bug#36405: 26.2.90; O_PATH problem on some versions of Cygwin

[Paul Eggert <eggert@cs.ucla.edu>]

[-- Attachment #1: Type: text/plain, Size: 634 bytes --]

Eli Zaretskii wrote:
> What problems did you see/envision on other platforms with the
> original patch?

Mainly the hassle of maintaining code with a lot of hard-to-read ifdefs in it. 
It's better if system-specific stuff is kept to a minimum in mainline code. 
Also, this O_PATH stuff is not that high of a priority, as the code works quite 
well in practice without O_PATH. (Also, if Cygwin botched O_PATH on FIFOs 
there's a reasonable chance it botched O_PATH elsewhere too. :-)

How about something like the attached patch instead? It's simpler and should be 
a bit faster on Cygwin than the original patch. I haven't tested it.

[-- Attachment #2: cygwin.diff --]
[-- Type: text/x-patch, Size: 1023 bytes --]

diff --git a/configure.ac b/configure.ac
index 8ff0e21fbf..a0e99ac393 100644
--- a/configure.ac
+++ b/configure.ac
@@ -5734,6 +5734,9 @@ AC_DEFUN
     AC_MSG_WARN([[building Emacs on Cygwin 1.5 is not supported.]])
            echo
 	   ;;
+  cygwin,3.0.[0-8]'('*)
+    AC_DEFINE([HAVE_CYGWIN_O_PATH_BUG], 1,
+      [Define to 1 if opening a FIFO with O_PATH causes a hang.]);;
 esac
 
 # Remove any trailing slashes in these variables.
diff --git a/src/dired.c b/src/dired.c
index b8197d36a0..b700013f6a 100644
--- a/src/dired.c
+++ b/src/dired.c
@@ -41,10 +41,6 @@
 #include "buffer.h"
 #include "coding.h"
 
-#ifdef __CYGWIN__
-# undef O_PATH /* Buggy in Cygwin 3.0.0 through 3.0.7.  */
-#endif
-
 #ifdef MSDOS
 #include "msdos.h"	/* for fstatat */
 #endif
@@ -941,7 +937,7 @@ file_attributes (int fd, char const *name,
 
   int err = EINVAL;
 
-#ifdef O_PATH
+#if defined O_PATH && !defined HAVE_CYGWIN_O_PATH_BUG
   int namefd = openat (fd, name, O_PATH | O_CLOEXEC | O_NOFOLLOW);
   if (namefd < 0)
     err = errno;