Re: [PATCH 04/13] utils: Use '@' for separating package names and version numbers.

[Ludovic Courtès]

Mathieu Lirzin <address@hidden> skribis:

> address@hidden (Ludovic Courtès) writes:
>
>> Mathieu Lirzin <address@hidden> skribis:
>>
>>> From 7fef43249b704db3c4d511b2f1b62428740cfa73 Mon Sep 17 00:00:00 2001
>>> From: Mathieu Lirzin <address@hidden>
>>> Date: Sun, 24 Jan 2016 01:54:44 +0100
>>> Subject: [PATCH] utils: Use '@' for separating package names and version
>>>  numbers.
>>>
>> [...]
>>>  (define (package-name->name+version name)
>>> -  "Given NAME, a package name like \"foo-0.9.1b\", return two values:
>>> -\"foo\" and \"0.9.1b\".  When the version part is unavailable, NAME and
>>> -#f are returned.  The first hyphen followed by a digit is considered to
>>> -introduce the version part."
>>> +  "Given NAME, a package name like \"address@hidden", return two values: 
>>> \"foo\"
>>> +and \"0.9.1b\".  When the version part is unavailable, NAME and #f are
>>> +returned.  Both parts must not contain any '@'."
>>
>> I think this one should remain unchanged, first because it triggers a
>> full rebuild ;-), and second because it has uses in
>> {emacs,gnu}-build-system that expect exactly these semantics.
>>
>> So I think we have to simply provide a different version of that in
>> (guix utils) or so.
>>
>> Also, I think that at least for some time, the new
>> ‘package-name->name+version’ (maybe we could call it
>> ‘package-specification->name+version’ for consistency) should fall back
>> to the old method when:
>>
>>   1. The spec has no @ sign, and
>>
>>   2. The specified package name was not found.
>>
>>
>> It could print a warning when the old method has been used *and* a
>> matching package was found, explaining that this is deprecated syntax.
>>
>> WDYT?
>
> I have finally took some time to look into this, and actually
> implementing ‘package-specification->name+version’ in (guix utils) with
> the suggested fall back feature seems not desirable since checking for
> package existence would require using the procedure
> ‘find-packages-by-name’ from (gnu packages) which will then create a
> circular dependency.  Did I miss something?

No, good point.

What about this:

  1. Put the new ‘package-name->name+version’ in (guix utils) and keep
     it dumb (i.e., it insists on having an address@hidden, as is the case with 
the
     patch you posted.)

  2. Add the fallback case in ‘specification->package’ and
     ‘specification->package+output’ in (gnu packages).

How does that sound?

Thanks!

Ludo’.