C-x v i bug

[Matt Mackall]

On Fri, 2009-12-18 at 07:54 -0800, Dan Nicolaescu wrote:
> Martin Geisler <mg at lazybytes.net> writes:
> 
>   > Dan Nicolaescu <dann at ics.uci.edu> writes:
>   > 
>   > > Let's first talk about the original problem that started this
>   > > discussion.
>   > >
>   > > When a file in a directory that is under mercurial control is opened
>   > > in emacs, emacs runs "hg status FILE" so that it knows if it's
>   > > registered or not, if it's modified, etc.
>   > >
>   > > Any user settings in .hgrc should be irrelevant to the above. Right?
>   > 
>   > Right. Many people use the color extension to get better feedback from
>   > 'hg status', but if Emacs sets TERM=dumb, then the extension will
>   > disable itself. I'm just mentioning color to say that there are useful
>   > extensions out there that modify even basic commands like 'hg status'.
>   > 
>   > > It's desirable that this is as fast as possible, so processing .hgrc,
>   > > initializing plugins will just waste time.
>   > > After that emacs will want to know the version number for the file, for that
>   > > it runs "hg log -l1 FILE", and parse it from the output.
>   > > Any user settings in .hgrc should be irrelevant for this command.  Right?
>   > 
>   > Right, and it's even quite important that you disable localization (run
>   > hg with LANGUAGE=C in the environment). Otherwise you'll end up parsing:
>   > 
>   >   % hg log -l1 README
>   >   ændring:     9586:a41f2840f9c6
>   >   bruger:      Lee Cantey <lcantey at gmail.com>
>   >   dato:        Tue Oct 13 12:27:50 2009 -0700
>   >   uddrag:      README: revert accidental commit
>   > 
>   > The user could also very well have installed a different default style
>   > by setting ui.style. On the command line it's done line this:
> 
> Thank you, this was very useful in taking care of some issues in emacs.
> 
>   >   % hg log -l1 README --style=compact
>   >   9586   a41f2840f9c6   2009-10-13 12:27 -0700   lcantey
>   >     README: revert accidental commit
>   > 
>   > > [too bad that the status and version number are not available from a
>   > > single command...]
>   > 
>   > Well, you know, files don't really have a version number with modern
>   > version control systems. The entire tree has a version number... You can
>   > of course ask about when a file was last touched, but I think that
>   > information is getting more and more irrelevant these days.
> 
> In emacs the generic Version Control layer needs a version number in some case.
> Here's an example from a bug report:
> 
> cd  /tmp
> mkdir hgtest2
> cd hgtest2
> hg init
> echo foo > foo.txt
> hg add foo.txt
> hg commit -m "Added foo.txt"
> hg branch bar
> echo bar > foo.txt
> hg commit -m "Changed foo to bar"
> hg update -r default
> echo frobozz > frobozz.txt
> hg add frobozz.txt
> hg commit -m "Added frobozz.txt"
> 
> 
> now open the file mkdir /tmp/hgtest2/foo.txt and ask to see the
> annotated version, emacs does that by running
> 
> hg annotate -r REVISION foo.txt
>
> How can REVISION be obtained in this case?
> It should be "0", but
> hg log -l1 foo.txt
> does not show that...

Version numbers are not per-file in Mercurial. The number you should use
is the global number (or numbers!) reported by hg parents. This revision
is also known as '.', eg 'hg annotate -r . foo.txt'. 

You're probably thinking "but I actually want to report the last
changeset this file was touched in to be more like CVS". No, you don't,
that's not the Mercurial way. Trying to emulate CVS is doing your users
a disservice.

As a side note, if you want history relative to the working directory
(and not just all of history), you'll want the -f flag to log.

-- 
http://selenic.com : development and support for Mercurial and Linux