Ed Santiago [off-list ref] writes:
Question mark character appears to be valid in a CVS tag,
but not a git one. Remove it. Leave open the possibility that there may
be more such characters; and comment (FIXME) that we may want to replace
those instead of removing them.
I agree that people may want to optionally replace them to avoid mapping
two originally different tags into the same one.
Does your regexp chain check other invalid refname sequences, such as ".name",
"na..me", etc.?
Also: if git tag command fails, do not include $! in our
error message: it is not useful after system(), and will
only serve as a red herring.
---
Sign-off?
On Wed, Apr 14, 2010 at 07:29:03AM -0700, Junio C Hamano wrote:
I agree that people may want to optionally replace them to avoid mapping
two originally different tags into the same one.
I considered that but decided that it was beyond the scope of
what I wanted to tackle. My intent was to suggest an incremental
fix, not a revolutionary one. I also wished to remove a misleading
diagnostic[1] and provide a way for others to do the same as needed.
[1] The error I saw was "Cannot create tag mumblefoo{?dist}: Illegal seek",
where "Illegal seek" is due to an uninitialized/invalid $!
A proper solution would, I believe, be uncomfortably complex.
Some of the approaches I considered are:
* command-line option '--tag-replace <from-char>=<to>', eg '?=-'
* replacing a (hardcoded) list of chars with %xx hex codes
* option a la -A, with a file listing <oldtag>=<newtag>
* like above, but the file lists perl expressions such as s/\?/./g
I used git-cvsimport because I've used tailor[2] before and
dreaded the thought of running it again. I love how easy
git-cvsimport was to use! I see it as a great tool for
quick jobs, for proof-of-concept situations: it has a low
cost of entry, works well with a remote CVS repo, and was
easy to debug. But if it were to become as configurable -- and
as complex -- as tailor, I think it might lose much of its value.
[2] http://wiki.darcs.net/RelatedSoftware/Tailor
Does your regexp chain check other invalid refname sequences, such as ".name",
"na..me", etc.?
Eek, no. Again, this was a purely incremental fix, an addition
to an existing (simple) check. Duplicating the functionality
of git-check-ref-format would be a terrible idea. But if there's
already a perlified library for this, or a portable hook into
git itself (I haven't looked), it might be wise to have
git-cvsimport use that to check/convert tags. At the very
least it would give a much better error to the end user.
Sign-off?
My apologies. I'm new to this.
How would you like me to proceed? I'm reluctant to code a
full general solution to the tag-validity problem, but am
happy to rework my changes into something better.
Thanks for your feedback,
Ed
--
Ed Santiago Software Engineer santiago@redhat.com