Federico Beffa <be...@ieee.org> skribis:
> For python2-matplotlib I've noticed a behavior that I do not
> understand: matplotlib depends on numpydoc which I pushed yesterday
> and appears not to be ready yet on hydra. If I build
> python2-matplotlib with
>
> ./pre-inst-env guix build python2-matplotlib
>
> guix tells me that it will build among other packages
>
> /gnu/store/gjfrir5iykxwxicp9dxjv3adk5vpifb4-python2-numpydoc-0.5.drv
>
> However, the build process for this package fails, because a test
> fails. Since I prepared the numpydoc package I know about this test
> and know that the package includes the parameter #:tests? #f.
>
> In fact, if I explicitly build python2-numpydoc with
>
> ./pre-inst-env guix build python2-numpydoc
>
> it finished successfully (skipping the tests).
>
> The two ways of causing numpydoc to be built (direct and indirect)
> generate two independent
>
> /gnu/store/...-python2-numpydoc-0.5-guile-builder
>
> The one generated by an explicit build command includes the #:tests?
> #f flag as expected. However, the indirect one includes a #:tests? #t
> flag! I'm confused. Am I missing something?
What happens is that ‘package-with-python2’ automatically generates
python2 variants of the dependencies of the package you give it. So
(package-with-python2 python-matplotlib) generates a package with an
input that is exactly (package-with-python2 python-numpydoc), hence with
#:tests? #t.
The fix is to explicitly use the right python2-numpydoc, along the lines
of:
(define-public python2-matplotlib
(let ((matplotlib (package-with-python2 python-matplotlib)))
(package (inherit matplotlib)
;; Make sure we use exactly PYTHON2-NUMPYDOC, which is
;; customized for Python 2.
(inputs `(("numpydoc" ,python2-numpydoc)
,@(alist-delete "numpydoc" (package-inputs matplotlib)))))))
> + ;; FIX-ME: Add backends when available.
Please write “FIXME” for easier grepping and highlighting.
Otherwise the patch looks good to me, thanks for working on it!
Ludo’.
