mercurial at 11238: 10 new changesets (3 stable)

[Cedric Duval]

Henrik Stuart wrote:
> On 01-06-2010 20:55, Cedric Duval wrote:
> >> http://selenic.com/repo/hg//rev/1e701ffd9df4
> >> changeset:   11229:1e701ffd9df4
> >> user:        Henrik Stuart <hg at hstuart.dk>
> >> date:        Sat May 29 20:32:39 2010 +0200
> >> summary:     mq: support multiple patch queues using qqueue

> > This is a very nice feature to have, thanks!

> > I am however a bit uncomfortable with the way those new queues are
> > stored, making it possible to have any arbitrary name inside the
> > .hg/ directory.

> > Just a few tested examples below:

> >   + hg init r
> >   + cd r
> >   + hg qqueue --create store
> >   + hg qnew journal
> >   abandon : abandoned transaction found - run hg recover !

> >   + hg init r
> >   + cd r
> >   + hg qqueue --create branch
> >   + hg qnew p
> >   + hg branch b
> >   abandon : Is a directory : /tmp/qq/r/.hg/branch

> >   + hg init r
> >   + cd r
> >   + hg qqueue --create dirstate
> >   + hg qnew p
> >   abandon : Is a directory

> > Those examples are obviously contrived (especially the last one),
> > and one could find them disingenuous name choices from the user,
> > but when anything is allowed (and given that the .hg layout may
> > evolve, adding new files or directories - which are not always
> > created right away, but sometimes after some specific actions),
> > I fear that there is a real concern that name clashes _will_ occur,
> > probably in rare and particular cases, but this might be a cause
> > for some future trouble.

> > The ideal would be to give those queues a dedicated namespace,
> > for instance by putting them in .hg/mq/, where no collision could
> > occur, and in order to retain compatibility .hg/patches/ would
> > simply become a symbolic link pointing there. But symbolic links
> > are not really a possibility on most Windows releases as far as I
> > know, so something else probably has to be found...

> > So I'm afraid I'm not coming up with a good solution or patch, just
> > some thoughts, before it becomes carved in stone with next release.

> Quite devious naming you thought up there. :o)

> I have pushed a revised patch that limits the allowed names somewhat and
> guards against these cases.

Actually, looking more closely at this patch, I'm not sure the test
added under "%% tricky cases" serves any purpose. Creating a queue
named "test" under .hg/store was not a failing case before (only
slightly messy). If an effective test is desired for tricky cases,
it should be for instance "journal" under store, or some equivalently
devious naming schemes like above.. :)

-- 
Cédric