Commit ed6c7b751dd4 causes test failures in debuginfo suite on Debian Stable

[kinnison]

Hi,

I have been trying to get the rust compiler tests running on my Debian/stable (Stretch) computer and was hitting continuous failure running the debuginfo suite, specifically the pretty-std-collections test. I kept hitting gdb complaining with Python Exception <class 'gdb.error'> No type named alloc::collections::btree::node::InternalNode<i32, ()>. and similar.

Looking at the gdb pretty printer python integration, I decided to try reverting the commit mentioned in the title and now the tests pass just fine.

I'm sadly not well enough versed in gdb/python bindings to begin to debug this, but would be pleased to be remote-hands to help someone else work it out.

[nikic]

The corresponding PR is #56144, cc @tromey.

[tromey]

Reverting the commit isn't really an option as the previous code was incorrect. Rewriting it in a different way is hard because BTreeSet and BTreeMap hide the real types of nodes.

What version of gdb are you using? gdb --version will say. I can build that locally and try it out. Perhaps some workaround can be found.

Failing a workaround, I think the only recourse will be to disable the test, and probably that pretty-printer, on whatever gdb releases show the problem.

[kinnison]

I guessed that a revert wasn't the right answer, sure :-) It just got me to a point of being able to run the rest of the test suite so I could play with another issue.

GDB is version GNU gdb (Debian 7.12-6) 7.12.0.20161007-git

It is the version packaged in Debian's Stretch (Stable) distribution.

[tromey]

I guessed that a revert wasn't the right answer, sure :-)

Sorry about that, I didn't mean to imply anything.

I've built 7.12 and I can reproduce the problem.

[kinnison]

I've built 7.12 and I can reproduce the problem.

That's excellent news. Let me know if you put a patch together and want a confirmation on my system.

[tromey]

I think this was this gdb bug: https://sourceware.org/bugzilla/show_bug.cgi?id=21763. I fixed it in 8.1. There isn't a workaround so I'll be disabling the test.

[kinnison]

Fair enough :-) Thanks for looking into this.