Versioning svn checkouts [Was: Re: alpha/beta software in Fedora 8?]

Toshio Kuratomi a.badger at gmail.com
Fri Nov 30 20:40:16 UTC 2007


Martin Marques wrote:
> Bill Nottingham escribió:
>> Petr Machata (pmachata at redhat.com) said:
>>> Steve Grubb wrote:
>>>> kdepim-enterprise-svn20070926.tar.bz2
>>> As a side note, I always wondered why to use date in the release tag of
>>> package, whose sources come from non-cvs versioning system.  For svn, in
>>> my opinion, it would make more sense to use the tree revision number;
>>> for git, similarly, sha1 id of the tree.
>>
>> Well, git<date> sorts sanely. git<sha1> does not. Comments in the spec
>> (or similar) might help with this.
> 
> git<date> from where? Remember that git is distributed and unless you 
> have a centralized copy git<date> can have more then one copy.
> 
This actually is a reasonable point but you might not like the answer 
:-).  Git is distributed but that doesn't mean we should be pulling from 
any git tree.  It doesn't even mean that we should be pulling from 
several different git trees.  In almost all cases there should be one 
upstream git tree that upstream either designates as canonical or one 
that we decide it would be sane to pull from for a reasonable amount of 
time (with reasonable amount purposely left vague so maintainers have 
leeway here.)

Once again, though, the information in the release field after the 
mandatory integer at the beginning is for the end user to indicate the 
age of the upstream pull.  Git hashes and other ids that aren't either 
well known to end users or have some bearing on age (incrementing 
integers show some idea of age, a checksum does not) don't fit this 
criteria.  They belong in the spec file as part of a comment on how to 
checkout the sources but are of limited use in the release tag itself.

-Toshio




More information about the devel mailing list