Re: building vanilla gnu.org-i18n

[Yavor Doganov]

В чт, 2007-12-27 в 18:03 +0200, Kaloian Doganov написа:
> Yavor Doganov <address@hidden> writes:
> 
>     We could check somehow if the diff is only POT-Creation-Date
> 
> This is not a major issue,

Maybe; but I still have my doubts how the thing will work with CVS.

> I think there is a simpler approach.  

I definitely disagree.  Nothing could be simpler than a make recipe that
adds *exactly* the same file it operates with.  This is not just a whim
-- the whole system would be much more powerful if we extend it taking
this road. 

> We do not need any hostname checks

This is just to prevent a translator (or a webmaster) of actually
modifying the repository (OK, provided that we don't add directories,
there's no way to modify the repository except "cvs commit" where
everyone with write access should check what she is checking in...)

The conditional variable that I'm talking about should be enough for
this purpose.

>     2. Rebuild translations (`make -k www/prep/i18n').

JFTR: I think it is better not to ignore errors.  An error in a rule
early in the process will break all translations down the chain (well,
at least those that depend on the file which was not built/updated
correctly).

>     3. Examine working copy and collect new filenames that should be put
>        under control (possibly grepping through `cvs status').

On a 300 MB repository with lots of files that is a horrible thing.  We
know what should be under CVS control at build time, so we'd better do
it then.

>     5. Invoke `cvs commit' on the working copy.

The script will basically update, invoke make (trapping all the output)
and commit only upon successful exit of the make process.  I've been
thinking in the past few days how to safeguard against possible bugs
(i.e. command failures), so I'll work on this.

> Does this makes sense to you?