Lele Gaifax wrote: > I recall that I had problems with cvsps, when I started with cvsync: > from time to time (probably due to new branch creations, or something > wierd) the revid was not "constant" (in other words, what cvsps tagged > as "123" a week ago may be different from the changeset "123" of today). > Has anyone ever seen this, or is this fixed in current cvps?
I'm not sure it is the same, but cvs 1.12.x (x maybe >= 10) introduces buggy old revisions, which confuse cvsps and make it produce matching old revisions (ie. -u and -x do not produce the same result). See http://bugs.debian.org/369870 for details. Also, cvsps patchsets are ordered by date, so it is (at least in theory) possible that a later commit with an old date gets into the repository (in much the same way as the buggy ones described above). But the cvs backend would have the same problems than cvsps here. >> It would have been smart to look at how cvsps dealt with tags before >> writing the tag handling code for the cvs source, but I didn't, so I'm >> not familiar with the funky/invalid classification you refer to. I'll >> take a look. > > I'm afraid the cvsps backend does not handle tags. I surely should spend > some time on that too, because I never got a precise idea of how tags > are currently handled :-) Those interested on working on the cvsps backend could be interested as well in a couple of nasty bugs I discovered when hunting for the one mentionned above, at http://bugs.debian.org/cvsps > Maybe the behaviour should be abstracted a > little bit so that other backends could play the same game. That would be great! A generic API to deal with branches would be another great addition, BTW:) -- Yann Dirson <[EMAIL PROTECTED]> <[EMAIL PROTECTED]> _______________________________________________ Tailor mailing list [email protected] http://lists.zooko.com/mailman/listinfo/tailor
