bug#64762: Guix sometimes doesn't support most packages for i686-linux and armhf-linux

[Christopher Baines <mail@cbaines.net>]

[-- Attachment #1: Type: text/plain, Size: 2078 bytes --]


Christopher Baines <mail@cbaines.net> writes:

> To confirm that this is an issue with the supported systems as reported
> by Guix, I had the data service print out the transitive supported
> systems for the guix package:
>
>   debug: Starting getting derivations for (i686-linux . #f)
>   looking at guix package (supported systems: (), system supported: #f, target supported: #t
>   debug: Finished getting derivations for (i686-linux . #f), took 12  seconds
>
>   debug: Starting getting derivations for (armhf-linux . #f)
>   looking at guix package (supported systems: (), system supported: #f, target supported: #t
>   debug: Finished getting derivations for (armhf-linux . #f), took 41 seconds

I realised that this could be non-deterministic because the order in
which the data service processes derivations is non-deterministic, from
the order of the systems returned by (guix platforms).

That led me to a reproducer:

  (use-modules (gnu packages) (guix packages) (gnu packages package-management))

  (fold-packages (lambda (pkg result)
                   (package-transitive-supported-systems pkg "i586-gnu")
                   #f)
                 #f)

  (peek "guix package supported systems" (package-transitive-supported-systems guix "i686-linux"))

If you look at the %final-inputs packages, there's an issue with libc:

  (use-modules (ice-9 match)
               (gnu packages) (guix packages) (gnu packages package-management)
               (gnu packages commencement))

  (fold-packages (lambda (pkg result)
                   (package-transitive-supported-systems pkg "i586-gnu")
                   #f)
                 #f)

  (for-each
   (match-lambda
     ((name pkg rest ...)
      (peek name
            (package-transitive-supported-systems
             pkg
             "i686-linux"))))
   (%final-inputs "i686-linux"))

I think this could be because %final-inputs is cached based on system,
but doesn't seem to use system. Setting %current-system to system seems
to help, so I've sent a patch for that [1].

1: https://issues.guix.gnu.org/64763

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 987 bytes --]