On Wed, 12 Feb 2025, Jeremy Drake via Cygwin-patches wrote:
> On Wed, 12 Feb 2025, Jeremy Drake via Cygwin-patches wrote:
>
> > It was *supposed* to not return the second one. Maybe I broke it when
> > trying to break out of the loop early... I will test this scenario and
> > see why it doesn't work as expected.
>
> Yeah, I never actually looked at how posix_sorted was sorted. It's sorted
> by length first, then by strcmp. This was probably a premature
> optimization anyway. conv_to_posix_path doesn't try to bail early, it
> just continues, and that's going to be a much more common code path than
> this...
Not only that, but there's a sash-vs-backslash mismatch between
native_path and mnt_fsname resulting in the strcasematch not matching. I
must have only tested the root-of-drive-letter case (where the paths are
like C:), because that was the case the existing code was handling:
diff --git a/winsup/cygwin/mount.cc b/winsup/cygwin/mount.cc
index 4be24fbe84..ef3070dbe1 100644
--- a/winsup/cygwin/mount.cc
+++ b/winsup/cygwin/mount.cc
@@ -1645,14 +1645,8 @@ fillout_mntent (const char *native_path, const char
*posix_path, unsigned flags)
struct mntent& ret=_my_tls.locals.mntbuf;
bool append_bs = false;
- /* Remove drivenum from list if we see a x: style path */
if (strlen (native_path) == 2 && native_path[1] == ':')
- {
- int drivenum = cyg_tolower (native_path[0]) - 'a';
- if (drivenum >= 0 && drivenum <= 31)
- _my_tls.locals.available_drives &= ~(1 << drivenum);
append_bs = true;
- }
I have a fix for these, and I also added a patch 3 on top which removes
the de-duplication code and calls cygdrive_posix_path instead of
conv_to_posix_path. You can not apply patch 3, apply patch 3 in case it
has to be reverted later, or squash it into 2 if you prefer, there's
options.
