[llvm-ranlib] crash when called on emscripten .a file produced by GHC's new JS backend

[sternenseemann]

When compiling just about any Haskell library using GHC's new JS backend which involves creating an ar file, llvm-ranlib is called (via Emscripten's emranlib), triggering the following crash:

LLVM ERROR: malformed uleb128, extends past end
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.	Program arguments: /nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib test.a
 #0 0x00007f808df7afe4 PrintStackTraceSignalHandler(void*) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0xeaefe4)
 #1 0x00007f808df7866b SignalHandler(int) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0xeac66b)
 #2 0x00007f808cbe0bf0 __restore_rt (/nix/store/lqz6hmd86viw83f9qll2ip87jhb7p1ah-glibc-2.35-224/lib/libc.so.6+0x3dbf0)
 #3 0x00007f808cc2dbc7 __pthread_kill_implementation (/nix/store/lqz6hmd86viw83f9qll2ip87jhb7p1ah-glibc-2.35-224/lib/libc.so.6+0x8abc7)
 #4 0x00007f808cbe0b46 gsignal (/nix/store/lqz6hmd86viw83f9qll2ip87jhb7p1ah-glibc-2.35-224/lib/libc.so.6+0x3db46)
 #5 0x00007f808cbcb4b5 abort (/nix/store/lqz6hmd86viw83f9qll2ip87jhb7p1ah-glibc-2.35-224/lib/libc.so.6+0x284b5)
 #6 0x00007f808de80314 llvm::report_fatal_error(llvm::Twine const&, bool) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0xdb4314)
 #7 0x00007f808de8046a (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0xdb446a)
 #8 0x00007f808fc1ffd1 readULEB128(llvm::object::WasmObjectFile::ReadContext&) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0x2b53fd1)
 #9 0x00007f808fc2efd7 llvm::object::WasmObjectFile::WasmObjectFile(llvm::MemoryBufferRef, llvm::Error&) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0x2b62fd7)
#10 0x00007f808fc2f442 llvm::object::ObjectFile::createWasmObjectFile(llvm::MemoryBufferRef) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0x2b63442)
#11 0x00007f808fc13fbe llvm::object::ObjectFile::createObjectFile(llvm::MemoryBufferRef, llvm::file_magic, bool) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0x2b47fbe)
#12 0x00007f808fc17879 llvm::object::SymbolicFile::createSymbolicFile(llvm::MemoryBufferRef, llvm::file_magic, llvm::LLVMContext*, bool) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0x2b4b879)
#13 0x00007f808fb79066 llvm::writeArchiveToStream(llvm::raw_ostream&, llvm::ArrayRef<llvm::NewArchiveMember>, bool, llvm::object::Archive::Kind, bool, bool) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0x2aad066)
#14 0x00007f808fb7b29f llvm::writeArchive(llvm::StringRef, llvm::ArrayRef<llvm::NewArchiveMember>, bool, llvm::object::Archive::Kind, bool, bool, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer> >) (/nix/store/338k2wvhmwgkw3ipqd8f9rrym4sg1ipj-llvm-14.0.6-lib/lib/libLLVM-14.so+0x2aaf29f)
#15 0x000000000040c1bc performWriteOperation(ArchiveOperation, llvm::object::Archive*, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer> >, std::vector<llvm::NewArchiveMember, std::allocator<llvm::NewArchiveMember> >*) (/nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib+0x40c1bc)
#16 0x000000000040e96d performOperation(ArchiveOperation, llvm::object::Archive*, std::unique_ptr<llvm::MemoryBuffer, std::default_delete<llvm::MemoryBuffer> >, std::vector<llvm::NewArchiveMember, std::allocator<llvm::NewArchiveMember> >*) (/nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib+0x40e96d)
#17 0x000000000040eb13 performOperation(ArchiveOperation, std::vector<llvm::NewArchiveMember, std::allocator<llvm::NewArchiveMember> >*) (/nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib+0x40eb13)
#18 0x0000000000406a5c main (/nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib+0x406a5c)
#19 0x00007f808cbcc24e __libc_start_call_main (/nix/store/lqz6hmd86viw83f9qll2ip87jhb7p1ah-glibc-2.35-224/lib/libc.so.6+0x2924e)
#20 0x00007f808cbcc309 __libc_start_main@GLIBC_2.2.5 (/nix/store/lqz6hmd86viw83f9qll2ip87jhb7p1ah-glibc-2.35-224/lib/libc.so.6+0x29309)
#21 0x0000000000407d75 _start (/nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib+0x407d75)

emranlib is just a wrapper around llvm-ranlib that transparently passes through all arguments to llvm-ranlib. /nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib is just a symlink to normal llvm-ranlib 14.0.6:

> realpath /nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib
/nix/store/p8wpdbr2m12ydclalh33sf8mlygxgh11-llvm-14.0.6/bin/llvm-ar
> /nix/store/b4z1yykly634q6ric5d3w6162slxzrbf-emscripten-llvm-3.1.24/bin/llvm-ranlib --version
LLVM (http://llvm.org/):
  LLVM version 14.0.6
  Optimized build.
  Default target: x86_64-unknown-linux-gnu
  Host CPU: skylake

It should be easy to reproduce the crash by running on llvm-ranlib on the archive file produced by GHC (had to upload externally due to GitHub's file size limit) by just executing llvm-ranlib repro-ranlib-crash.a.

llvm-ranlib comes from nixpkgs, of course.

Reproducing this using just GHC

1. You need to build GHC as a cross-compiler for javascript-unknown-ghcjs (for older revisions js-unknown-ghcjs) – using LLVM 14.0.6 of course. Build instructions are available upstream. I used GHC revision a203ad854ffee802e6bf0aca26e6c9a99bec3865.
   If you want to use Nix, you can checkout e23efa222d1d7af7fa4380a392afd359f9e5116c of my nixpkgs fork and build nix-build -A pkgsCross.ghcjs.buildPackages.haskell.compiler.ghcHEAD.
2. Create the following Test.hs:

   module Test (test) where
    
   test x = x + 41

3. Run javascript-unknown-ghcjs-ghc -outputdir build/ -v -o repro-ranlib-crash.a --make -staticlib Test.hs. You should see the crash already.
4. You can reproduce the crash again by running llvm-ranlib repro-ranlib-crash.a.

Reproducible with:

• 14.0.6
• 15.0.7
• main (85dbb3f)

[llvmbot]

@llvm/issue-subscribers-tools-llvm-ar-llvm-ranlib

[EugeneZelenko]

Could you please try 16 or main branch?

[sternenseemann]

The same crash happens with llvm-ranlib built from 85dbb3f.

[sternenseemann]

GHC ticket

[voidus]

Another repro nix build 'github:nixos/nixpkgs?rev=6c43a3495a11e261e5f41e5d7eda2d71dae1b2fe#legacyPackages.x86_64-linux.pkgsCross.ghcjs.haskell.packages.ghc96.hashable'

[danwdart]

This is just because libraries can't be compiled by js backend yet without them being executables.