On Fri, 8 Aug 2025, Jeremy Drake via Cygwin wrote:
> On Fri, 8 Aug 2025, Brian Inglis via Cygwin wrote:
>
> > On 2025-08-08 00:53, Thomas Wolff via Cygwin wrote:
> > > Am 08.08.2025 um 02:31 schrieb Jeremy Drake via Cygwin:
> > >> On a case-insensitive but case-preserving filesystem, is there a Cygwin
> > >> API to get the on-disk case for a given path? It seems like `realpath`
> > >> ought to do it but running
> > >> $ touch case-test
> > >> $ realpath CASE-TEST
> > >> returns CASE-TEST.
> > > On the command line, you could use
> > > ls | grep -i
> > >
> > >> Regardless, canonicalize_file_name or realpath may not
> > >> be what I want because it would dereference symlinks.
> > >>
> > >> Background: I'm trying to debug some test failures in Clang, due to a
> > >> warning that's supposed to be issued when you #include "foo.h" but the
> > >> file on disk that it opened is "Foo.h".
> >
> > Looks like if you use wildcards, it should work correctly:
> >
> > $ lsattr -dl .
> > . ---
> > $ l *_exit*
> > _Exit.2 _exit.3
> > $ l _exit.?
> > _Exit.2 _exit.3
> > $ l _exit.[23]
> > _Exit.2 _exit.3
> > $ l *EXIT-TEST*
> > exit-test
> > $ l *exit*
> > _Exit.2 _exit.3 EXIT exit-test
> >
> > also, you could just opendir(3)/readdir(3)/closedir(3) and strcasecmp(3).
> >
>
> Yeah, globbing is opendir/readdir/closedir, but it'd have to be done on
> each path component to recover the on-disk case for a path. I'll explain
> now that I have a better idea what clang is doing under the hood.
>
> llvm has an API that opens a file with an out parameter for the "real"
> path. On Windows, it uses GetFinalPathNameByHandleW
> (FILE_NAME_NORMALIZED|VOLUME_NAME_DOS), with some massaging for UNC. On
> Unix, it first would prefer fcntl with F_GETPATH. It doesn't look like
> Cygwin provides that. If that's not available, available, it tries
> readlink on /proc/self/fd/%d. If that's not available, it falls back to
> realpath on the name input. I had a breakpoint on realpath that was not
> hit, so it appears that readlink is what it's doing.
>
I came up with this hack that seems to work. It's pretty stupid, but
maybe a start of a discussion? Maybe it'd make more sense to factor out
the GetFinalPathNameByHandleW-handling code from symlink_info::check and
call that to convert it to a posix path?
diff --git a/winsup/cygwin/fhandler/base.cc b/winsup/cygwin/fhandler/base.cc
index 5321ad7ff1..cbdd20d743 100644
--- a/winsup/cygwin/fhandler/base.cc
+++ b/winsup/cygwin/fhandler/base.cc
@@ -29,6 +29,7 @@ details. */
#include "shared_info.h"
#include <asm/socket.h>
#include "cygwait.h"
+#include "tls_pbuf.h"
static const int CHUNK_SIZE = 1024; /* Used for crlf conversions */
@@ -133,6 +134,27 @@ char *fhandler_base::get_proc_fd_name (char *buf)
stpcpy (stpcpy (buf, get_name ()), " (deleted)");
return buf;
}
+ if (get_device () == FH_FS && get_name ())
+ {
+ tmp_pathbuf tp;
+ PWCHAR fpbuf = tp.w_get ();
+ DWORD ret;
+
+ ret = GetFinalPathNameByHandleW (get_handle (), fpbuf, NT_MAX_PATH, 0);
+ if (ret)
+ {
+ PWCHAR ubuf = tp.w_get ();
+ UNICODE_STRING uc = {2, 2, ubuf + sys_mbstowcs (ubuf, NT_MAX_PATH,
get_name ())},
+ fc = {2, 2, fpbuf + ret + 1};
+ while (--uc.Buffer >= ubuf && --fc.Buffer >= fpbuf &&
+ (RtlCompareUnicodeString (&uc, &fc, TRUE) == 0 ||
+ (iswdirsep (*uc.Buffer) && iswdirsep (*fc.Buffer))))
+ if (!iswdirsep (*uc.Buffer))
+ *uc.Buffer = *fc.Buffer;
+ sys_wcstombs (buf, NT_MAX_PATH, ubuf);
+ return buf;
+ }
+ }
if (get_name ())
return strcpy (buf, get_name ());
if (dev ().name ())
--
Problem reports: https://cygwin.com/problems.html
FAQ: https://cygwin.com/faq/
Documentation: https://cygwin.com/docs.html
Unsubscribe info: https://cygwin.com/ml/#unsubscribe-simple
