Bug Report: hg {copy, rename, mv} with ".." to an empty directory fails with obscure error

[Matt Mackall]

On Sun, Oct 07, 2007 at 11:29:13AM -0700, Mike Neuman wrote:
> I made a one line change to canonpath which seems to fix the problem, 
> but introduce a new one:
> 
> *** /usr/local/lib/python2.4/site-packages/mercurial/util.py    Mon Jun 
> 25 19:18:18 2007
> --- util.py     Sun Oct  7 11:24:01 2007
> ***************
> *** 366,371 ****
> --- 366,372 ----
>       if not os.path.isabs(name):
>           name = os.path.join(root, cwd, name)
>       name = os.path.normpath(name)
> +     name = os.path.realpath(name)

Good spotting!

Hmmm, that means someone else is using realpath. We should probably be
only using one or the other project-wide.

$ hgrep normpath | wc
     20      82    1273
$ hgrep realpath | wc
      5      22     355

Normpath is preferred as it's just a string op. But we can get into
trouble when someone goes through a symlink. We can probably be clever
and use realpath only as a fallback but it will take some care.

> Now the problem is the directory you're in gets blown away right from 
> under you. Semantically, this makes sense, but it's pretty confusing to 
> the user:
> 
> % pwd
> /usr/home/mcn/devel/mercurial-0.9.4/test-rep
> % hg init .
> % mkdir dir1 dir2
> % touch dir1/file1
> % hg add dir1/file1
> % hg commit -m test
> % cd dir1
> % hg mv file1 ../dir2
> % pwd
> pwd: .: No such file or directory
> % cd ..
> /home

That's a feature.

-- 
Mathematics is the supreme nostalgia of our time.