Tests fail with file-5.47

[orlitzky]

file-5.47 was released a few days ago: https://mailman.astron.com/pipermail/file/2026-February/001570.html

Given the limited scope of the changes, I feel like I should be able to identify the problem, but so far I haven't had any luck:

t/oo-api.t .............................. 
        #   Failed test 'describe_contents /home/mjo/src/File-LibMagic-1.23/t/samples/tiny.pdf'
        #   at t/oo-api.t line 137.
        #          got: PDF document, version %.%
        #     expected: any one of ...
        #              PDF document, version 1.4

        #   Failed test 'describe_contents /home/mjo/src/File-LibMagic-1.23/t/samples/tiny.pdf as ref'
        #   at t/oo-api.t line 137.
        #          got: PDF document, version %.%
        #     expected: any one of ...
        #              PDF document, version 1.4
        # Looks like you failed 2 tests of 6.

    #   Failed test 'old API with /home/mjo/src/File-LibMagic-1.23/t/samples/tiny.pdf'
    #   at t/oo-api.t line 138.

            #   Failed test 'description'
            #   at t/oo-api.t line 214.
            #          got: PDF document, version %.%
            #     expected: any one of ...
            #              PDF document, version 1.4
            # Looks like you failed 1 test of 4.

        #   Failed test 'info_from_string'
        #   at t/oo-api.t line 216.

            #   Failed test 'description'
            #   at t/oo-api.t line 225.
            #          got: PDF document, version %.%
            #     expected: any one of ...
            #              PDF document, version 1.4
            # Looks like you failed 1 test of 4.

        #   Failed test 'info_from_string as ref'
        #   at t/oo-api.t line 227.

            #   Failed test 'description'
            #   at t/oo-api.t line 233.
            #          got: PDF document, version %.%
            #     expected: any one of ...
            #              PDF document, version 1.4
            # Looks like you failed 1 test of 5.

        #   Failed test 'info_from_handle'
        #   at t/oo-api.t line 241.

            #   Failed test 'description'
            #   at t/oo-api.t line 252.
            #          got: PDF document, version %.%
            #     expected: any one of ...
            #              PDF document, version 1.4
            # Looks like you failed 1 test of 4.

        #   Failed test 'info_from_handle - handle from scalar ref'
        #   at t/oo-api.t line 256.
        # Looks like you failed 4 tests of 5.

    #   Failed test 'new API with /home/mjo/src/File-LibMagic-1.23/t/samples/tiny.pdf'
    #   at t/oo-api.t line 143.
    # Looks like you failed 2 tests of 6.
t/oo-api.t .............................. 1/? 
#   Failed test 'standard magic file'
#   at t/oo-api.t line 46.
# Looks like you failed 1 test of 3.

[autarch]

This is odd. It looks like a bug in the file command's output.

[orlitzky]

The standalone "file" actually looks OK,

$ file t/samples/tiny.pdf
t/samples/tiny.pdf: PDF document, version 1.4

which makes this more interesting (i.e. harder to debug). I looked through the entire file-5.46 -> file-5.47 diff and nothing stood out to me. If it was easy I would have reported a solution instead of the problem :)

[autarch]

Oh, I misread the test failure. Yeah, it seems like something weird on the Perl side.

I only have file 5.45 on my Ubuntu system. How did you test with the new file version?

[orlitzky]

The new version is available on Gentoo linux. It could probably also be compiled by hand, though you may have to install it to e.g. /usr/local rather than use LD_LIBRARY_PATH hacks so that it finds the correct magic files.

[orlitzky]

I just compiled and installed both versions of file and

# Works
$ LD_LIBRARY_PATH="/home/mjo/tmp/file-5.46/lib/" make test

# Doesn't work
$ LD_LIBRARY_PATH="/home/mjo/tmp/file-5.47/lib/" make test

with the latter giving me a lot of

magic_load((null)): libmagic could not find any valid magic files! at /home/mjo/src/File-LibMagic-1.23/blib/lib/File/LibMagic.pm line 101.

So that's... interesting.

[autarch]

So it looks like this might be a bug in libmagic 5.47. There are a few bug reports about regressions in 5.47 that seems relevant. And Claude wrote some C code to test the magic_buffer function and seemed to find that this was a libmagic bug in the magic_buffer func.

[orlitzky]

It looks like file/file@08ab5f6 is the culprit. If I revert that, the tests pass again.

[autarch]

Yes, Claude agrees with you. I've been sick for a few days. If you're up for reporting this to the file bug tracker, that'd be great. Otherwise, I will probably get around to it later.

[orlitzky]

No problem: https://bugs.astron.com/view.php?id=735