Re: Testsuite broken by svntest changes in svn trunk

Michael Haggerty <[email protected]> Thu, 17 Jun 2010 09:05:43 +0200
Newsgroups gmane.comp.version-control.subversion.cvs2svn.devel
Message-ID <[email protected]>
Max Bowsher wrote:
> Since the inception of cvs2svn, or near enough, we've depended on 
> 'svntest' via svn:externals. Anyone stopping and thinking for a bit can 
> see that depending on an unspecified version of an external piece of 
> code is rather silly.
> 
> And indeed, we're currently broken, because svntest has changed 
> underneath us.

This has happened before, not too long ago, when the output of "svn
proplist" changed format.  I'm partly to blame because I don't even
notice the breakage; I work with git-svn and rarely update the svn:external.

> I think we should do one of two things:
> 
> EITHER: Use a pinned revision in the svn:externals.
> 
> OR: Embed svntest into our tree.
> 
> 
> I've never liked the idea of svn:externals much, so I favour the second, 
> but I could be persuaded to the first.

Either solution is OK with me.  If you choose the latter, then it would
be great if you would set up an update.sh script like the one in
cvs2svn_rcsparse.  And I suppose we should decide on a policy--are we
forking svntest (basically sticking to the current version and adding
our own modifications) or are we going to try to track upstream svntest,
just at our own pace?  I think the latter is more realistic, given our
limited resources.

I've also been wondering how much we really use from svntest, and
whether the Subversion-specific stuff there is rather holding us back.
Maybe with time we should migrate more towards python's unittest module,
using the stuff in svntest only for parsing SVN output etc.

Michael

------------------------------------------------------
http://cvs2svn.tigris.org/ds/viewMessage.do?dsForumId=1667&dsMessageId=2622962

To unsubscribe from this discussion, e-mail: [[email protected]].