gdb 8.1 fails debuginfo tests (output changed).

[eddyb]

AFAICT, it appears to print module::Enum::Variant {...} instead of module::Variant {...} now.

cc @tromey @michaelwoerister

[eddyb]

@Pazzaz Hmm, do you specifically want more information? The build server I use is NixOS and it was updated recently, which appears to have picked up gdb 8.1, which has slightly different output.

[nnethercote]

I have the same or a similar problem:

failures:
    [debuginfo-gdb] debuginfo/borrowed-c-style-enum.rs
    [debuginfo-gdb] debuginfo/borrowed-enum.rs
    [debuginfo-gdb] debuginfo/c-style-enum-in-composite.rs
    [debuginfo-gdb] debuginfo/gdb-pretty-struct-and-enums-pre-gdb-7-7.rs
    [debuginfo-gdb] debuginfo/generic-enum-with-different-disr-sizes.rs
    [debuginfo-gdb] debuginfo/generic-struct-style-enum.rs
    [debuginfo-gdb] debuginfo/generic-tuple-style-enum.rs
    [debuginfo-gdb] debuginfo/struct-style-enum.rs
    [debuginfo-gdb] debuginfo/tuple-style-enum.rs
    [debuginfo-gdb] debuginfo/unique-enum.rs

I'm on Ubuntu 18.04, using gdb 8.1.

[Pazzaz]

@eddyb It was that I encountered the bug on my own machine, running Ubuntu 18.04 and gdb 8.1. (same as nnethercote)

[joshtriplett]

Encountered this on current Debian unstable with gdb 8.1.

[tromey]

I think the new output is more correct and so the thing to do is update the tests.

I looked through the gdb 8.1 git log, searching for a change that would affect this. All I found, though, was this commit: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=commit;h=ec8df23454873916c6b6b918967af631b268acd8

... which claims that the change was in rustc, not gdb.

[eddyb]

cc @alexcrichton Can we reasonably update the tests?

[alexcrichton]

So long as we're still running tests on CI seems fine by me (I forget if we have min-gdb-version yet)

[michaelwoerister]

I would be awesome to mostly base our tests on gimli or llvm-dwarfdump...

[tromey]

So long as we're still running tests on CI seems fine by me (I forget if we have min-gdb-version yet)

Yes, we do. Somewhere also there was a PR that built a local gdb in travis. Though now that 8.2 is officially out perhaps it isn't too hard to upgrade some config somewhere.

[tromey]

It was #53497, but I'm also talking with Matthias Klose about getting a newer gdb into the Ubuntu toolchain PPA, which would be a lot simpler I think.

[alexcrichton]

@tromey ok great! All of our builds are pretty self-contained and independently upgradeable, so it should be fine to move a builder to Ubuntu 18.04 from where it's at now (that's what I'm running locally now and see these failures). We could then tag everything with a minimum version and they'll only get run on this one bot

[tromey]

FWIW testing locally I can only get failures with gdb 8.2 and later, not 7.11 or 8.1. Perhaps Ubuntu's "8.1" is actually based on something later. This may complicate the version check.

[tromey]

I think the new output is more correct and so the thing to do is update the tests.

Well, that seems wrong to me now.

The test says:

// gdbr-check:$1 = borrowed_c_style_enum::ABC::TheA

And when I update the Dockerfile and run the tests I get:

$1 = borrowed_c_style_enum::TheA

I think what is happening here is that LLVM is not emitting DW_AT_enum_class on these enums. I'm not sure how to dig this out of the docker build but I saw this while working on #54004, which is why that patch has the IsFixed enum stuff.

This applies to the C-style enum fails, but the others seem similar. More later after I rebuild & see if I can provoke it in a more debuggable way.

[tromey]

I can't reproduce with a clean build of git master and gdb 8.1, but I can via an update Dockerfile. I wonder if there's some LLVM difference between these two builds somehow.