Skip to content

Test not running to completion on aarch64-linux #31

Open

Description

@zimward

The test that was introduced in 97724a424116e4401ac8da7878699885352c46ac

./src/splitcode --assign --lift ./func_tests/vcf_validation.fa.gz ./func_tests/test_1.vcf.gz CAST_EiJ --kmer-length=31 --kmer-output=./func_tests/test.kmers.1.txt
[fai_load] build FASTA index.

does never run to completion on aarch64-linux systems. The process is stuck at 100% cpu core usage.
Here is the non-changing part of the stack-trace:

#5  0x0000000000472ea0 in fai_build_core (bgzf=bgzf@entry=0x571c90) at faidx.c:103
#6  0x00000000004732dc in fai_build (fn=fn@entry=0x571970 ".//splitcode.lift.0h2kxalMGgWUjpbVk",
    fai_file=fai_file@entry=0x571c00 ".//splitcode.lift.bBityyoVsF9AmnlRV.fai") at faidx.c:189
#7  0x000000000047366c in fai_load (fn=0x571970 ".//splitcode.lift.0h2kxalMGgWUjpbVk",
    fai_file=0x571c00 ".//splitcode.lift.bBityyoVsF9AmnlRV.fai") at faidx.c:270
#8  0x00000000004524d8 in LiftWorkflow::modify_fasta (this=this@entry=0x7fffffa2a0)
    at /nix/store/qbvp2h1f5a03nwmw0nwqimnk4kjhzv19-gcc-14-20241116/include/c++/14-20241116/bits/basic_string.h:227
#9  0x00000000004227c8 in ParseOptions (argc=argc@entry=8, argv=argv@entry=0x7fffffc758, opt=...)
    at /home/zimward/splitcode/src/main.cpp:748
#10 0x000000000041a294 in main (argc=8, argv=0x7fffffc758) at /home/zimward/splitcode/src/main.cpp:1483

Activity

  1. Yenaled commented on Jan 8, 2025

    @Yenaled
    Collaborator

    It probably has to do with zlib-ng (now enabled by default) and should run if you compile splitcode with: cmake .. ZLIBNG_DISABLE=ON -BUILD_FUNCTESTING=ON

    I'll see if I can fix it.

  2. zimward commented on Jan 8, 2025

    @zimward
    ContributorAuthor

    The behavior doesn't change when built without zlib-ng.

  3. Yenaled commented on Jan 8, 2025

    @Yenaled
    Collaborator

    Thanks for checking — looks like I’ll need to look into it some more. A bit difficult since it works on all my systems.

  4. self-assigned this
    on Jan 8, 2025
  5. zimward commented on Jan 8, 2025

    @zimward
    ContributorAuthor

    The way in which it's broken seems quite strange indeed as it passes most platforms (NixOS/nixpkgs#370167 (comment))
    I have yet to try if it also happens using binfmt emulation (qemu aarch64 emulation) but it probably does which should make debugging easier

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions