The OS is the one that's meant to take the umask into account. When you write out a file and ask for 0777, the OS will apply the umask and leave that in e.g. 0755. It seems really strange that we would have to do it ourselves.

No, we explicitly call chmod.

I think @carlosmn is correct and we should probably just not call chmod in this case. I'm not quite sure what the implications of that are without doing a more thorough audit of the places that use this code, though...

So you want to move the mode to open? Does win32 preserve mode across
renames?

On Mon, Nov 4, 2013 at 5:08 PM, Russell Belfer [email protected]:

I think @carlosmn https://github.com/carlosmn is correct and we should
probably just not call chmod in this case. I'm not quite sure what the
implications of that are without doing a more thorough audit of the places
that use this code, though...

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/1940#issuecomment-27727651
.

Ah, you're right.When calling chmod do we have to it ourselves, but we probably shouldn't be doing that at all, as git_futils_creat_locked already opens with the specified mode. The chmod call overwrites the OS' hard work.

If Win32 doesn't preserve the mode across renames, then we're still buggy (and have been for a while), as the rename call happens after the chmod of the locked file.

Ok, so it's a pretty large change to refactor this now to fix the issue. We should do this in the future, as the call to chmod is silly when we always know beforehand what mode we want to write with and should just do the sensible thing and tell open; but we missed that when reviewing the incoming changes.

So let's go with this and we can do the refactor later on, as it's purely an internal API.

Well, we also call mkstemp, so we'd need to call chmod either way in
that case.

I think that the only advantage to doing it in git_filebuf_open is that
it is reminiscent of open(2). Which would be very nice indeed, but yes,
getting this fixed would be very nice and fixing the API would be nice to
have.

On Mon, Nov 4, 2013 at 5:36 PM, Carlos Martín Nieto <
[email protected]> wrote:

Ok, so it's a pretty large change to refactor this now to fix the issue.
We should do this in the future, as the call to chmod is silly when we
always know beforehand what mode we want to write with and should just do
the sensible thing and tell open; but we missed that when reviewing the
incoming changes.

So let's go with this and we can do the refactor later on, as it's purely
an internal API.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/1940#issuecomment-27729890
.

Okay, let's move that then so that we can get this merged. World writable files in the git repo are for real bad.

Yo lets merge this so the build turns green and we stop writing world writable files. @cmn are you happy with the API change?

👍 ❤️