Hi,

On Fri, 08 Dec 2023 at 16:50, Simon Tournier <zimon.toutoune@gmail.com> wrote:

>>>  @example
>>> -guix shell --pure
>>> +guix shell -CPW
>>>  @end example
>>
>> I would not recommend that or adding also the option -N.  Else, the
>> development experience can be annoying.
>
> As said, I disagree with this change pushed by
> 01361d46b8e0481ad56665d7a06c276b08f59c6d.
>
> Could you revert?  Because as explained by…
>
>> For instance, “make check” will probably fail because some substitutes
>> are missing (see #67532 [1], e.g., tests/derivations.scm).
>
> …the option ’-CPW’ is not a solution.  Other said, it would require in
> addition ’-N’.

Ludo’s answer from bug#67532 [1]:

--8<---------------cut here---------------start------------->8---
I had come to forget about that because for the ‘guix’ package
(obviously built in a network-less environment), we “pre-seed” the store
with the 4 or 5 files needed to run the test suite.

So I propose:
[ text/x-patch ]
diff --git a/doc/contributing.texi b/doc/contributing.texi
index 0833b5d32c..0072e5d42d 100644
--- a/doc/contributing.texi
+++ b/doc/contributing.texi
@@ -76,13 +76,13 @@ Building from Git
 hack on Guix:

 @example
-guix shell -D guix -CPW
+guix shell -D guix -CPWN
 @end example

 or even, from within a Git worktree for Guix:

 @example
-guix shell -CPW
+guix shell -CPWN
 @end example

 If @option{-C} (short for @option{--container}) is not supported on your
[ text/plain ]

(And not going back to ‘--pure’ because as we know it depends too much
of the goodwill of the user’s shell.)
--8<---------------cut here---------------end--------------->8---

Well, IMHO it falls into…

> To me, this recommendation of container is like a bazooka for chasing
> the real bug: about why --pure does not clean all as expected.  Or why
> “guile” is found elsewhere?

…a bazooka for killing a fly.  For instance, in the two message of the
initial bug report, I do not see the output of the current
recommendation: run “guix shell --check”.

Yes, “guix shell” depends on the goodwill of the user’s shell but
recommending first --pure is an opportunity for improving --check and/or
--pure itself, IMHO.

Considering this reported instance of issue, it is not clear for me that
--check would not have reported the problem.  Or why --pure does not
correctly clean all the environment variables?  Or why ./configure set
something system wide?

For me, these questions are the bug.  And -CPWN is just a temporary
workaround avoiding the bug and not a fix.

Somehow, if the use of “-CPWN” instead of “--pure“ is not an idea that
user has when hitting a failure with “guix shell --pure”, yeah maybe it
could be helpful to recommend it in addition.  And not the contrary. :-)
Something like: