Re: [PATCH] errno: make EEXIST != ENOTEMPTY on AIX
Hi Paul, > +2024-07-31 Paul Eggert > + > + errno: make EEXIST != ENOTEMPTY on AIX > + ... > + * tests/test-errno.c (e1, ..., e131): Remove, replacing with ... > + (CHECK_POSIX_ERRNOS, POSITIVE_INTEGER_CONSTANT_EXPRESSION) > + (INDEXED_BY_ERRNO, ERRNO_COUNT): These new macros. > + Check that all errno values are positive integer constant expressions. > + Check that they are all distinct, except perhaps for > + EWOULDBLOCK == EAGAIN and ENOTSUP == EOPNOTSUPP. > + Also check ESOCKTNOSUPPORT, added in POSIX.1-2024. > + Also, check that errno values are distinct except when POSIX says > + they needn’t be distinct, since POSIX.1-2024 gives license to > + GNU/Linux’s non-distinct values. This patch makes 'test-errno' uncompilable on GNU/Hurd. The object file test-errno.o is 1073745616 bytes large (> 1 GB), and I had to kill the link command that was meant to produce the 'test-errno' program before it caused an out-of-memory on my machine. I think this change needs to be reworked. Did it possibly embody the assumption that errno values are small integers or all near together? Bruno
fchmodat: Fix cross-compilation guess
When checking a $host_os, it's better to allow a kernel version number than not. 2024-08-07 Bruno Haible fchmodat: Fix cross-compilation guess. * m4/fchmodat.m4 (gl_FUNC_FCHMODAT): Tolerate Linux version number in $host_os. diff --git a/m4/fchmodat.m4 b/m4/fchmodat.m4 index e71ee8e8ef..99dcf58aa7 100644 --- a/m4/fchmodat.m4 +++ b/m4/fchmodat.m4 @@ -1,5 +1,5 @@ # fchmodat.m4 -# serial 8 +# serial 9 dnl Copyright (C) 2004-2024 Free Software Foundation, Inc. dnl This file is free software; the Free Software Foundation dnl gives unlimited permission to copy and/or distribute it, @@ -70,12 +70,12 @@ AC_DEFUN([gl_FUNC_FCHMODAT] esac ], [case "$host_os" in - # Guess no on Linux with glibc and Cygwin. -linux-gnu* | cygwin*) gl_cv_func_fchmodat_works="guessing no" ;; - # Guess 'nearly' on AIX. -aix*) gl_cv_func_fchmodat_works="guessing nearly" ;; - # If we don't know, obey --enable-cross-guesses. -*) gl_cv_func_fchmodat_works="$gl_cross_guess_normal" ;; + # Guess no on Linux with glibc and Cygwin. +linux*-gnu* | cygwin*) gl_cv_func_fchmodat_works="guessing no" ;; + # Guess 'nearly' on AIX. +aix*) gl_cv_func_fchmodat_works="guessing nearly" ;; + # If we don't know, obey --enable-cross-guesses. +*) gl_cv_func_fchmodat_works="$gl_cross_guess_normal" ;; esac ]) rm -f conftest.fchmodat])
Re: [PATCH] errno: make EEXIST != ENOTEMPTY on AIX
On 2024-08-07 02:58, Bruno Haible wrote:
Did it possibly embody the
assumption that errno values are small integers or all near together?
Yes it did. My preference for static checking went overboard, it seems.
I installed the attached, which I hope fixes it.From 37fcd9c9d822150f807211de0fec6c7d86de60ef Mon Sep 17 00:00:00 2001
From: Paul Eggert
Date: Wed, 7 Aug 2024 07:23:51 -0700
Subject: [PATCH] errno-tests: port to GNU/Hurd
Test for errno distinctness dynamically rather than statically,
since the static test blows up the compiler on Hurd.
Problem reported by Bruno Haible in:
https://lists.gnu.org/r/bug-gnulib/2024-08/msg00039.html
Also, test that errno values can all be used in #if,
and improve diagnostics.
* tests/test-errno.c: Include stdio.h, stdlib.h, string.h.
(USABLE_IN_IF): New macro. Use it to check errno values in #if.
(ERRTAB): New macro.
(struct nameval): New type.
(errtab): New global variable.
(errtab_cmp): New function.
(main): Test for errno distinctness dynamically not statically.
Diagnose lack of distinctness better.
---
ChangeLog | 18 +++
tests/test-errno.c | 57 +++---
2 files changed, 62 insertions(+), 13 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index 8e0896f127..df710ddfa6 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,21 @@
+2024-08-07 Paul Eggert
+
+ errno-tests: port to GNU/Hurd
+ Test for errno distinctness dynamically rather than statically,
+ since the static test blows up the compiler on Hurd.
+ Problem reported by Bruno Haible in:
+ https://lists.gnu.org/r/bug-gnulib/2024-08/msg00039.html
+ Also, test that errno values can all be used in #if,
+ and improve diagnostics.
+ * tests/test-errno.c: Include stdio.h, stdlib.h, string.h.
+ (USABLE_IN_IF): New macro. Use it to check errno values in #if.
+ (ERRTAB): New macro.
+ (struct nameval): New type.
+ (errtab): New global variable.
+ (errtab_cmp): New function.
+ (main): Test for errno distinctness dynamically not statically.
+ Diagnose lack of distinctness better.
+
2024-08-07 Bruno Haible
fchmodat: Fix cross-compilation guess.
diff --git a/tests/test-errno.c b/tests/test-errno.c
index 7403ee1432..acd07ab95b 100644
--- a/tests/test-errno.c
+++ b/tests/test-errno.c
@@ -20,6 +20,10 @@
#include
+#include
+#include
+#include
+
/* Check all POSIX-defined errno values, using M (v) to check value v. */
#define CHECK_POSIX_ERRNOS(m) \
m (E2BIG) \
@@ -107,24 +111,51 @@
#define POSITIVE_INTEGER_CONSTANT_EXPRESSION(e) static_assert (0 < (e) << 0);
CHECK_POSIX_ERRNOS (POSITIVE_INTEGER_CONSTANT_EXPRESSION)
+/* Verify that errno values can all be used in #if. */
+#define USABLE_IN_IF(e) ^ e
+#if 0 CHECK_POSIX_ERRNOS (USABLE_IN_IF)
+#endif
+
+/* Check that errno values all differ, except possibly for
+ EWOULDBLOCK == EAGAIN and ENOTSUP == EOPNOTSUPP. */
+#define ERRTAB(e) { #e, e },
+static struct nameval { char const *name; int value; }
+ errtab[] = { CHECK_POSIX_ERRNOS (ERRTAB) };
+
+static int
+errtab_cmp (void const *va, void const *vb)
+{
+ struct nameval const *a = va, *b = vb;
+
+ /* Sort by value first, then by name (to simplify later tests).
+ Subtraction cannot overflow as both are positive. */
+ int diff = a->value - b->value;
+ return diff ? diff : strcmp (a->name, b->name);
+}
+
int
main ()
{
+ int test_exit_status = EXIT_SUCCESS;
+
/* Verify that errno can be assigned. */
errno = EOVERFLOW;
/* Check that errno values all differ, except possibly for
- EWOULDBLOCK == EAGAIN and ENOTSUP == EOPNOTSUPP. */
- #define INDEXED_BY_ERRNO(e) [e] = 1,
- #define ERRNO_COUNT(e) 0,
- static char const
-indexed_by_errno[] = { CHECK_POSIX_ERRNOS (INDEXED_BY_ERRNO) },
-errno_count[] = { CHECK_POSIX_ERRNOS (ERRNO_COUNT) };
- int distinct_errnos = 0;
- for (int i = 0; i < sizeof indexed_by_errno; i++)
-distinct_errnos += indexed_by_errno[i];
- return ((sizeof errno_count
- - (EWOULDBLOCK == EAGAIN)
- - (ENOTSUP == EOPNOTSUPP))
- != distinct_errnos);
+ EAGAIN == EWOULDBLOCK and ENOTSUP == EOPNOTSUPP. */
+ int nerrtab = sizeof errtab / sizeof *errtab;
+ qsort (errtab, nerrtab, sizeof *errtab, errtab_cmp);
+ for (int i = 1; i < nerrtab; i++)
+if (errtab[i - 1].value == errtab[i].value)
+ {
+fprintf (stderr, "%s == %s == %d\n",
+ errtab[i - 1].name, errtab[i].name, errtab[i].value);
+if (! ((strcmp ("EAGAIN", errtab[i - 1].name) == 0
+&& strcmp ("EWOULDBLOCK", errtab[i].name) == 0)
+ || (strcmp ("ENOTSUP", errtab[i - 1].name) == 0
+ && strcmp ("EOPNOTSUPP", errtab[i].name) == 0)))
+ test_exit_status = EXIT_FAILURE;
+ }
+
+ return test_exit_status;
}
--
2.43.0
Re: [PATCH] errno: make EEXIST != ENOTEMPTY on AIX
Hi Paul, > > Did it possibly embody the > > assumption that errno values are small integers or all near together? > > Yes it did. My preference for static checking went overboard, it seems. > I installed the attached, which I hope fixes it. Yes, it fixes it. Thanks. 'test-errno' links fine, succeeds, and prints EAGAIN == EWOULDBLOCK == 1073741859 Bruno
Re: test-pthread-rwlock failure on Pop!_OS 22.04 LTS
On 2024-07-07, I stated: > This provides enough justification for gnulib to override the glibc behaviour > here. I will therefore go ahead and override > PTHREAD_RWLOCK_INITIALIZER > pthread_rwlock_init > pthread_rwlock_attr_init > to use PREFER_WRITER or PREFER_WRITER_NONRECURSIVE, which both behave in a > sane way even in glibc ≥ 2.25. Done through these two patches: 2024-08-07 Bruno Haible pthread-rwlock tests: Strengthen tests. * tests/test-pthread-rwlock-waitqueue.c: New file. * modules/pthread-rwlock-tests (Files): Add it. (Depends-on): Add extensions, nanosleep, stdbool. (Makefile.am): Arrange to test test-pthread-rwlock-waitqueue. 2024-08-07 Bruno Haible pthread-rwlock: Fix default wait queue behaviour on glibc/Linux. * lib/pthread.in.h: If REPLACE_PTHREAD_RWLOCK_INIT is 1 but REPLACE_PTHREAD_RWLOCK_DESTROY is 0, override PTHREAD_RWLOCK_INITIALIZER. * lib/pthread-rwlock.c (pthread_rwlockattr_init) [PTHREAD_RWLOCK_BAD_WAITQUEUE]: New function. (pthread_rwlock_init) [PTHREAD_RWLOCK_BAD_WAITQUEUE]: New function. * m4/pthread-rwlock.m4 (gl_PTHREAD_RWLOCK): Check for reasonable pthread_rwlock wait queue handling. Set REPLACE_PTHREAD_RWLOCK_INIT and REPLACE_PTHREAD_RWLOCKATTR_INIT and define PTHREAD_RWLOCK_BAD_WAITQUEUE if not. * modules/pthread-rwlock (configure.ac): Update GL_COND_OBJ_PTHREAD_RWLOCK condition. * doc/posix-functions/pthread_rwlock_rdlock.texi: Mark the glibc problem as fixed. * doc/posix-functions/pthread_rwlock_tryrdlock.texi: Likewise. * doc/posix-functions/pthread_rwlock_timedrdlock.texi: Likewise. From f7576a2e4bc63fc0b15801a82abe865304ca Mon Sep 17 00:00:00 2001 From: Bruno Haible Date: Wed, 7 Aug 2024 19:53:37 +0200 Subject: [PATCH 1/2] pthread-rwlock: Fix default wait queue behaviour on glibc/Linux. * lib/pthread.in.h: If REPLACE_PTHREAD_RWLOCK_INIT is 1 but REPLACE_PTHREAD_RWLOCK_DESTROY is 0, override PTHREAD_RWLOCK_INITIALIZER. * lib/pthread-rwlock.c (pthread_rwlockattr_init) [PTHREAD_RWLOCK_BAD_WAITQUEUE]: New function. (pthread_rwlock_init) [PTHREAD_RWLOCK_BAD_WAITQUEUE]: New function. * m4/pthread-rwlock.m4 (gl_PTHREAD_RWLOCK): Check for reasonable pthread_rwlock wait queue handling. Set REPLACE_PTHREAD_RWLOCK_INIT and REPLACE_PTHREAD_RWLOCKATTR_INIT and define PTHREAD_RWLOCK_BAD_WAITQUEUE if not. * modules/pthread-rwlock (configure.ac): Update GL_COND_OBJ_PTHREAD_RWLOCK condition. * doc/posix-functions/pthread_rwlock_rdlock.texi: Mark the glibc problem as fixed. * doc/posix-functions/pthread_rwlock_tryrdlock.texi: Likewise. * doc/posix-functions/pthread_rwlock_timedrdlock.texi: Likewise. --- ChangeLog | 20 + .../pthread_rwlock_rdlock.texi| 12 +- .../pthread_rwlock_timedrdlock.texi | 10 +- .../pthread_rwlock_tryrdlock.texi | 10 +- lib/pthread-rwlock.c | 64 +++- lib/pthread.in.h | 9 +- m4/pthread-rwlock.m4 | 358 +- modules/pthread-rwlock| 2 +- 8 files changed, 463 insertions(+), 22 deletions(-) diff --git a/ChangeLog b/ChangeLog index df710ddfa6..64e680886a 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,3 +1,23 @@ +2024-08-07 Bruno Haible + + pthread-rwlock: Fix default wait queue behaviour on glibc/Linux. + * lib/pthread.in.h: If REPLACE_PTHREAD_RWLOCK_INIT is 1 but + REPLACE_PTHREAD_RWLOCK_DESTROY is 0, override + PTHREAD_RWLOCK_INITIALIZER. + * lib/pthread-rwlock.c + (pthread_rwlockattr_init) [PTHREAD_RWLOCK_BAD_WAITQUEUE]: New function. + (pthread_rwlock_init) [PTHREAD_RWLOCK_BAD_WAITQUEUE]: New function. + * m4/pthread-rwlock.m4 (gl_PTHREAD_RWLOCK): Check for reasonable + pthread_rwlock wait queue handling. Set REPLACE_PTHREAD_RWLOCK_INIT and + REPLACE_PTHREAD_RWLOCKATTR_INIT and define PTHREAD_RWLOCK_BAD_WAITQUEUE + if not. + * modules/pthread-rwlock (configure.ac): Update + GL_COND_OBJ_PTHREAD_RWLOCK condition. + * doc/posix-functions/pthread_rwlock_rdlock.texi: Mark the glibc problem + as fixed. + * doc/posix-functions/pthread_rwlock_tryrdlock.texi: Likewise. + * doc/posix-functions/pthread_rwlock_timedrdlock.texi: Likewise. + 2024-08-07 Paul Eggert errno-tests: port to GNU/Hurd diff --git a/doc/posix-functions/pthread_rwlock_rdlock.texi b/doc/posix-functions/pthread_rwlock_rdlock.texi index 6cbeabddce..ad07611917 100644 --- a/doc/posix-functions/pthread_rwlock_rdlock.texi +++ b/doc/posix-functions/pthread_rwlock_rdlock.texi @@ -13,17 +13,15 @@ Minix 3.1.8, mingw, MSVC 14, Android 4.3. But the provided replacement is just a dummy on some of these platforms: Minix 3.1.8. -@end itemize - -Portability problems not fixed by Gnulib: -@itemize @item This function prefers readers to writers (meaning, when this function is called on an rwlock that is alr
Re: TODO lists regarding POSIX:2024 work
I wrote: > There are many changes in POSIX:2024 that can have an impact on Gnulib; Some (biased) summary of the changes can be found here: https://sortix.org/blog/posix-2024/
Avoid compiler warnings in some configure tests
In some config.log I was seeing this warning:
configure:24322: checking for pthread_mutex_timedlock
configure:24346: gcc -o conftest -g -O2 -I/inst-x86_64-64/include -Wall
-L/inst-x86_64-64/lib conftest.c >&5
conftest.c: In function 'main':
conftest.c:149:25: warning: argument 2 null where non-null expected [-Wnonnull]
149 | return pthread_mutex_timedlock (&lock, (struct
timespec *) 0);
| ^~~
This patch fixes it.
2024-08-07 Bruno Haible
Avoid compiler warnings in some configure tests.
* m4/pthread-rwlock.m4 (gl_PTHREAD_RWLOCK): Allocate more room for local
array 'name'.
* m4/pthread_mutex_timedlock.m4 (gl_FUNC_PTHREAD_MUTEX_TIMEDLOCK): Don't
pass a NULL pointer to pthread_mutex_timedlock.
diff --git a/m4/pthread-rwlock.m4 b/m4/pthread-rwlock.m4
index 51d84419c7..cbd08790e1 100644
--- a/m4/pthread-rwlock.m4
+++ b/m4/pthread-rwlock.m4
@@ -1,5 +1,5 @@
# pthread-rwlock.m4
-# serial 5
+# serial 6
dnl Copyright (C) 2019-2024 Free Software Foundation, Inc.
dnl This file is free software; the Free Software Foundation
dnl gives unlimited permission to copy and/or distribute it,
@@ -249,7 +249,7 @@ AC_DEFUN([gl_PTHREAD_RWLOCK]
char **names = (char **) malloc (n * sizeof (char *));
for (size_t i = 0; i < n; i++)
{
- char name[10];
+ char name[12];
sprintf (name, "%c%u", rw_string[i], (unsigned int) (i+1));
names[i] = strdup (name);
}
diff --git a/m4/pthread_mutex_timedlock.m4 b/m4/pthread_mutex_timedlock.m4
index 46b59d793c..9b175d5474 100644
--- a/m4/pthread_mutex_timedlock.m4
+++ b/m4/pthread_mutex_timedlock.m4
@@ -1,5 +1,5 @@
# pthread_mutex_timedlock.m4
-# serial 5
+# serial 6
dnl Copyright (C) 2019-2024 Free Software Foundation, Inc.
dnl This file is free software; the Free Software Foundation
dnl gives unlimited permission to copy and/or distribute it,
@@ -28,7 +28,8 @@ AC_DEFUN([gl_FUNC_PTHREAD_MUTEX_TIMEDLOCK]
#include
]],
[[pthread_mutex_t lock;
- return pthread_mutex_timedlock (&lock, (struct timespec *) 0);
+ struct timespec ts = { 0 };
+ return pthread_mutex_timedlock (&lock, &ts);
]])
],
[gl_cv_func_pthread_mutex_timedlock_in_LIBMULTITHREAD=yes],
doc: Update for OpenBSD 7.5
This patch updates the doc regarding OpenBSD 7.5. This version now has - the header file , - the functions mbrtoc16, mbrtoc32, c16rtomb, c32rtomb, waitid. 2024-08-07 Bruno Haible doc: Update for OpenBSD 7.5. * doc/posix-headers/*.texi: Update. * doc/glibc-headers/*.texi: Likewise. * doc/posix-functions/*.texi: Likewise. * doc/pastposix-functions/*.texi: Likewise. * doc/glibc-functions/*.texi: Likewise. https://git.savannah.gnu.org/gitweb/?p=gnulib.git;a=commitdiff;h=496a7aee40885a2e7bbb8863f525738a0d6095cc
Re: math: Fix INFINITY and NAN on FreeBSD and AIX
And on mingw 5.0, I see this test failure:
FAIL: test-math
===
../../gltests/test-math.c:103: assertion 'sizeof (INFINITY) == sizeof (float)'
failed
FAIL test-math.exe (exit status: 3)
And once this is fixed, likewise for NAN.
This patch fixes it.
2024-08-07 Bruno Haible
math: Fix INFINITY and NAN on mingw.
* doc/posix-headers/math.texi: Mention this mingw bug.
* lib/math.in.h (INFINITY, NAN): Replace also on mingw.
diff --git a/doc/posix-headers/math.texi b/doc/posix-headers/math.texi
index 6f07b1e58c..fb72b94a08 100644
--- a/doc/posix-headers/math.texi
+++ b/doc/posix-headers/math.texi
@@ -55,7 +55,7 @@
@item
@code{INFINITY} and @code{NAN} are of type @code{double} instead of
@code{float}
on some platforms:
-FreeBSD 7.1.
+FreeBSD 7.1, mingw 5.0.
@item
The macros @code{NAN}, @code{HUGE_VALL}, and @code{INFINITY} are not
diff --git a/lib/math.in.h b/lib/math.in.h
index 7b1302d883..2fcba31c76 100644
--- a/lib/math.in.h
+++ b/lib/math.in.h
@@ -161,7 +161,7 @@ static void (*_gl_math_fix_itold) (long double *, int) =
_Qp_itoq;
/* Ensure that INFINITY is a constant expression, of type 'float'. */
-#if !defined INFINITY || (defined __FreeBSD__ && __FreeBSD__ < 8) || defined
_AIX
+#if !defined INFINITY || (defined __FreeBSD__ && __FreeBSD__ < 8) || defined
_AIX || defined __MINGW32__
# undef INFINITY
# if defined __GNUC__ || defined __clang__
# define INFINITY (__builtin_inff ())
@@ -176,7 +176,7 @@ static void (*_gl_math_fix_itold) (long double *, int) =
_Qp_itoq;
it on platforms like Solaris 10, where NAN is present but defined
as a function pointer rather than a floating point constant.
Also ensure that it is a constant expression, of type 'float'. */
-#if !defined NAN || @REPLACE_NAN@ || (defined __FreeBSD__ && __FreeBSD__ < 8)
|| defined _AIX
+#if !defined NAN || @REPLACE_NAN@ || (defined __FreeBSD__ && __FreeBSD__ < 8)
|| defined _AIX || defined __MINGW32__
# if !GNULIB_defined_NAN
# undef NAN
/* The Compaq (ex-DEC) C 6.4 compiler and the Microsoft MSVC 9 compiler
