Paul Eggert reported that unlink/rmdir vs. e.g., /proc/self/fd/N/FILE
fails with ENOSYS. This makes openat revert to using save-cwd.c
functions in that case.
(dev_from_mount_options)
[defined MOUNTED_GETMNTENT1 || defined MOUNTED_GETMNTENT2]:
New function. It no longer assumes "dev=" has the System V meaning
on Linux (since it doesn't). It also parses "dev=" more carefully.
(read_file_system_list)
[defined MOUNTED_GETMNTENT1 || defined MOUNTED_GETMNTENT2]: Use it.
MOUNTED_GETMNTENT2 is new here; the code didn't used to look for
dev= in that case.
days most people use C99-compatible compilers to debug, so it's
not worth worrying about catering to older compilers for that.
This works around some porting problems with HP-UX compilers.
(false, true) [defined __BEOS__]: Don't #undef; no longer needed.
(_Bool): typedef to bool if C++ or BeOS, and #define to signed char
otherwise.
(openat_permissive): New function -- used in remove.c rewrite.
(all functions): Set errno just before returning, only if there
was an actual failure.
Use EXPECTED_ERRNO rather than comparing against only ENOTDIR.
Idea and some code from Ulrich Drepper's glibc changes.
(BUILD_PROC_NAME): New macro.
Include <stdio.h>, <string.h>, "alloca.h" and "intprops.h".
(rpl_openat): Emulate by trying to open /proc/self/fd/%d/%s,
before falling back on save_cwd and restore_cwd.
(fdopendir, fstatat, unlinkat): Likewise.
business.
(gethrxtime) [! (HAVE_NANOUPTIME
|| (defined CLOCK_MONOTONIC && HAVE_CLOCK_GETTIME)
|| HAVE_MICROUPTIME)]: Fall back on gettime rather than rolling
our own approximation.
