Check for a branded ELF note when OS/ABI is NONE #1344
Conversation
Currently the auditor is only checking for the correct OS/ABI and producing a warning on FreeBSD when it doesn't match. This warning is incorrect on AArch64 though, as FreeBSD instead uses an ELF note to declare the OS/ABI. The change implemented here looks for such a note when the OS/ABI in the ELF header is NONE.
|
I haven't looked at the implementation yet, but I want to quickly answer the question
Yeah, the problem is that we don't know (or forgot? who knows) how libraries with the wrong OS/ABI field were generated on FreeBSD. An alternative should be to compile a simple dummy library (something like |
|
Ah rip, the patchelf we ship is old and doesn't have |
|
Sigh, never mind the test then, we can't upgrade patchelf because newer version is broken (JuliaPackaging/Yggdrasil#7728 (comment), caused by NixOS/patchelf#492) |
|
Finally, 168 days after this PR was first opened, it actually works, is tested, and passes CI. |
| end | ||
|
|
||
| os_from_elf_note(::ObjectHandle) = nothing | ||
|
|
There was a problem hiding this comment.
Good to know a blank line was not covered by tests!
There was a problem hiding this comment.
@test isempty(readlines("src/auditor/dynamic_linkage.jl")[27])
Co-authored-by: Mosè Giordano <[email protected]>
|
Bump :) |
Currently the auditor is only checking for the correct OS/ABI and producing a warning on FreeBSD when it doesn't match. This warning is incorrect on AArch64 though, as FreeBSD instead uses an ELF note to declare the OS/ABI. The change implemented here looks for such a note when the OS/ABI in the ELF header is NONE.
Note that there are no tests added... I'd be interested in any suggestions for how best to test this given that the functions I modified are not currently tested, at least directly (AFAICT).