Reporting "nearly-exact" adds (Was: Windows test suite state)

[Adrian Buehlmann]

On 06.06.2008 01:42, Adrian Buehlmann wrote:
> Sigh. It seems we have some more serious issues with hg add
> than just test output differences:
> 
> With your patches applied (plus that known bug fixed), on Windows
> (hgt = hg -- just my wrapper bat):
> 
>> hgt sta
> --- running hg from W:\hg-crew-paul
> ? x\a.txt
> 
>> hgt add x\a.txt
> --- running hg from W:\hg-crew-paul
> adding x\a.txt
> 
>> hgt sta
> --- running hg from W:\hg-crew-paul
> A x\a.txt
> ? x\a.txt
> 
> last output looks like there is some serious confusion
> in the dirstate (same file both "unkown" and "added").

Paul, this has been outdated by your newest resending of your
patches (which is good). But I just wanted to add something to the
above, which I think is important (I used that same outdated codebase
as above):

Looking at:

> hgt debugstate
--- running hg from W:\hg-crew-paul
a   0         -1 unset               x\a.txt

I see that in this old version of the codebase, the paths
in the dirstate used *backslases* for the path separator.

With your newest codebase (c8b1fb9e1957, checked by using your
binary I downloaded) I see debugstate now returns *forward* slashes
-- which seems very good to me.

Since the dirstate is stored as a file at .hg/dirstate,
I think I now understand that the path separators inside the
dirstate by all means must be *forward* slahes (*not* backslashes).

The reason for this seems clear: the whole repo
-- that is, the directory which contains the .hg dir --
with all its contents ("working directory" plus .hg dir) must
be:

* Compatible from one Mercurial release to the next
* Transferrable from unix to Windows and vice versa

Unless there is a mechanism I don't know of which detects and
rebuilds the dirstate in both of these cases...