Re: [PATCH v3 1/1] packages: Support for full Guix specification

[Ricardo Wurmus]

Hi Olivier,

thanks for the new patch and my apologies for the delay!

It looks fine, but I can’t help but feel a little confused about what
variables are actual package values and what are wrappers.  Sometimes we
have a variable called “packages”, but it’s seemingly always going to be
package wrappers, so we map “package-unwrap”.

I wonder if there’s a way to hide this machinery somewhat.  On the other
hand, “package-unwrap” is a no-op for actual packages, so it doesn’t
really matter.

> +(define package-native-inputs
> +  (match-lambda
> +    ((? package? pkg)
> +     (package-native-inputs pkg))
> +    ((? inferior-package? pkg)
> +     (inferior-package-native-inputs pkg))
> +    ((? package-wrapper? pkg)
> +     (package-native-inputs (package-wrapper-package pkg)))))

I’m confused about the naming.  Is this a recursive application of
PACKAGE-NATIVE-INPUTS?  Or are these two different implementations of
PACKAGE-NATIVE-INPUTS, one from GWL and the other from Guix?

>  (define (lookup-package specification)
>    (log-event 'guix (G_ "Looking up package `~a'~%") specification)
> -  (match (lookup-inferior-packages (current-guix) specification)
> -    ((first . rest) first)
> -    (_ (raise (condition
> -               (&gwl-package-error
> -                (package-spec specification)))))))
> +  (let-values (((name version output)
> +                (package-specification->name+version+output
> +                 specification)))
> +    (let* ((inferior-package
> +            (lookup-inferior-packages (current-guix)
> +                                      name version))
> +           (package (match inferior-package
> +                      ((first . rest) first)
> +                      (_ (raise (condition
> +                                 (&gwl-package-error
> +                                  (package-spec specification))))))))
> +      (make-package-wrapper package output))))

I think this would be slightly improved by using SRFI-71 instead of
LET-VALUES.  SRFI-71 replaces LET and LET* so that you can assign
multiple values without needing to be explicit about it.

Since this procedure is now a bit more complicated I think it would be
good to have a docstring that mentions the input and output values.

>  (define default-guile
>    (mlambda ()
>      "Return the variant of Guile that was used to build the \"guix\"
>  package, which provides all library features used by the GWL.  We use
>  this Guile to run scripts."
> -    (and=> (assoc-ref (inferior-package-native-inputs (build-time-guix))
> +    (and=> (assoc-ref (package-native-inputs (build-time-guix))
>                        "guile") first)))

Is this correct?  BUILD-TIME-GUIX has been changed so that the return
value is an unwrapped wrapper — so it really is an inferior package.
Does PACKAGE-NATIVE-INPUTS refer to the implementation here or that in
Guix?

I think it would be best if we can make all this unambiguous.

-- 
Ricardo