[PATCH] filemerge: wrap quotes around tool path

[Matt Mackall]

On Mon, 2008-02-04 at 18:15 -0200, Alexis S. L. Carvalho wrote:
> Thus spake Matt Mackall:
> > 
> > On Mon, 2008-02-04 at 12:33 -0600, Steve Borho wrote:
> > > # HG changeset patch
> > > # User Steve Borho <steve at borho.org>
> > > # Date 1202147114 21600
> > > # Node ID 1da25064cd8eb0278fd3414b78a16a0f8729cf96
> > > # Parent  288ec2f6faa2a362dafa8da81bd03d96e36be5ee
> > > filemerge: wrap quotes around tool path
> > > 
> > > diff --git a/mercurial/filemerge.py b/mercurial/filemerge.py
> > > --- a/mercurial/filemerge.py
> > > +++ b/mercurial/filemerge.py
> > > @@ -184,7 +184,7 @@
> > >          replace = dict(local=a, base=b, other=c, output=out)
> > >          args = re.sub("\$(local|base|other|output)",
> > >                        lambda x: '"%s"' % replace[x.group()[1:]], args)
> > > -        r = util.system(toolpath + ' ' + args, cwd=repo.root, environ=env)
> > > +        r = util.system('"'+toolpath+'" '+args, cwd=repo.root, environ=env)
> > >  
> > >      if not r and _toolbool(ui, tool, "checkconflicts"):
> > >          if re.match("^(<<<<<<< .*|=======|>>>>>>> .*)$", fcm.data()):
> > 
> > Hmm, that breaks things like:
> > 
> > HGMERGE=script -with -some -other -args
> > 
> > We were using precisely that trick in run-tests.py until yesterday, so I
> > suspect other people are using it too.
> 
> We're still using it in some tests:
> 
> $ grep HGMERGE=\" *
> test-filebranch:HGMERGE="python ../merge"; export HGMERGE
> test-merge-symlinks:HGMERGE="python ../echo.py" hg merge
> test-merge-symlinks:HGMERGE="python ../echo.py" hg up 3
> test-merge1:HGMERGE="python ../merge"; export HGMERGE
> test-merge6:HGMERGE="python ../merge"; export HGMERGE
> test-rename-merge2:HGMERGE="python ../merge"; export HGMERGE
> 
> And AFAICS they only work because the script happens to be in another
> directory - we call util.find_exe on the value of $HGMERGE to see if the
> tool is around; that function sees an os.sep in the string and assumes
> it's a full path.

Hmm, indeed. We should instead always call HGMERGE exactly as set. Doing
this right probably means having _picktool return (tool, toolpath). That
has the advantage of only calling _findtool once, which is nice.

> [ui]
> merge = command -arg

But this case is a little different. We can treat this a couple ways:

a) make ui.merge always take precedence over tools from merge-tools,
regardless of symlink/binary/gui/available tests
b) try to be clever about spaces (tricky!)
c) disallow this trick
d) leave it all to os.system to figure out

-- 
Mathematics is the supreme nostalgia of our time.