patchRe: [PATCH] CVSps fixed; git-cvsimport works, too

[for original thread see http://www.gelato.unsw.edu.au/archives/git/0511/11812.html]

Hi Chris,

I was looking at your patch and test script to decide whether to include it in my cvsps patch collection, but it seems there are still problems:

* when run with "-x" it gives the expected result, even with "-A"

* when run without "-x" and without "-A", the patchset introducing
file "b" and branch "stable" does not appear, without even a warning

* when run with "-A" cvsps notices the inconsistency:

invalid initial_branch for file a, probably from old cache, run with -x.

I have included the current version of your patch in branch "to-check" in http://ydirson.free.fr/soft/git/cvsps.git/. I did not tak time to look at it closely yet - do you plan to work on this issue ?

Best regards,

On Sat, Jun 17, 2006 at 02:21:58PM +0200, Yann Dirson wrote:

This is expected.

With out "-A" I don't think the patch is operative.

good advice.

I wondered if you'd find that.  Evidentally, you're being thorough. :)

The intent of my patch is only to affect the branch ancestry calculation. IIRC, the patch version is incompatible with cached results from unpatched versions. Hence I would only expect it to work with '-x -A' (unless there was no pre-existing cache file).

I've had one independent report of success from someone who found and used my patch.

I haven't had to use cvsps recently, so I don't know how much time I'll have for cvsps hacking, but I'm interested in getting feedback on my current patch.

There's probably some nicer way to handle the cache incompatibility, but my current approach has been to tell people to delete their old cache file or run wih -x. Suggestions are welcome.

-chris

On Sat, Jun 17, 2006 at 09:20:23AM -0400, Chris Shoemaker wrote:

All those tests were done with your patch applied, so the cache was generated and re-read using the pached version.

If the cache format is changed, this would be a good candidate for the "v2" cache format. I already have 2 (incomplete) patches that well need to bump the cache format version - see branches multitag (allow several tags per patchset) and semicolon (allow semicolons in filenames).

Best regards,

On Sat, Jun 17, 2006 at 03:47:17PM +0200, Yann Dirson wrote:

Are you saying there was no pre-existing cache file when the patched version was first run? Did you delete it? If so, then there's a bug somewhere.

IIRC, it's not syntactically incompatible - it's just that the patched version can tell that it doesn't trust the initial branch determinations that are stored in the cache file by an unpatched version.

The complaint is basically, Hey, there's an initial branch determination already made for this file, but it's not the one I would have made, so it must have been restored from a cache file that I didn't make.

-chris

On Sat, Jun 17, 2006 at 10:34:43AM -0400, Chris Shoemaker wrote:

It was the sample script you provided to demonstrate the problem which showed the new issue: it does a "cvsps -x" first, then runs git-cvsimport, which runs "cvsps -u". That one does fails. I'm pretty sure the cache does not exist beforehand, since your script creates the repo in a temporary directory.

BTW, I'll see about adding a testsuite to cvsps, based on the framework used by git. All these scriptlets used to exhibit the various problems have to be kept somewhere.

I do think there is a bug somewhere :)

OK, thanks for the clarification.

Best regards,