Re: [PATCH] remote-hg: store converted URL

From: Max Horn <max@quendi.de>
Date: Jan 15, 2013, 11:41 UTC
Message-ID: <64C81CD0-960A-47F2-89FC-8D3126B1F4D5@quendi.de>
In-Reply-To: <7vmwwbd43o.fsf@alter.siamese.dyndns.org>

On 14.01.2013, at 19:14, Junio C Hamano wrote:

Show 14 quoted lines

> Max Horn <max@quendi.de> writes:
> 
>> From: Felipe Contreras <felipe.contreras@gmail.com>
>> 
>> Mercurial might convert the URL to something more appropriate, like an
>> absolute path.
> 
> "What it is converted *TO*" is fairly clear with ", like an ...",
> but from the first reading it was unclear to me "what it is
> converted *FROM*" and "*WHEN* the conversion happens".  Do you mean
> that the user gives "git clone" an URL "../hg-repo" via the command
> line (e.g. the argument to "git clone" is spelled something like
> "hg::../hg-repo"), and that "../hg-repo" is rewritten to something
> else (an absolute path, e.g. "/srv/project/hg-repo")?

Yes, that was meant.

Show 16 quoted lines

> 
>> Lets store that instead of the original URL, which won't
>> work from a different working directory if it's relative.
> 
> What is lacking from this description is why it even needs to work
> from a different working directory.  I am guessing that remote-hg
> later creates a hidden Hg repository or something in a different
> place and still tries to use the URL to interact with the upstream,
> and that is what breaks, but with only the above description without
> looking at your original report, people who will read the "git log"
> output and find this change will not be able to tell why this was
> needed, I am afraid.
> 
> Of course, the above guess of mine may even be wrong, but then that
> is yet another reason that the log needs to explain the change
> better.

Fully agreed. How about this commit message:

-- >8 -- remote-hg: store converted URL of hg repo in git config

When remote-hg is invoked, read the remote repository URL from the git config, give Mercurial a chance to expand it, and if changed, store it back into the git config.

This fixes the following problem: Suppose you clone a local hg repository
using a relative path, e.g.
  git clone hg::hgrepo gitrepo
This stores "hg::hgrepo" in gitrepo/.git/config. However, no information
about the PWD is stored, making it impossible to correctly interpret the
relative path later on. Thus when latter attempting to, say, "git pull"
from inside gitrepo, remote-hg cannot resolve the relative path correctly,
and the user sees an unexpected error.

With this commit, the URL "hg::hgrepo" gets expanded (during cloning, but also during any other remote operation) and the resulting absolute URL (e.g. "hg::/abspath/hgrepo") is stored in gitrepo/.git/config. Thus the git clone of hgrepo becomes usable. -- >8 --

Show 9 quoted lines

> 
>> Suggested-by: Max Horn <max@quendi.de>
>> Signed-off-by: Felipe Contreras <felipe.contreras@gmail.com>
>> Signed-off-by: Max Horn <max@quendi.de>
>> ---
>> For a discussion of the problem, see also
>>  http://article.gmane.org/gmane.comp.version-control.git/210250
> 
> I do not see any discussion; only your problem report.

Aha, an english language issue on my side I guess: For me, a single person can "discuss" a problem (often, a research paper is said to be "discussing a problem"). Sorry for that. Anyway, the reason I gave that link was because it attempts explains the problem and one solution (which this patch ended up implementing), but also express that I feel a bit uncomfortable with this. Which I still do. Relying on the remote helper to invoke "git config" feels like a hack and I was wondering whether this is deemed an acceptable solution -- or whether one should instead extend the remote-helper protocol, allowing the remote helper to signal a rewritten remote URL (perhaps only directly after a clone?). As it is, the remote helper seems (?) to have no way to distinguish whether it is being called duri
 ng a clone or a pull; hence it has to "expand" and rewrite the URL every time it is called, just in case.

Anyway, as long as this particular command works somehow, I am fine:

  git clone hg::../relative/path/to/hg-repo  git-repo

> Was this work done outside the list?  I just want to make sure this
> patch is not something Felipe did not want to sign off for whatever
> reason but you are passing it to the list as a patch signed off by
> him.

The work was done by Felipe's and published in his github repository:
  https://github.com/felipec/git/commit/605bad5b52d2fcf3d8f5fd782a87d7c97d1b040a
See also the discussion (yeah, this time a real one ;-) leading to this:
  https://github.com/felipec/git/issues/2

I took his sign-off from there and interpreted it as saying that Felipe was OK with this being pushed to git.git. But perhaps this is not what I should have done? In that case I am very sorry :-(. It's just that I feel this patch is quite useful and important for daily use (which is why I suggested it in the first place ;-), so I was/am quite eager to see it in.

Cheers, Max

PS: recently, yet another tool has (re)emerged for using hg repos from inside git:
  https://github.com/buchuki/gitifyhg
This is partially based on Felipe's work, but has several bug fixes atop that. It is also seems to be a priority for its author, so it os more actively developed... anyway, that's now, what, "solution" #5 or #6? I really hope the dust on this will settle soon and we'll have just one (or maybe two) tools doing a decent job, instead of attention splitting over so many different ones...