Ok. Continuing in the same shell:
U:\foo>git status
# On branch master
# Changes to be committed:
# (use "git reset HEAD <file>..." to unstage)
#
# new file: trace1
#
# Untracked files:
# (use "git add <file>..." to include in what will be committed)
#
# trace2
U:\foo>git commit -m "test b"
error: Trying to write ref refs/heads/master with nonexistent object 428dafc292b8396fc7b3c7f692dc9dfe1196a40a
fatal: cannot update HEAD ref
U:\foo>dir /s *a40a
Volume in drive U is Users
Volume Serial Number is FC97-3DA5
Directory of U:\foo\.git\objects\42
05/06/2013 12:26 PM 152 8dafc292b8396fc7b3c7f692dc9dfe1196a40a
1 File(s) 152 bytes
Total Files Listed:
1 File(s) 152 bytes
0 Dir(s) 1,594,477,576,192 bytes free
U:\foo>date
The current date is: Mon 05/06/2013
Enter the new date: (mm-dd-yy)
U:\foo>time
The current time is: 13:10:49.55
Enter the new time:
U:\foo>git cat-file -p 428dafc292b8396fc7b3c7f692dc9dfe1196a40a
tree cecae5b4c87ea21aef513fcfcd5c27fe87e0536f
parent f6a0de35d12a3b58f12bf1af4ff629b8b004ad82
author David Goldfarb <deg@xxxxxxxxx> 1367834997 +0300
committer David Goldfarb <deg@xxxxxxxxx> 1367834997 +0300
test b
U:\foo>U:\foo>
David
-----Original Message-----
From: Thomas Rast [mailto:trast@xxxxxxxxxxx]
Sent: Monday, May 06, 2013 1:08 PM
To: David Goldfarb
Cc: git@xxxxxxxxxxxxxxx
Subject: Re: trouble on windows network share
David Goldfarb <deg@xxxxxxxxx> writes:
> Looks like it works.
>
> From the windows machine:
> U:\foo>git cat-file -p 0b89efdeef245ed6a0a7eacc5c578629a141f856
> 100644 blob b02e7c87fe376a353ea4f014bdb3f5200a946b37 foo1
> 100644 blob 2cbf64f759a62392ad9dfe1fb9c2cdb175876014 foo2
>
> U:\foo>
>
>
> Double-checking that nothing was fixed or changed when I earlier committed the file from Linux, here's a second test:
>
>
> U:\foo>git cat-file -p 0b89efdeef245ed6a0a7eacc5c578629a141f856
> 100644 blob b02e7c87fe376a353ea4f014bdb3f5200a946b37 foo1
> 100644 blob 2cbf64f759a62392ad9dfe1fb9c2cdb175876014 foo2
>
> U:\foo>git status
> # On branch master
> # Untracked files:
> # (use "git add <file>..." to include in what will be committed)
> #
> # trace1
> # trace2
> nothing added to commit but untracked files present (use "git add"
> to track)
>
> U:\foo>git add trace1
>
> U:\foo>git commit trace1 -m "testing"
Note that specifying 'trace1' here is redundant, since you already added it and had no other staged changes. Perhaps you can re-run a test like this without the extra argument for comparison. That would tell us if it matters that the write and read happen in the same process.
> error: unable to find cecae5b4c87ea21aef513fcfcd5c27fe87e0536f
> fatal: cecae5b4c87ea21aef513fcfcd5c27fe87e0536f is not a valid
> object
>
> U:\foo>git cat-file -p cecae5b4c87ea21aef513fcfcd5c27fe87e0536f
> 100644 blob b02e7c87fe376a353ea4f014bdb3f5200a946b37 foo1
> 100644 blob 2cbf64f759a62392ad9dfe1fb9c2cdb175876014 foo2
> 100644 blob 19102815663d23f8b75a47e7a01965dcdc96468c test.txt
> 100644 blob c9009b02950964cf1d5281125e6e2f647dd9dc16 trace1
I'm inclined to just say that your FS is crazy.
What's unsatisfactory is that we already have a bunch of crazy FS workarounds in move_temp_to_file(), which is one obvious candidate for what is going on here. So this wouldn't be something new; just another craziness to work around.
For example, you could test the theory that rename() has something to do with it by patching this into move_temp_to_file():
diff --git i/sha1_file.c w/sha1_file.c
index 67e815b..22af015 100644
--- i/sha1_file.c
+++ w/sha1_file.c
@@ -2635,6 +2635,10 @@ int move_temp_to_file(const char *tmpfile, const char *filename)
/* FIXME!!! Collision check here ? */
}
+ if (access(filename, R_OK) < 0)
+ return error("access(%s, R_OK) failed immediately after rename(): %s",
+ filename, strerror(errno));
+
out:
if (adjust_shared_perm(filename))
return error("unable to set permission to '%s'", filename);
--
Thomas Rast
trast@{inf,student}.ethz.ch
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html