[PATCH] issue 1364

[Patrick Mézard]

Petr Kodl a écrit :
> On Mon, Oct 27, 2008 at 12:49 PM, Andrei Vermel <andrei.vermel at gmail.com
> <mailto:andrei.vermel at gmail.com>> wrote:
> 
>     > The patch posted works as far as 'it does not break' anything, but
>     testing
>     on the
>     > actual machine that shows the original problem would definitely help.
> 
>     I confirm that Petr's patch fixes the problem on my kit.
>     The patch appears to be corrupted by mailer though, I had to edit it to
>     apply.
> 
> 
> I looked through the fstat call variants in CRT source - they seem to be
> 'broken' the same way in both VC7 (2003) and VC9 (2008) - which covers
> the compilation platforms used by current Python distributions from 2.3
> -> 3.0. In order to make the code behave the same way as they do we
> would have to replicate their approach:
> 
> FILETIME -> LOCAL FILE TIME -> SYSTEM TIME -> custom conversion to UTC time
> 
> The code manipulates the timezone globals along the way by using several
> undocumented internal functions - not very nice code to look at in
> general. It also uses custom calculation of time conversions that
> are different than NT kernel level versions.
> 
> Comparing to reimplementation of this mess both options bellow look very
> clean ;-(.
> 
> 1) do nothing and disable the fast osutil.c for Python <= 2.4

Call me lazy but I really prefer this one. Who would it be painful for ? People building Mercurial from sources, under Windows, with huge repositories and stuck with python < 2.4. Probably not a very large crowd. Andrei, are you in this case ? I am really trying to tell whether a "lazy" decision would be a real showstopper for some people.

--
Patrick Mézard