encode/decode filter weirdness

[Matt Mackall]

On Tue, Sep 20, 2005 at 02:40:06PM -0700, Maquelin, Olivier wrote:
> >The filters are applied in exactly two places:
> >
> > - reading files from the working directory
> > - writing files to the working directory
> >
> >Thus, the Mercurial internals never see files except in their
> >canonicalized form. For instance, checking whether a file is changed
> >is done by reading it from the working directory (w/filter) and
> >comparing it to what's in the repo (unfiltered).
> >
> >> Now let's try to extract the canonical representation from the
> >> repository by deleting the file and having it re-created:
> >> 
> >> [~/foo] rm foo.txt 
> >> [~/foo] hg status
> >> R foo.txt
> >> [~/foo] hg revert
> >> [~/foo] hg status
> >> M foo.txt
> >> [~/foo] hg diff
> >> [~/foo] od -a foo.txt
> >> 0000000   o   n   e  nl   t   w   o  nl
> >> 0000010
> >> 
> >> The file contents are as expected, but why is 'hg status' 
> >reporting the
> >> file as being modified?
> >
> >Because the revert command changed the size of the file without
> >changing the dirstate. The file is indeed changed.
> 
> Wouldn't this issue go away if the size of the filtered contents were
> used instead of the size of the original file? Of course, this might
> cause a significant performance hit.

This isn't an "issue", it's correct behavior. The file in your working
directory does not match what was last checked out or committed and hg
status is reporting it as such.

> >If your filters are not matched, you'll get interesting results.
> 
> Hmm... I had not realized there was such a limitation. I guess this
> means I will have to be careful with clone/push/pull between Windows and
> Linux, at least when communicating through a shared filesystem. For
> example, instead of cloning a repository on Windows for later use under
> Linux it seems I should create an empty repository from Linux, push (or
> pull) into it and then update from Linux.

The only place where you get into trouble is if you check out files on
Windows (with r/w filters) and then work on them on UNIX (w/o r/w
filters). But you can push/pull/clone between the two as much as you like.

> Btw, I just tried that and had to use 'hg push -f' to push a single,
> initial changeset into an empty repository. Is that the expected
> behavior (because the number of heads increases from 0 to 1)?

Yes.

-- 
Mathematics is the supreme nostalgia of our time.