Re: Updating mumi on berlin

[Ludovic Courtès]

Hi,

Thiago Jung Bauermann <bauermann@kolabnow.com> skribis:

> Ludovic Courtès <ludo@gnu.org> writes:
>
>> Hi,
>>
>> Maxim Cournoyer <maxim.cournoyer@gmail.com> skribis:
>>
>>> Yes, that would be the correct and general solution (I hope you get to
>>> fix it, else I may look into it in some time).  Workarounds currently in
>>> use can be seen in our mutter package definition (it's a bit convoluted,
>>> having to fork a process in which we set the child reaping property and
>>> invoke the test suite via tini, acting as a fake PID 1).
>>
>> The Guile process running the build script (with the build phases) runs
>> as PID 1.  So how about doing one of these:
>>
>>   1. Add a final phase that does (waitpid WAIT_ANY).
>>
>>   2. Add a SIGCHLD handler that calls ‘waitpid’.
>>
>> I suspect the former is enough.
>
> I don't think it is, unless I'm misunderstanding something. The
> testsuite expects the daemon to die within a few seconds of sending it
> the quit command (which it does several times over the course of a
> number of the tests), so having it reaped at the end of the build
> process won't help.

I see, makes sense.

> I tried the latter a couple of weeks ago and it didn't work either,
> though it's highly likely that I did something wrong in my Scheme code.
> I also didn't use the 1-second timer hack so perhaps that was it.
>
> I just tested this (adapted from the mutter package) and it worked:
>
> @@ -4137,6 +4142,26 @@ (define-public public-inbox
>               ;; the builder.
>               (substitute* "t/httpd-unix.t"
>                 (("^SKIP: \\{") "SKIP: { skip('Guix');"))))
> +         (replace 'check
> +           (lambda* (#:key target
> +                     (tests? (not target)) (test-flags '())
> +                     #:allow-other-keys)
> +             (if tests?
> +                 (match (primitive-fork)
> +                   (0                     ;child process
> +                    (set-child-subreaper!)
> +                    ;; Use tini so that signals are properly handled and
> +                    ;; doubly-forked processes get reaped; otherwise,
> +                    ;; lei-daemon is kept as a zombie and the testsuite
> +                    ;; fails thinking that it didn't quit as it should.
> +                    (apply execlp "tini" "--"
> +                           "make" "check" test-flags))

How does a basic SIGCHLD handler fare?

  (sigaction SIGCHLD
    (lambda _ (false-if-exception (waitpid WAIT_ANY WNOHANG))))

>> Eventually, we should do one of these in gnu-build-system.scm, as in:
>>
>>   https://issues.guix.gnu.org/30948#2
>>
>> WDYT?
>
> I agree, possibly with the 1-second timer hack as well.

Heh, I hate that one, we need to find a better solution.

Thanks,
Ludo’.