Hi Nick,

Sorry, I somehow missed this message.

On Thu, 2024-04-18 at 17:32 +0100, Nick Alcock wrote:
> I'm getting back to libctf development again (some fairly big things
> coming, I hope), so I've just tried the trybots for the first time.
> 
> First impressions: really useful to be able to test on piles of arches
> without needing to set up dead slow QEMU emulations of them myself;
> might be even more useful if FAILs/ERRORs were explicitly filtered into
> their own category rather than carefully filtering out warnings (which
> are usually useless junk from dejagnu), while you have to explicitly
> download the stdio output to figure out which testsuite any failures
> actually happened in...

Yeah, the logs and mails sent by the buildbot could use some work.
But the results are also stored in bunsen. Try looking there:
https://builder.sourceware.org/testruns/
https://builder.sourceware.org/testruns/?has_keyvalue_op=glob&has_keyvalue_k=testrun.git_describe&has_keyvalue_v=buildbot/binutils-%2A

> ... but the check-libctf tests are not run at all, which means the only
> testing I get of libctf is of its interoperation with ld, which is only
> about half the codebase and about as much of the API surface (counting
> uses internal to libctf itself).
> 
> So... could make check-libctf be added to the stuff routinely run when
> binutils try-builds are done? I'd write a patch but I'm not honestly
> sure where the repo is or anything :) it's pretty fast (much faster than
> make check-ld, about as fast as the gas tests) and doesn't have spurious
> failures anywhere that I know of. (And if there are failures, I'll fix
> them.)

The builder sources can be found here:
https://sourceware.org/cgit/builder
And there is a mailinglist (CCed):
https://sourceware.org/mailman/listinfo/buildbot

The attached patch (committed) makes it so that make check-libctf is
always run now. Lets see how that turns out.

Cheers,

Mark