using the same repo with different OS

I have my repos on a USB key (fat32 fs) and I work with git under
linux and windows (maybe also Solaris in the future).

Considering also the new MinGW port, should I use a different repo for
every OS I work with???

Are planned/expected OS-specific files in the repo preventing me to
use the same whenever I'm working???

thanks in advance

---
Stefano Spinucci

On 2/25/07, Stefano Spinucci [off-list ref] wrote:

Are planned/expected OS-specific files in the repo preventing me to
use the same *whenever* I'm working???

:-(

obviously, I mean "*wherever* I'm working???"

---
Stefano Spinucci

Hi,

On Sun, 25 Feb 2007, Stefano Spinucci wrote:

I have my repos on a USB key (fat32 fs) and I work with git under
linux and windows (maybe also Solaris in the future).

Considering also the new MinGW port, should I use a different repo for 
every OS I work with???

You don't need to. But remember that the stat informations are likely to 
be borked. For example, when you do a "git-status" on a sizable working 
tree on that stick, it will take a long, long time whenever you switch 
OSes. Basically the same amount of time as if git did not have an index!

Are planned/expected OS-specific files in the repo preventing me to use 
the same whenever I'm working???

It could be possible that the index is saved in a slightly different 
format, due to different alignment strategies. Maybe 64-bit vs 32-bit has 
differences there. I have no idea.

Ciao,
Dscho

Johannes Schindelin [off-list ref] writes:

It could be possible that the index is saved in a slightly different 
format, due to different alignment strategies. Maybe 64-bit vs 32-bit has 
differences there. I have no idea.

As far as I am aware of, there should be no such difference,
although we probably rely on int being 32 and long being 64
bit.  We do store our numbers in network byte order.

However, did somebody say FAT-32?  I seem to recall that that
filesystem lacks uid/gid, so I am reasonably sure there would be
different return values for the same path from lstat(2) between
git on Cygwin/MinGW and on Linux that mounts the same filesystem
using vfat driver with different uid= & gid= mount options.
That difference which would make the paths stat-dirty.

Also there is the problem of i-num we discussed lately.  The
i-num we get for the same path from lstat(2) can be different
when same filesystem is mounted by different operating systems
(remember that NTFS on Linux and Cygwin?).

So even though file format of the index may be designed and
implemented to be portable, it might not be practical to use
the same physical repository from different OSes, depending on
the combination of things.

Hi,

On Sun, 25 Feb 2007, Junio C Hamano wrote:

So even though file format of the index may be designed and implemented 
to be portable, it might not be practical to use the same physical 
repository from different OSes, depending on the combination of things.

Yes.

However, in the case of a _bare_ repo, which is meant purely to transport 
(possibly huge) changes from A to B, a USB stick can be used _without_ 
problems. I do that myself quite often.

Ciao,
Dscho