Re: 02/02: services: Add Gitolite.

[Mark H Weaver <mhw@netris.org>]

Hi Christopher,

Christopher Baines <mail@cbaines.net> writes:

> Mark H Weaver <mhw@netris.org> writes:
>
>> mail@cbaines.net (Christopher Baines) writes:
>>
>>> cbaines pushed a commit to branch master
>>> in repository guix.
>>>
>>> commit 258a6d944ed891fa92fa87a16731e5dfe0bac477
>>> Author: Christopher Baines <mail@cbaines.net>
>>> Date:   Fri Jul 13 20:39:46 2018 +0100
>>>
>>>     services: Add Gitolite.
>>>
>>>     * gnu/services/version-control.scm (<gitolite-configuration>,
>>>     <gitolite-rc-file>): New record types.
>>>     (gitolite-accounts, gitolite-activation): New procedures.
>>>     (gitolite-service-type): New variables.
>>>     * gnu/tests/version-control.scm (%gitolite-test-admin-keypair, %gitolite-os,
>>>     %test-gitolite): New variables.
>>>     (run-gitolite-test): New procedure.
>>>     * doc/guix.texi (Version Control): Document the gitolite service.
>>
>> This commit has a flaw which broke evaluations on Hydra, so I reverted
>> it for now.
>>
>>> +(define-gexp-compiler (gitolite-rc-file-compiler
>>> +                       (file <gitolite-rc-file>) system target)
>>> +  (match file
>>> +    (($ <gitolite-rc-file> umask git-config-keys roles enable)
>>> +     (apply text-file* "gitolite.rc"
>>> +      `("%RC = (\n"
>>> +        "    UMASK => " ,(format #f "~4,'0o" umask) ",\n"
>>
>> The problem is here, in the call to 'format'.  Guile has two variants of
>> the 'format' procedure: a very simple one which is always available
>> (also known as 'simple-format'), and a much more complex and featureful
>> variant in (ice-9 format).  In the code above, you are assuming that
>> (ice-9 format) has been loaded, but you have not arranged for that to
>> happen.  During the Hydra evaluation, this led to the following error:
>>
>> --8<---------------cut here---------------start------------->8---
>>            0 (simple-format #f "~4,'0o" 63)
>>
>> ERROR: In procedure simple-format:
>> In procedure simple-format: FORMAT: Unsupported format option ~4 - use (ice-9 format) instead
>> --8<---------------cut here---------------end--------------->8---
>
> Thanks for looking at this Mark, do you know where I can find out what
> Hydra is doing when it encounters this error? I tried looking around the
> web interface, but the latest evaluation I could find was from 2017
> apparently.

Hydra's web interface is the wrong place to look for information on
this problem.

I guess the error happens when Hydra tries to create a derivation for
the system test '%test-gitolite'.

The relevant code that Hydra runs to generate an evaluation is in
build-aux/hydra/gnu-system.scm.  Specifically, 'system-test-jobs', which
calls (all-system-tests) from gnu/tests.scm, which looks for public
variables bound to 'system-test' records.  That includes
'%test-gitolite'.

>> I guess that the fix will involve 'with-imported-modules'.  Could you
>> take a look?
>
> Sure. The confusing thing to me here is that this code works when using
> the service, including the system test (at least when I run it
> locally). Also, as far as I understand, even though the code is within a
> gexp-compiler, it doesn't even involve the store, as the call to format
> happens before the g-expression is generated.
>
> It sounds to me like adding #:use-modules (ice-9 format) to (gnu
> services version-control) would fix this, but I'll wait until I can
> reproduce the failure before re-adding the service.

I'm not yet sufficiently familiar with gexp compilation to know where
the call to 'format' is ultimately being executed, and I don't have time
right now to investigate further, so I hoping that Ludovic can chime in
here with a definitive answer on how to fix this properly.

       Mark