Re: [PATCH] system: Remove spec->file-system.

[ludo@gnu.org (Ludovic Courtès)]

David Craven <david@craven.ch> skribis:

> * gnu/system/file-systems.scm (spec->file-system): Remove variable.
> * gnu/system/linux-container.scm (container-script): Refactor.

[...]

> --- a/gnu/system/linux-container.scm
> +++ b/gnu/system/linux-container.scm
> @@ -81,8 +81,7 @@ MAPPINGS is a list of <file-system> objects that specify the files/directories
>  that will be shared with the host system."
>    (let* ((os           (containerized-operating-system os mappings))
>           (file-systems (filter file-system-needed-for-boot?
> -                               (operating-system-file-systems os)))
> -         (specs        (map file-system->spec file-systems)))
> +                               (operating-system-file-systems os))))
>  
>      (mlet* %store-monad ((os-drv (operating-system-derivation
>                                    os
> @@ -94,10 +93,9 @@ that will be shared with the host system."
>                                    (gnu build linux-container)))
>            #~(begin
>                (use-modules (gnu build linux-container)
> -                           (gnu system file-systems) ;spec->file-system
>                             (guix build utils))
>  
> -              (call-with-container (map spec->file-system '#$specs)
> +              (call-with-container #$file-systems

AFAICS that doesn’t work because <file-system> and records in general
are not automatically marshalled/unmarshalled when staging code.  Thus,
the above would leave to an invalid on-disk s-expression like this:

  (begin
    (use-modules …)
    (call-with-container (#<<file-system> …> #<<file-system> …>)
      …))

Manually calling ‘file-system->spec’ on the host side, and then
‘spec->file-system’ on the build side is a way to explicitly
marshall/unmarshall the record (I mentioned it as a current limitation
of gexps in my Scheme Workshop talk).

Makes sense?

Thanks,
Ludo’.