Re: CI scripts

[Simon Marchi <simark@simark.ca>]

On 1/18/23 07:09, Jan Vrany via Gdb wrote:
> Hi Simon,
> 
> On Thu, 2023-01-12 at 08:55 -0500, Simon Marchi wrote:
>>
>> On 1/12/23 07:56, Jan Vrany via Gdb wrote:
>>> Hi Simon,
>>>
>>> about a month ago when you helped me with use-after-free fix
>>> you wrote:
>>>
>>>> I ran the patch through my CI job, looks good.
>>>
>>> I'm wondering whether the CI scripts for your job are available?
>>> I'd like to setup a similar job on my CI to avoid similar problems
>>> in future.
>>>
>>> I'm especially interested in "configure ..." and "make check ..."
>>> commands.
>>>
>>> Thanks!
>>>
>>> Jan
>>>
>>
>> Hi Jan,
>>
>> Yes, it's on Jenkins, the job description (it uses Jenkins Job Builder)
>> is here:
>>
>> https://github.com/simark/lttng-ci/blob/master/jobs/binutils-gdb.yaml
>>
>> The build script is runs is here (which is where the commands you are
>> intereted in are):
>>
>> https://github.com/simark/lttng-ci/blob/master/scripts/binutils-gdb/build.sh
>>
> 
> Perfect, thanks a lot! I managed to setup a job on my CI. I just call your
> script to build GDB and run and process test.
> 
> There's one little thing - the script uses `--with-guile=guile-2.2`. My build
> nodes have no guile-2.2, only `guile` which is an alias for `guile-3.0` so I
> have to patch the script to use just `--with-guile`.

Oh, I'll update to just --with-guile and let the system pick the default
guile.

I just noticed I provided you links to my personal fork, which is not up
to date.  The files on the real repo are more up to date:

https://github.com/lttng/lttng-ci/blob/master/scripts/binutils-gdb/build.sh

Notably, this version generates a gdb.fail.sum, which contains just the
tests that caused the build to be considered a failure.  So it's only
FAILs/UNRESOLVEDs/XPASSes/DUPLICATEs, after filtering out the known
pre-existing ones.

>> The job itself is hosted here:
>>
>> https://ci.lttng.org/view/GDB/job/binutils-gdb_master_linuxbuild/
> 
> Thanks! Interestingly, my build of 1b1be68b9 (using your script) shows
> only 6 failures wheres "your" [1] build shows 72750 failures...

I don't know, if you use -D_GLIBCXX_DEBUG (like the script does), you
should see loads of failures in the C++ tests.  I reported the problem
here, and it should go back to normal once Tom provides a fix.

https://inbox.sourceware.org/gdb-patches/871qnt2bob.fsf@tromey.com/T/#mceb9a4e339cd209f7b6d8d361ac5b4523d71635c

I have some local modifications to the result parsing script, which I
need to clean up and push, so maybe there's a bit of difference in how
the results are analyzed.

Simon