Cyril Roelandt <tipec...@gmail.com> skribis:
> * gnu/packages/python.scm (python-2): patch Lib/subprocess.py to use
> /nix/.../sh when /bin/sh is not available (most of the time, in Guix chroot)
> ---
> gnu/packages/python.scm | 19 ++++++++++++++++++-
> 1 file changed, 18 insertions(+), 1 deletion(-)
>
> diff --git a/gnu/packages/python.scm b/gnu/packages/python.scm
> index 8f92dc2..a663b5e 100644
> --- a/gnu/packages/python.scm
> +++ b/gnu/packages/python.scm
> @@ -138,7 +138,24 @@
> (with-directory-excursion out
> (for-each (cut augment-rpath <> lib)
> (find-files "bin" ".*")))))
> - %standard-phases)))
> + (alist-replace
> + 'configure
> + (lambda* (#:key outputs #:allow-other-keys #:rest args)
> + (let ((configure (assoc-ref %standard-phases 'configure)))
> + ;; Some libraries try to call subprocess.Popen(), which uses
> + ;; /bin/sh, in their tests. Since /bin/sh is not available in
> the
> + ;; chroot, the tests fail. Instead of disabling them, add
> + ;; '/nix/.../sh' as an alternative shell. The default shell is
> + ;; still /bin/sh, so the subprocess module works exactly as
> + ;; expected. The extra 'if' condition will slow down things, but
> + ;; not enough to be noticed.
> + (substitute* "Lib/subprocess.py"
> + (("args = \\[\"/bin/sh")
> + (string-append
> + "args = [\"/bin/sh\" if os.path.exists(\"/bin/sh\") else "
> + "\"" (which "sh"))))
> + (apply configure args)))
I’m in favor of just (which "sh").
As discussed on IRC, all one can expect from /bin/sh is to point to a
Bourne-compatible shell, and that’s what we’d guarantee by explicitly
using the build-time (which "sh").
Fine with you?
Thanks,
Ludo’.
