Chase Phillips wrote:
I discovered that Tailor doesn't have CVS create directories in a
working installation by running 'cvs .. update -d foo/' nor delete empty
directories by running 'cvs .. update -dP foo/'. Since Tailor does this
by hand (I think Tailor should have CVS handle this), I reasoned that
the problem was when the directory was supposed to be removed, rather
than the second time it was added. I began searching for where it would
delete the directory.
I found that Tailor doesn't delete empty directories... at least in the
test case as I ran it. Tailor just schedules a removal event to occur
on the target tree. This leads to a situation where the source and
target trees get out of sync. I've attached a patch which fixes that
problem. With the patch applied, the CvsReappearedDirectory test
completes successfully for me. Also, in my tests my repository gets
past the problem that brought about my original email.
Thanks Chase, but I'm too tired now to understand why your patch is
needed, and before applying it I'd really like to understand why the
test is working for me.
I pushed two patches that implement a --debug for "tailor test" too, and
check for error in the setUp() methods.
Could you please rerun the test with --debug and send the whole output
to me?
Some comments:
1. Why the code at cvsps.py:336? It looks like it's trying to do what
maybeDeleteDirectory() tries. With my patch applied, rmtree() appears
in two cases when a directory is to be removed at a deleted event. If
it makes sense to do so, maybe the code could be reorganized to only
call rmtree() once.
I'll look closer tomorrow.
2. It wasn't clear in the verbose or debug output how a new directory
was added. For that I had to track the code. Maybe debugging output
could be added which calls out when a new directory is being added in
the CVS working directory?
Latest patch adds some noise there.
ciao, lele.
_______________________________________________
Tailor mailing list
[email protected]
http://lists.zooko.com/mailman/listinfo/tailor
