[Bug] libpulsarwithdeps.a misses some symbols on Linux or macOS

[BewareMyPower]

Search before asking

• I searched in the issues and found nothing similar.

Version

https://dist.apache.org/repos/dist/dev/pulsar/pulsar-client-cpp/pulsar-client-cpp-3.8.0-candidate-2/

Minimal reproduce step

Download the macos-arm64.zip and follow the guide here to verify it.

What did you expect to see?

The application run successfully

What did you see instead?

% clang++ main.cc ./lib/libpulsarwithdeps.a -std=c++11 -I./include
Undefined symbols for architecture arm64:
  "_bitset_resize", referenced from:
      _roaring_bitmap_to_bitset in libpulsarwithdeps.a[1287](roaring.c.o)
      _roaring_bitmap_to_bitset in libpulsarwithdeps.a[1287](roaring.c.o)
ld: symbol(s) not found for architecture arm64

Anything else?

This should be caused by the static library of roaring is not included in the *.a file.

This dependency was introduced since #497, so it's a regression

Are you willing to submit a PR?

• I'm willing to submit a PR!

[BewareMyPower]

The debian package also misses the symbol.

# g++ main.cc  -std=c++11 /usr/lib/libpulsarwithdeps.a
/usr/bin/ld: /usr/lib/libpulsarwithdeps.a(roaring.c.o): in function `roaring_bitmap_to_bitset':
roaring.c:(.text+0xe79c): undefined reference to `bitset_resize'
collect2: error: ld returned 1 exit status

The root cause is we use ar -x to uncompress static libraries to object files (*.o). However, libroaring has duplicated bitset.c.o. For example, on my macOS laptop,

% ar t vcpkg/packages/roaring_arm64-osx/lib/libroaring.a
__.SYMDEF SORTED
isadetection.c.o
array_util.c.o
bitset_util.c.o
art.c.o
bitset.c.o
array.c.o
bitset.c.o

• https://github.com/RoaringBitmap/CRoaring/blob/master/src/bitset.c
• https://github.com/RoaringBitmap/CRoaring/blob/master/src/containers/bitset.c

After ar -x, only one bitset.c.o is retained, which means the other one is lost

[BewareMyPower]

The workaround is not to use ar -x to extract *.a files to *.o files. For example, on macOS, the following trick works:

diff --git a/build-support/merge_archives.sh b/build-support/merge_archives.sh
index 065130d..d7e3c53 100755
--- a/build-support/merge_archives.sh
+++ b/build-support/merge_archives.sh
@@ -20,6 +20,7 @@

 set -e

+ROOT="$pwd/.."
 MERGED_LIBRARY=$1
 shift
 ITEMS=$*
@@ -44,7 +45,9 @@ for ITEM in $ITEMS; do
     LIBS="$LIBS $LIB_NAME/*.o"
     cd ..
 done
+LIBS="$LIBS $ROOT/vcpkg/buildtrees/roaring/arm64-osx-dbg/src/CMakeFiles/roaring.dir/bitset.c.o"

+set -x
 ar -qc $MERGED_LIBRARY $LIBS

P.S. Change dbg to rel in release.

However, it's a bit complicated that we need to write a new script to scan object files (*.o) from dependencies directly under the vcpkg/buildtrees directory. In addition, the build on Windows is different that I don't have time to verify if the Windows library has the same issue.

A quick fix is to remove the libroaring dependency and use an alternative for Roaring64Map from libroaring for now. /cc @shibd