): smatAction(" runOptimize", smatWrap(smatRunOptimize)),
+
+	smat.ActionID('e'): smatAction(" isEmpty", smatWrap(smatIsEmpty)),
+
+	smat.ActionID('i'): smatAction(" intersects", smatWrap(smatIntersects)),
+
+	smat.ActionID('f'): smatAction(" flip", smatWrap(smatFlip)),
+
+	smat.ActionID('-'): smatAction(" difference", smatWrap(smatDifference)),
+}
+
+var smatRunningPercentActions []smat.PercentAction
+
+func init() {
+	var ids []int
+	for actionId := range smatActionMap {
+		ids = append(ids, int(actionId))
+	}
+	sort.Ints(ids)
+
+	pct := 100 / len(smatActionMap)
+	for _, actionId := range ids {
+		smatRunningPercentActions = append(smatRunningPercentActions,
+			smat.PercentAction{pct, smat.ActionID(actionId)})
+	}
+
+	smatActionMap[smat.ActionID('S')] = smatAction("SETUP", smatSetupFunc)
+	smatActionMap[smat.ActionID('T')] = smatAction("TEARDOWN", smatTeardownFunc)
+}
+
+// We only have one smat state: running.
+func smatRunning(next byte) smat.ActionID {
+	return smat.PercentExecute(next, smatRunningPercentActions...)
+}
+
+func smatAction(name string, f func(ctx smat.Context) (smat.State, error)) func(smat.Context) (smat.State, error) {
+	return func(ctx smat.Context) (smat.State, error) {
+		c := ctx.(*smatContext)
+		c.actions++
+
+		smatLog("  ", "%s\n", name)
+
+		return f(ctx)
+	}
+}
+
+// Creates an smat action func based on a simple callback.
+func smatWrap(cb func(c *smatContext)) func(smat.Context) (next smat.State, err error) {
+	return func(ctx smat.Context) (next smat.State, err error) {
+		c := ctx.(*smatContext)
+		cb(c)
+		return smatRunning, nil
+	}
+}
+
+// Invokes a callback function with the input v bounded to len(c.pairs).
+func (c *smatContext) withPair(v int, cb func(*smatPair)) {
+	if len(c.pairs) > 0 {
+		if v < 0 {
+			v = -v
+		}
+		v = v % len(c.pairs)
+		cb(c.pairs[v])
+	}
+}
+
+// ------------------------------------------------------------------
+
+func smatSetupFunc(ctx smat.Context) (next smat.State, err error) {
+	return smatRunning, nil
+}
+
+func smatTeardownFunc(ctx smat.Context) (next smat.State, err error) {
+	return nil, err
+}
+
+// ------------------------------------------------------------------
+
+func smatPushPair(c *smatContext) {
+	c.pairs = append(c.pairs, &smatPair{
+		bm: NewBitmap(),
+		bs: bitset.New(100),
+	})
+}
+
+func smatPopPair(c *smatContext) {
+	if len(c.pairs) > 0 {
+		c.pairs = c.pairs[0 : len(c.pairs)-1]
+	}
+}
+
+func smatSetBit(c *smatContext) {
+	c.withPair(c.x, func(p *smatPair) {
+		y := uint32(c.y)
+		p.bm.AddInt(int(y))
+		p.bs.Set(uint(y))
+		p.checkEquals()
+	})
+}
+
+func smatRemoveBit(c *smatContext) {
+	c.withPair(c.x, func(p *smatPair) {
+		y := uint32(c.y)
+		p.bm.Remove(y)
+		p.bs.Clear(uint(y))
+		p.checkEquals()
+	})
+}
+
+func smatAnd(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c.withPair(c.y, func(py *smatPair) {
+			px.bm.And(py.bm)
+			px.bs = px.bs.Intersection(py.bs)
+			px.checkEquals()
+			py.checkEquals()
+		})
+	})
+}
+
+func smatOr(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c.withPair(c.y, func(py *smatPair) {
+			px.bm.Or(py.bm)
+			px.bs = px.bs.Union(py.bs)
+			px.checkEquals()
+			py.checkEquals()
+		})
+	})
+}
+
+func smatAndCardinality(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c.withPair(c.y, func(py *smatPair) {
+			c0 := px.bm.AndCardinality(py.bm)
+			c1 := px.bs.IntersectionCardinality(py.bs)
+			if c0 != uint64(c1) {
+				panic("expected same add cardinality")
+			}
+			px.checkEquals()
+			py.checkEquals()
+		})
+	})
+}
+
+func smatOrCardinality(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c.withPair(c.y, func(py *smatPair) {
+			c0 := px.bm.OrCardinality(py.bm)
+			c1 := px.bs.UnionCardinality(py.bs)
+			if c0 != uint64(c1) {
+				panic("expected same or cardinality")
+			}
+			px.checkEquals()
+			py.checkEquals()
+		})
+	})
+}
+
+func smatRunOptimize(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		px.bm.RunOptimize()
+		px.checkEquals()
+	})
+}
+
+func smatClear(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		px.bm.Clear()
+		px.bs = px.bs.ClearAll()
+		px.checkEquals()
+	})
+}
+
+func smatCardinality(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c0 := px.bm.GetCardinality()
+		c1 := px.bs.Count()
+		if c0 != uint64(c1) {
+			panic("expected same cardinality")
+		}
+	})
+}
+
+func smatIsEmpty(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c0 := px.bm.IsEmpty()
+		c1 := px.bs.None()
+		if c0 != c1 {
+			panic("expected same is empty")
+		}
+	})
+}
+
+func smatIntersects(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c.withPair(c.y, func(py *smatPair) {
+			v0 := px.bm.Intersects(py.bm)
+			v1 := px.bs.IntersectionCardinality(py.bs) > 0
+			if v0 != v1 {
+				panic("intersects not equal")
+			}
+
+			px.checkEquals()
+			py.checkEquals()
+		})
+	})
+}
+
+func smatFlip(c *smatContext) {
+	c.withPair(c.x, func(p *smatPair) {
+		y := uint32(c.y)
+		p.bm.Flip(uint64(y), uint64(y)+1)
+		p.bs = p.bs.Flip(uint(y))
+		p.checkEquals()
+	})
+}
+
+func smatDifference(c *smatContext) {
+	c.withPair(c.x, func(px *smatPair) {
+		c.withPair(c.y, func(py *smatPair) {
+			px.bm.AndNot(py.bm)
+			px.bs = px.bs.Difference(py.bs)
+			px.checkEquals()
+			py.checkEquals()
+		})
+	})
+}
+
+func (p *smatPair) checkEquals() {
+	if !p.equalsBitSet(p.bs, p.bm) {
+		panic("bitset mismatch")
+	}
+}
+
+func (p *smatPair) equalsBitSet(a *bitset.BitSet, b *Bitmap) bool {
+	for i, e := a.NextSet(0); e; i, e = a.NextSet(i + 1) {
+		if !b.ContainsInt(int(i)) {
+			fmt.Printf("in a bitset, not b bitmap, i: %d\n", i)
+			fmt.Printf("  a bitset: %s\n  b bitmap: %s\n",
+				a.String(), b.String())
+			return false
+		}
+	}
+
+	i := b.Iterator()
+	for i.HasNext() {
+		v := i.Next()
+		if !a.Test(uint(v)) {
+			fmt.Printf("in b bitmap, not a bitset, v: %d\n", v)
+			fmt.Printf("  a bitset: %s\n  b bitmap: %s\n",
+				a.String(), b.String())
+			return false
+		}
+	}
+
+	return true
+}

vendor/github.com/RoaringBitmap/roaring/util.go

@@ -0,0 +1,315 @@
+package roaring
+
+import (
+	"math/rand"
+	"sort"
+)
+
+const (
+	arrayDefaultMaxSize        = 4096 // containers with 4096 or fewer integers should be array containers.
+	arrayLazyLowerBound        = 1024
+	maxCapacity                = 1 << 16
+	serialCookieNoRunContainer = 12346 // only arrays and bitmaps
+	invalidCardinality         = -1
+	serialCookie               = 12347 // runs, arrays, and bitmaps
+	noOffsetThreshold          = 4
+
+	// Compute wordSizeInBytes, the size of a word in bytes.
+	_m              = ^uint64(0)
+	_logS           = _m>>8&1 + _m>>16&1 + _m>>32&1
+	wordSizeInBytes = 1 << _logS
+
+	// other constants used in ctz_generic.go
+	wordSizeInBits = wordSizeInBytes << 3 // word size in bits
+)
+
+const maxWord = 1<<wordSizeInBits - 1
+
+// doesn't apply to runContainers
+func getSizeInBytesFromCardinality(card int) int {
+	if card > arrayDefaultMaxSize {
+		// bitmapContainer
+		return maxCapacity / 8
+	}
+	// arrayContainer
+	return 2 * card
+}
+
+func fill(arr []uint64, val uint64) {
+	for i := range arr {
+		arr[i] = val
+	}
+}
+func fillRange(arr []uint64, start, end int, val uint64) {
+	for i := start; i < end; i++ {
+		arr[i] = val
+	}
+}
+
+func fillArrayAND(container []uint16, bitmap1, bitmap2 []uint64) {
+	if len(bitmap1) != len(bitmap2) {
+		panic("array lengths don't match")
+	}
+	// TODO: rewrite in assembly
+	pos := 0
+	for k := range bitmap1 {
+		bitset := bitmap1[k] & bitmap2[k]
+		for bitset != 0 {
+			t := bitset & -bitset
+			container[pos] = uint16((k*64 + int(popcount(t-1))))
+			pos = pos + 1
+			bitset ^= t
+		}
+	}
+}
+
+func fillArrayANDNOT(container []uint16, bitmap1, bitmap2 []uint64) {
+	if len(bitmap1) != len(bitmap2) {
+		panic("array lengths don't match")
+	}
+	// TODO: rewrite in assembly
+	pos := 0
+	for k := range bitmap1 {
+		bitset := bitmap1[k] &^ bitmap2[k]
+		for bitset != 0 {
+			t := bitset & -bitset
+			container[pos] = uint16((k*64 + int(popcount(t-1))))
+			pos = pos + 1
+			bitset ^= t
+		}
+	}
+}
+
+func fillArrayXOR(container []uint16, bitmap1, bitmap2 []uint64) {
+	if len(bitmap1) != len(bitmap2) {
+		panic("array lengths don't match")
+	}
+	// TODO: rewrite in assembly
+	pos := 0
+	for k := 0; k < len(bitmap1); k++ {
+		bitset := bitmap1[k] ^ bitmap2[k]
+		for bitset != 0 {
+			t := bitset & -bitset
+			container[pos] = uint16((k*64 + int(popcount(t-1))))
+			pos = pos + 1
+			bitset ^= t
+		}
+	}
+}
+
+func highbits(x uint32) uint16 {
+	return uint16(x >> 16)
+}
+func lowbits(x uint32) uint16 {
+	return uint16(x & 0xFFFF)
+}
+
+const maxLowBit = 0xFFFF
+
+func flipBitmapRange(bitmap []uint64, start int, end int) {
+	if start >= end {
+		return
+	}
+	firstword := start / 64
+	endword := (end - 1) / 64
+	bitmap[firstword] ^= ^(^uint64(0) << uint(start%64))
+	for i := firstword; i < endword; i++ {
+		//p("flipBitmapRange on i=%v", i)
+		bitmap[i] = ^bitmap[i]
+	}
+	bitmap[endword] ^= ^uint64(0) >> (uint(-end) % 64)
+}
+
+func resetBitmapRange(bitmap []uint64, start int, end int) {
+	if start >= end {
+		return
+	}
+	firstword := start / 64
+	endword := (end - 1) / 64
+	if firstword == endword {
+		bitmap[firstword] &= ^((^uint64(0) << uint(start%64)) & (^uint64(0) >> (uint(-end) % 64)))
+		return
+	}
+	bitmap[firstword] &= ^(^uint64(0) << uint(start%64))
+	for i := firstword + 1; i < endword; i++ {
+		bitmap[i] = 0
+	}
+	bitmap[endword] &= ^(^uint64(0) >> (uint(-end) % 64))
+
+}
+
+func setBitmapRange(bitmap []uint64, start int, end int) {
+	if start >= end {
+		return
+	}
+	firstword := start / 64
+	endword := (end - 1) / 64
+	if firstword == endword {
+		bitmap[firstword] |= (^uint64(0) << uint(start%64)) & (^uint64(0) >> (uint(-end) % 64))
+		return
+	}
+	bitmap[firstword] |= ^uint64(0) << uint(start%64)
+	for i := firstword + 1; i < endword; i++ {
+		bitmap[i] = ^uint64(0)
+	}
+	bitmap[endword] |= ^uint64(0) >> (uint(-end) % 64)
+}
+
+func flipBitmapRangeAndCardinalityChange(bitmap []uint64, start int, end int) int {
+	before := wordCardinalityForBitmapRange(bitmap, start, end)
+	flipBitmapRange(bitmap, start, end)
+	after := wordCardinalityForBitmapRange(bitmap, start, end)
+	return int(after - before)
+}
+
+func resetBitmapRangeAndCardinalityChange(bitmap []uint64, start int, end int) int {
+	before := wordCardinalityForBitmapRange(bitmap, start, end)
+	resetBitmapRange(bitmap, start, end)
+	after := wordCardinalityForBitmapRange(bitmap, start, end)
+	return int(after - before)
+}
+
+func setBitmapRangeAndCardinalityChange(bitmap []uint64, start int, end int) int {
+	before := wordCardinalityForBitmapRange(bitmap, start, end)
+	setBitmapRange(bitmap, start, end)
+	after := wordCardinalityForBitmapRange(bitmap, start, end)
+	return int(after - before)
+}
+
+func wordCardinalityForBitmapRange(bitmap []uint64, start int, end int) uint64 {
+	answer := uint64(0)
+	if start >= end {
+		return answer
+	}
+	firstword := start / 64
+	endword := (end - 1) / 64
+	for i := firstword; i <= endword; i++ {
+		answer += popcount(bitmap[i])
+	}
+	return answer
+}
+
+func selectBitPosition(w uint64, j int) int {
+	seen := 0
+
+	// Divide 64bit
+	part := w & 0xFFFFFFFF
+	n := popcount(part)
+	if n <= uint64(j) {
+		part = w >> 32
+		seen += 32
+		j -= int(n)
+	}
+	w = part
+
+	// Divide 32bit
+	part = w & 0xFFFF
+	n = popcount(part)
+	if n <= uint64(j) {
+		part = w >> 16
+		seen += 16
+		j -= int(n)
+	}
+	w = part
+
+	// Divide 16bit
+	part = w & 0xFF
+	n = popcount(part)
+	if n <= uint64(j) {
+		part = w >> 8
+		seen += 8
+		j -= int(n)
+	}
+	w = part
+
+	// Lookup in final byte
+	var counter uint
+	for counter = 0; counter < 8; counter++ {
+		j -= int((w >> counter) & 1)
+		if j < 0 {
+			break
+		}
+	}
+	return seen + int(counter)
+
+}
+
+func panicOn(err error) {
+	if err != nil {
+		panic(err)
+	}
+}
+
+type ph struct {
+	orig int
+	rand int
+}
+
+type pha []ph
+
+func (p pha) Len() int           { return len(p) }
+func (p pha) Less(i, j int) bool { return p[i].rand < p[j].rand }
+func (p pha) Swap(i, j int)      { p[i], p[j] = p[j], p[i] }
+
+func getRandomPermutation(n int) []int {
+	r := make([]ph, n)
+	for i := 0; i < n; i++ {
+		r[i].orig = i
+		r[i].rand = rand.Intn(1 << 29)
+	}
+	sort.Sort(pha(r))
+	m := make([]int, n)
+	for i := range m {
+		m[i] = r[i].orig
+	}
+	return m
+}
+
+func minOfInt(a, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}
+
+func maxOfInt(a, b int) int {
+	if a > b {
+		return a
+	}
+	return b
+}
+
+func maxOfUint16(a, b uint16) uint16 {
+	if a > b {
+		return a
+	}
+	return b
+}
+
+func minOfUint16(a, b uint16) uint16 {
+	if a < b {
+		return a
+	}
+	return b
+}
+
+func maxInt(a, b int) int {
+	if a > b {
+		return a
+	}
+	return b
+}
+
+func maxUint16(a, b uint16) uint16 {
+	if a > b {
+		return a
+	}
+	return b
+}
+
+func minUint16(a, b uint16) uint16 {
+	if a < b {
+		return a
+	}
+	return b
+}