CentOS 7 build (#13)

mlin · Jan 5, 2021 · bf8ce0a · bf8ce0a
1 parent b9ca880
commit bf8ce0a
Show file tree

Hide file tree

Showing 6 changed files with 190 additions and 55 deletions.
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -5,6 +5,7 @@ env:
   APT_DEPS: libzstd-dev samtools tabix libhts-dev pigz python3-pip
   BREW_DEPS: python@3.8 sqlite zstd samtools pigz coreutils
   PIP_DEPS: pytest pytest-xdist
+  SQLITE_VERSION: 3340000
 
 jobs:
 
@@ -59,7 +60,7 @@ jobs:
     - name: test
       run: env -C build ctest -V
 
-  linux-so:
+  linux-x86-64-so:
     runs-on: ubuntu-20.04
     steps:
     - uses: actions/checkout@v2
@@ -70,18 +71,25 @@ jobs:
         sudo apt-get -qq update
         sudo apt-get install -y $APT_DEPS
         sudo pip3 install --system $PIP_DEPS
-        cmake -DCMAKE_BUILD_TYPE=Release -B build . && cmake --build build -j $(nproc)
+        cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_FLAGS='-D_GLIBCXX_USE_CXX11_ABI=0' -B build . && cmake --build build -j $(nproc)
         rm build/libgenomicsqlite.so
         sudo apt-get remove -y libzstd-dev  # should be linked into libgenomicsqlite.so
     - name: build portable .so in docker
-      run: docker build -t genomicsqlite . && docker run -v $(pwd):/mnt --rm genomicsqlite cp build/libgenomicsqlite.so /mnt/build/
-    - name: ldd
+      run: docker build --build-arg "SQLITE_VERSION=${SQLITE_VERSION}" -t genomicsqlite . && docker run -v $(pwd):/mnt --rm genomicsqlite cp /usr/local/lib/{libgenomicsqlite.so,libsqlite3.so.0} /mnt/build/
+    - name: symbols & digests
       run: |
-        docker run genomicsqlite ldd -v -r build/libgenomicsqlite.so
+        readelf -d build/libgenomicsqlite.so
+        ldd -r build/libgenomicsqlite.so
         echo "----"
-        ldd -v -r build/libgenomicsqlite.so
+        echo "libgenomicsqlite.so GLIBC version requirements:"
+        objdump -t build/libgenomicsqlite.so | grep -o 'GLIBC_.*' | sort -Vr | head -n1
+        objdump -t build/libgenomicsqlite.so | grep -o 'GLIBCXX_.*' | sort -Vr | head -n1
+        echo "libsqlite3.so.0 GLIBC version requirements:"
+        objdump -t build/libsqlite3.so.0 | grep -o 'GLIBC_.*' | sort -Vr | head -n1
         echo "----"
-        sha256sum build/libgenomicsqlite.so
+        sha256sum build/libgenomicsqlite.so build/libsqlite3.so.0
+        sudo cp build/libsqlite3.so.0 /usr/local/lib
+        sudo ldconfig
     - name: rust toolchain
       uses: actions-rs/toolchain@v1
       with:
@@ -96,13 +104,17 @@ jobs:
         bindings/rust/cargo test --release
         bindings/rust/cargo package --list | grep libgenomicsqlite.so
     - name: prepare artifacts
-      run: cp build/libgenomicsqlite.so include/genomicsqlite.h .
+      run: cp build/{libgenomicsqlite.so,libsqlite3.so.0} .
     - uses: actions/upload-artifact@v2
       with:
         name: libgenomicsqlite.so
         path: libgenomicsqlite.so
+    - uses: actions/upload-artifact@v2
+      with:
+        name: libsqlite3.so.0
+        path: libsqlite3.so.0
 
-  macOS-dylib:
+  macOS-x86-64-dylib:
     runs-on: macOS-10.15
     steps:
     - uses: actions/checkout@v2
@@ -111,9 +123,10 @@ jobs:
     - name: dependencies
       run: |
         rm -f /usr/local/bin/2to3  # https://github.com/pypa/pipenv/issues/3831
-        brew update
-        brew upgrade
-        brew install $BREW_DEPS
+        brew update || (sleep 2; brew update)
+        for dep in $BREW_DEPS ; do
+          brew install $dep || brew upgrade $dep
+        done
         /usr/local/bin/pip3 install $PIP_DEPS
     - name: rust toolchain
       uses: actions-rs/toolchain@v1
@@ -136,7 +149,7 @@ jobs:
               -DCMAKE_PREFIX_PATH=$(brew --prefix)/opt/sqlite \
               -B build .
         cmake --build build -j 4
-    - name: otool
+    - name: symbols & digests
       run: |
         otool -L build/libgenomicsqlite.dylib
         sha256sum build/libgenomicsqlite.dylib
@@ -160,13 +173,17 @@ jobs:
 
   artifacts:
     runs-on: ubuntu-20.04
-    needs: [linux-so, macOS-dylib]
+    needs: [linux-x86-64-so, macOS-x86-64-dylib]
     steps:
     - uses: actions/checkout@v2
     - name: fetch libgenomicsqlite.so
       uses: actions/download-artifact@v2
       with:
         name: libgenomicsqlite.so
+    - name: fetch libsqlite3.so.0
+      uses: actions/download-artifact@v2
+      with:
+        name: libsqlite3.so.0
     - name: fetch libgenomicsqlite.dylib
       uses: actions/download-artifact@v2
       with:
@@ -188,7 +205,7 @@ jobs:
         cp libgenomicsqlite.so libgenomicsqlite.dylib bindings/jdbc/genomicsqlite-jdbc/src/main/resources/
         mvn deploy -Drevision=${{ env.GIT_REVISION }} -B -f bindings/jdbc/genomicsqlite-jdbc/pom.xml
     - name: digests
-      run: sha256sum *genomicsqlite* bindings/jdbc/genomicsqlite-jdbc/target/genomicsqlite*.jar
+      run: sha256sum *genomicsqlite* bindings/jdbc/genomicsqlite-jdbc/target/genomicsqlite*.jar libsqlite3.so.0
     - uses: actions/upload-artifact@v2
       with:
         name: genomicsqlite-jdbc-${{ env.GIT_REVISION }}

diff --git a/Dockerfile b/Dockerfile
@@ -1,19 +1,36 @@
-# Builds libgenomicsqlite.so in old ubuntu (16.04) to maximize portability. Also bakes in Zstandard.
-# This recipe works on 14.04 too, but there are C++ symbol differences before & after g++ 5.1:
-#   https://gcc.gnu.org/onlinedocs/gcc-5.2.0/libstdc++/manual/manual/using_dual_abi.html
-FROM ubuntu:16.04
+# Builds libgenomicsqlite.so in CentOS 7 (+ devtoolset-8) to maximize compatibility. Bakes in
+# Zstandard, and also builds a compatible libsqlite3.so.0 for users unable/unwilling to upgrade
+# their host's.
+FROM centos:7 as builder
 
 ARG CMAKE_VERSION=3.17.3
-ARG SQLITE_VERSION=3310000
-ARG ZSTD_VERSION=1.4.7
-ARG CPU_ARCH=haswell
-
+ARG SQLITE_VERSION=3340000
+ARG ZSTD_VERSION=1.4.8
+ARG CPU_ARCH=ivybridge
 ENV CFLAGS="-march=${CPU_ARCH} -O3"
 ENV CXXFLAGS=${CFLAGS}
+# https://www.sqlite.org/compile.html
+ENV SQLITE_CFLAGS="\
+        -DSQLITE_ENABLE_LOAD_EXTENSION \
+        -DSQLITE_USE_URI \
+        -DSQLITE_DQS=0 \
+        -DSQLITE_LIKE_DOESNT_MATCH_BLOBS \
+        -DSQLITE_DEFAULT_MEMSTATUS=0 \
+        -DSQLITE_MAX_EXPR_DEPTH=0 \
+        -DSQLITE_ENABLE_NULL_TRIM \
+        -DSQLITE_USE_ALLOCA \
+        -DSQLITE_HAVE_ISNAN \
+        -DSQLITE_ENABLE_UPDATE_DELETE_LIMIT \
+        -DSQLITE_ENABLE_COLUMN_METADATA \
+        -DSQLITE_ENABLE_DBSTAT_VTAB \
+        -DSQLITE_ENABLE_FTS5 \
+        -DSQLITE_ENABLE_RTREE \
+        -DSQLITE_ENABLE_PREUPDATE_HOOK \
+        -DSQLITE_ENABLE_SESSION \
+"
 
-# apt
-RUN apt-get -qq update && DEBIAN_FRONTEND=noninteractive apt-get -qq install -y \
-        build-essential wget zip rsync git-core
+RUN yum install -y -q wget unzip rsync git centos-release-scl
+RUN yum install -y -q devtoolset-8-gcc devtoolset-8-gcc-c++ devtoolset-8-make
 
 RUN mkdir -p /work/GenomicSQLite
 
@@ -22,16 +39,16 @@ WORKDIR /work
 RUN wget -nv https://www.sqlite.org/2020/sqlite-amalgamation-${SQLITE_VERSION}.zip \
         && unzip -o sqlite-amalgamation-${SQLITE_VERSION}.zip
 WORKDIR /work/sqlite-amalgamation-${SQLITE_VERSION}
-RUN gcc -shared -o libsqlite3.so.0 -fPIC -shared -Wl,-soname,libsqlite3.so.0 -g ${CFLAGS} sqlite3.c
-RUN gcc -o sqlite3 -g ${CFLAGS} sqlite3.c shell.c -lpthread -ldl
+RUN bash -c "scl enable devtoolset-8 'gcc -shared -o libsqlite3.so.0 -fPIC -shared -Wl,-soname,libsqlite3.so.0 -g ${CFLAGS} ${SQLITE_CFLAGS} sqlite3.c' & pid=$? \
+             scl enable devtoolset-8 'gcc -o sqlite3 -g ${CFLAGS} ${SQLITE_CFLAGS} sqlite3.c shell.c -lpthread -ldl -lm' && wait $pid"
 RUN cp libsqlite3.so.0 /usr/local/lib && cp *.h /usr/local/include && cp sqlite3 /usr/local/bin
 RUN ln -s /usr/local/lib/libsqlite3.so.0 /usr/local/lib/libsqlite3.so
 
 # Zstandard -- hacked with -fPIC for use with ZSTD_WHOLE_ARCHIVE
 WORKDIR /work
 RUN wget -nv -O - https://github.com/facebook/zstd/releases/download/v${ZSTD_VERSION}/zstd-${ZSTD_VERSION}.tar.gz | tar zx
 WORKDIR /work/zstd-${ZSTD_VERSION}
-RUN CFLAGS="${CFLAGS} -fPIC" make install -j $(nproc)
+RUN scl enable devtoolset-8 "CFLAGS='${CFLAGS} -fPIC' make install -j $(nproc)"
 
 # CMake
 WORKDIR /work
@@ -44,7 +61,83 @@ RUN ldconfig
 # libgenomicsqlite.so
 ADD . /work/GenomicSQLite
 WORKDIR /work/GenomicSQLite
-RUN rm -rf build && cmake -DCMAKE_BUILD_TYPE=Release -DZSTD_WHOLE_ARCHIVE=true -B build . && cmake --build build --target genomicsqlite -j $(nproc)
-RUN ldd -v -r build/libgenomicsqlite.so
-RUN sqlite3 -cmd '.load build/libgenomicsqlite.so' :memory: 'select genomicsqlite_version()'
-RUN sha256sum build/libgenomicsqlite.so
+RUN rm -rf build
+RUN scl enable devtoolset-8 "cmake -DCMAKE_BUILD_TYPE=Release -DZSTD_WHOLE_ARCHIVE=true -B build . && cmake --build build --target genomicsqlite -j $(nproc)"
+RUN sqlite3 -cmd '.load build/libgenomicsqlite.so' :memory: 'select sqlite_version(); select genomicsqlite_version()'
+
+
+
+
+###################################################################################################
+# Run-up in fresh centos:7, to confirm no dependencies crept into the .so's
+
+FROM centos:7
+
+COPY --from=builder /usr/local/bin/sqlite3 /usr/local/bin/
+COPY --from=builder /usr/local/include/sqlite3.h /usr/local/include/
+
+# NOTE: /usr/local/lib is NOT a default ld path in CentOS -- https://unix.stackexchange.com/q/356624
+RUN mkdir -p /work/lib
+WORKDIR /work
+COPY --from=builder /usr/local/lib/libsqlite3.so.0 ./lib/
+COPY --from=builder /work/GenomicSQLite/build/libgenomicsqlite.so ./
+# the following approximates how bindings will usually load the extension at runtime:
+#     - libsqlite3.so.0 in the ld search path (if not already resident in the process)
+#     - given absolute path to libgenomicsqlite3.so
+RUN LD_LIBRARY_PATH=$(pwd)/lib sqlite3 -cmd ".load $(pwd)/libgenomicsqlite.so" :memory: 'select sqlite_version(); select genomicsqlite_version()'
+        # if troubleshooting that: set LD_DEBUG=libs for extremely detailed dlopen() logs
+
+# now try capi_smoke_test, with some hoops to ensure it's not dependent on implicit RPATHs
+RUN yum install -y -q gcc && gcc -v
+RUN mv libgenomicsqlite.so lib/ && ln -s libsqlite3.so.0 lib/libsqlite3.so
+ADD ./test/capi_smoke_test.c ./include/genomicsqlite.h ./
+RUN gcc -o genomicsqlite_capi_smoke_test ${CFLAGS} -L$(pwd)/lib -Wl,-rpath,\$ORIGIN -Wl,-z,origin \
+        capi_smoke_test.c -lgenomicsqlite -lsqlite3 \
+        && readelf -d genomicsqlite_capi_smoke_test
+RUN mv lib/{libgenomicsqlite.so,libsqlite3.so.0} . && rm -rf lib
+RUN ./genomicsqlite_capi_smoke_test
+        # ^ we didn't need LD_LIBRARY_PATH even though the .so's aren't at their linktime locations
+
+# test rust bindings too
+FROM centos:7
+
+RUN yum install -y -q gcc git
+ADD https://sh.rustup.rs /usr/local/bin/rustup-init.sh
+RUN chmod +x /usr/local/bin/rustup-init.sh && rustup-init.sh -y
+ENV PATH=${PATH}:/root/.cargo/bin
+
+RUN mkdir -p /work/lib
+WORKDIR /work
+COPY --from=builder /usr/local/lib/libsqlite3.so.0 ./lib/
+RUN ln -s libsqlite3.so.0 lib/libsqlite3.so
+ADD ./.git ./.git
+ADD ./bindings/rust ./rust
+COPY --from=builder /work/GenomicSQLite/build/libgenomicsqlite.so ./rust/
+RUN LD_LIBRARY_PATH=$(pwd)/lib LIBRARY_PATH=$(pwd)/lib rust/cargo test --release
+
+###################################################################################################
+# Run-up in ubuntu 16.04
+
+FROM ubuntu:16.04
+
+COPY --from=builder /usr/local/bin/sqlite3 /usr/local/bin/
+COPY --from=builder /usr/local/include/sqlite3.h /usr/local/include/
+COPY --from=builder /usr/local/lib/libsqlite3.so.0 /work/GenomicSQLite/build/libgenomicsqlite.so /usr/local/lib/
+RUN ln -s libsqlite3.so.0 /usr/local/lib/libsqlite3.so
+
+RUN sqlite3 -cmd '.load /usr/local/lib/libgenomicsqlite.so' :memory: 'select sqlite_version(); select genomicsqlite_version()'
+
+RUN mkdir /work
+WORKDIR /work
+
+RUN apt-get -qq update && apt-get install -qq -y build-essential && gcc -v
+ADD ./test/capi_smoke_test.c ./include/genomicsqlite.h /work/
+RUN gcc -o genomicsqlite_capi_smoke_test ${CFLAGS} capi_smoke_test.c -lgenomicsqlite -lsqlite3
+RUN ./genomicsqlite_capi_smoke_test
+
+# display dependencies & digests in docker build log
+RUN ldd -v -r /usr/local/lib/libgenomicsqlite.so
+RUN objdump -t /usr/local/lib/libgenomicsqlite.so | grep -o 'GLIBC_.*' | sort -Vr | head -n1
+RUN objdump -t /usr/local/lib/libgenomicsqlite.so | grep -o 'GLIBCXX_.*' | sort -Vr | head -n1
+RUN objdump -t /usr/local/lib/libsqlite3.so.0 | grep -o 'GLIBC_.*' | sort -Vr | head -n1
+RUN sha256sum /usr/local/lib/libgenomicsqlite.so /usr/local/lib/libsqlite3.so.0
diff --git a/docs/guide.md b/docs/guide.md
@@ -42,6 +42,13 @@ The Genomics Extension integrates with your programming language's existing SQLi
     Link the program to `sqlite3` and `genomicsqlite` libraries; optionally,
     [SQLiteCpp](https://github.com/SRombauts/SQLiteCpp).
 
+    GNU/Linux: to link the prebuilt `libgenomicsqlite.so` distributed from our GitHub Releases, you
+    may have to compile your source with `CXXFLAGS=-D_GLIBCXX_USE_CXX11_ABI=0`. This is because the
+    library is built against an old libstdc++ version to improve runtime compatibility. The
+    function of this flag is explained in the libstdc++ docs on
+    [Dual ABI](https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html). If you build
+    `libgenomicsqlite.so` from source, then the flag will not be needed.
+
     General note: GenomicSQLite C++ routines are liable to throw exceptions.
 
 === "C"

diff --git a/docs/index.md b/docs/index.md
@@ -23,13 +23,14 @@ Our **[Colab notebook](https://colab.research.google.com/drive/1OlHPOcRQBhDmEnS1
 
 **To use the Genomics Extension you might first need to upgrade SQLite itself.** The host program must link [SQLite version 3.31.0 (2020-01-22)](https://www.sqlite.org/releaselog/3_31_0.html) or newer. In your shell, `sqlite3 --version` displays the version installed with your OS, which is probably what your programs use; if in doubt, cause a program to report the result of `SELECT sqlite_version()`.
 
-If this is too old, then upgrade SQLite3 using your preferred binary package manager (e.g. apt, yum, conda, brew), if possible. Otherwise, modify your program's linking step or runtime environment to cause it to use an up-to-date version, for example by setting rpath or LD_LIBRARY_PATH to the location of an up-to-date shared library file. Resources:
+If this is too old, then upgrade SQLite3 using your preferred binary package manager (e.g. apt, yum, conda, brew), if possible. Otherwise, modify your program's linking step or runtime environment to cause it to use an up-to-date version, for example by setting rpath or `LD_LIBRARY_PATH` to the location of an up-to-date shared library file. Resources:
 
 * [How To Compile SQLite](https://www.sqlite.org/howtocompile.html)
 * [DreamHost Knowledge Base - Installing a custom version of SQLite3](https://help.dreamhost.com/hc/en-us/articles/360028047592-Installing-a-custom-version-of-SQLite3)
 * [Rpmfind: libsqlite3](https://rpmfind.net/linux/rpm2html/search.php?query=libsqlite3&submit=Search+...&system=&arch=)
 * [Sqlite :: Anaconda Cloud](https://anaconda.org/anaconda/sqlite)
 * Homebrew [formula/sqlite](https://formulae.brew.sh/formula/sqlite), [formula-linux/sqlite](https://formulae.brew.sh/formula-linux/sqlite)
+* As a last resort for GNU/Linux, our [GitHub Releases](https://github.com/mlin/GenomicSQLite/releases) include a `libsqlite3.so.0` that should be compatible with modern (2016+) hosts.
 
 You can always `SELECT sqlite_version()` to verify the upgrade in your program.
 
@@ -82,6 +83,13 @@ It's usually easiest to obtain the extension as a pre-compiled shared library (L
 
     Recommendation: *also* install the Python package, which includes a useful command-line shell and smoke-test script.
 
+    GNU/Linux: to link the prebuilt `libgenomicsqlite.so` distributed from our GitHub Releases, you
+    may have to compile your source with `CXXFLAGS=-D_GLIBCXX_USE_CXX11_ABI=0`. This is because the
+    library is built against an old libstdc++ version to improve runtime compatibility. The
+    function of this flag is explained in the libstdc++ docs on
+    [Dual ABI](https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html). If you build
+    `libgenomicsqlite.so` from source, then the flag will not be needed.
+
 See our [GitHub README](https://github.com/mlin/GenomicSQLite) for the source build procedure, if needed.
 
 We welcome community contributions to the available language bindings; see the Language Bindings Guide if interested.