[PATCH 1/7] spawn: Use special invocation for <spawn.h> on OS/2 kLIBC

[PATCH 1/7] spawn: Use special invocation for <spawn.h> on OS/2 kLIBC

[KO Myung-Hun]

On OS/2 kLIBC, <spawn.h> includes <signal.h>. Then <signal.h> ->
<pthread.h> -> <sched.h> -> <spawn.h> are included by GNULIB.
In this situation, `struct sched_param' is not yet defined.

* lib/spawn.in.h: Use special invocation to include <spawn.h> properly
on OS/2 kLIBC.
---
 lib/spawn.in.h | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)

diff --git a/lib/spawn.in.h b/lib/spawn.in.h
index 326b79e6c..35e78e6cf 100644
--- a/lib/spawn.in.h
+++ b/lib/spawn.in.h
@@ -15,16 +15,33 @@
    You should have received a copy of the GNU General Public License
    along with this program.  If not, see <https://www.gnu.org/licenses/>.  */
 
-#ifndef _@GUARD_PREFIX@_SPAWN_H
-
 #if __GNUC__ >= 3
 @PRAGMA_SYSTEM_HEADER@
 #endif
 @PRAGMA_COLUMNS@
 
+#if defined _GL_ALREADY_INCLUDING_SPAWN_H
+/* Special invocation convention:
+   On OS/2 kLIBC, <spawn.h> includes <signal.h>. Then <signal.h> ->
+   <pthread.h> -> <sched.h> -> <spawn.h> are included by GNULIB.
+   In this situation, struct sched_param is not yet defined.  */
+
+#@INCLUDE_NEXT@ @NEXT_SPAWN_H@
+
+#else
+
+#ifndef _@GUARD_PREFIX@_SPAWN_H
+/* Normal invocation convention.  */
+
 /* The include_next requires a split double-inclusion guard.  */
 #if @HAVE_SPAWN_H@
+
+# define _GL_ALREADY_INCLUDING_SPAWN_H
+
 # @INCLUDE_NEXT@ @NEXT_SPAWN_H@
+
+# define _GL_ALREADY_INCLUDING_SPAWN_H
+
 #endif
 
 #ifndef _@GUARD_PREFIX@_SPAWN_H
@@ -972,3 +989,4 @@ _GL_WARN_ON_USE (posix_spawn_file_actions_addfchdir,
 
 #endif /* _@GUARD_PREFIX@_SPAWN_H */
 #endif /* _@GUARD_PREFIX@_SPAWN_H */
+#endif
-- 
2.22.0

[PATCH 2/7] yield: Fix linkage to threadlib on OS/2 kLIBC

[KO Myung-Hun]

* lib/sched_yield.c (sched_yield) [OS/2 kLIBC]: New.
* modules/yield (Depends-on): Add sched_yield.
---
 lib/sched_yield.c | 13 +++++++++++++
 modules/yield     |  1 +
 2 files changed, 14 insertions(+)

diff --git a/lib/sched_yield.c b/lib/sched_yield.c
index 0a0ff3c9c..880711405 100644
--- a/lib/sched_yield.c
+++ b/lib/sched_yield.c
@@ -34,6 +34,19 @@ sched_yield (void)
   return 0;
 }
 
+#elif defined __KLIBC__
+/* OS/2 kLIBC implementation */
+
+# define INCL_DOS
+# include <os2.h>
+
+int
+sched_yield (void)
+{
+  DosSleep (0);
+  return 0;
+}
+
 #else
 /* Provide a dummy implementation for single-threaded applications.  */
 
diff --git a/modules/yield b/modules/yield
index 5ca862067..4481fde25 100644
--- a/modules/yield
+++ b/modules/yield
@@ -6,6 +6,7 @@ lib/glthread/yield.h
 m4/yield.m4
 
 Depends-on:
+sched_yield
 threadlib
 
 configure.ac:
-- 
2.22.0

[PATCH 3/7] stddef: Fix compilation for max_align_t on OS/2 kLIBC

[KO Myung-Hun]

On OS/2 kLIBC, `max_align_t' is defined only in C++ mode.

* lib/stddef.in.h [OS/2 kLIBC]: Include <cstddef> in C++ mode.
---
 lib/stddef.in.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/lib/stddef.in.h b/lib/stddef.in.h
index 77147e9bc..c415cbb85 100644
--- a/lib/stddef.in.h
+++ b/lib/stddef.in.h
@@ -104,7 +104,8 @@ typedef long max_align_t;
    On MSVC, max_align_t is defined only in C++ mode, after <cstddef> was
    included.  Its definition is good since it has an alignment of 8 (on x86
    and x86_64).  */
-#if defined _MSC_VER && defined __cplusplus
+#if (defined _MSC_VER || (defined __KLIBC__ && !defined __LIBCN__)) \
+  && defined __cplusplus
 # include <cstddef>
 #else
 # if ! (@HAVE_MAX_ALIGN_T@ || defined _GCC_MAX_ALIGN_T)
-- 
2.22.0

[PATCH 4/7] stdlib: putenv() needs a cast on OS/2 kLIBC

[KO Myung-Hun]

On OS/2 kLIBC, the first parameter of putenv () is `const char *string'
not `char *string'.

* lib/stdlib.in.h: Use _GL_CXXALIAS_SYS_CAST() for putenv().
---
 lib/stdlib.in.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/lib/stdlib.in.h b/lib/stdlib.in.h
index 49fc44e14..a0e113665 100644
--- a/lib/stdlib.in.h
+++ b/lib/stdlib.in.h
@@ -711,7 +711,9 @@ _GL_CXXALIAS_RPL (putenv, int, (char *string));
 #  endif
 _GL_CXXALIAS_MDA (putenv, int, (char *string));
 # else
-_GL_CXXALIAS_SYS (putenv, int, (char *string));
+/* Need to cast, because on OS/2 kLIBC, the first parameter is
+                                     const char *string.  */
+_GL_CXXALIAS_SYS_CAST (putenv, int, (char *string));
 # endif
 _GL_CXXALIASWARN (putenv);
 #elif @GNULIB_MDA_PUTENV@
-- 
2.22.0

[PATCH 5/7] zerosize-ptr: Fix compilation on OS/2 kLIBC

[KO Myung-Hun]

OS/2 kLIBC has <sys/mman.h> and mprotect() but not mmap(). As a result,
it's not enough to test with HAVE_SYS_MMAN_H and HAVE_MPROTECT. It's
necessary to test with HAVE_MMAP on OS/2 kLIBC.

* tests/zerosize-ptr.h [OS/2 kLIBC]: Test with HAVE_MMAP, too.
(zerosize_ptr) [OS/2 kLIBC]: Likewise.
---
 tests/zerosize-ptr.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/zerosize-ptr.h b/tests/zerosize-ptr.h
index 45a1729d3..b8e3a9e06 100644
--- a/tests/zerosize-ptr.h
+++ b/tests/zerosize-ptr.h
@@ -37,7 +37,7 @@
    We don't use HAVE_MMAP, because AC_FUNC_MMAP would not define it on HP-UX.
    HAVE_MPROTECT is not enough, because mingw does not have mmap() but has an
    mprotect() function in libgcc.a.  */
-#if HAVE_SYS_MMAN_H && HAVE_MPROTECT
+#if HAVE_SYS_MMAN_H && HAVE_MPROTECT && (!defined __KLIBC__ || HAVE_MMAP)
 # include <fcntl.h>
 # include <unistd.h>
 # include <sys/types.h>
@@ -58,7 +58,7 @@ zerosize_ptr (void)
 {
 /* Use mmap and mprotect when they exist.  Don't test HAVE_MMAP, because it is
    not defined on HP-UX 11 (since it does not support MAP_FIXED).  */
-#if HAVE_SYS_MMAN_H && HAVE_MPROTECT
+#if HAVE_SYS_MMAN_H && HAVE_MPROTECT && (!defined __KLIBC__ || HAVE_MMAP)
 # if HAVE_MAP_ANONYMOUS
   const int flags = MAP_ANONYMOUS | MAP_PRIVATE;
   const int fd = -1;
-- 
2.22.0

[PATCH 6/7] dirent: Redefine dirfd macro as a rpl function always

[KO Myung-Hun]

On OS/2 kLIBC, dirfd is defined as a macro not a function. Because of
this, dirfd is not redefined as rpl function in C++ mode. As a result,
the following error occurs:

-----
./../gllib/dirent.h:689:1: error: 'dirfd' was not declared in this scope; did you mean 'gnulib::dirfd'?
  689 | _GL_CXXALIASWARN (dirfd);
      | ^~~~~~~~~~~~~~~~
./../gllib/dirent.h:669:1: note: 'gnulib::dirfd' declared here
  669 | _GL_CXXALIAS_RPL (dirfd, int, (DIR *));
      | ^~~~~~~~~~~~~~~~
-----

* lib/dirent.in.h (dirfd): Redefine dirfd macro as a rpl function always.
---
 lib/dirent.in.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/lib/dirent.in.h b/lib/dirent.in.h
index 2e2c5119a..14ce5b987 100644
--- a/lib/dirent.in.h
+++ b/lib/dirent.in.h
@@ -154,7 +154,8 @@ _GL_WARN_ON_USE (closedir, "closedir is not portable - "
 /* Return the file descriptor associated with the given directory stream,
    or -1 if none exists.  */
 # if @REPLACE_DIRFD@
-#  if !(defined __cplusplus && defined GNULIB_NAMESPACE)
+#  if !(defined __cplusplus && defined GNULIB_NAMESPACE) || defined dirfd
+    /* If dirfd is a macro, re-define it as a rpl function always.  */
 #   undef dirfd
 #   define dirfd rpl_dirfd
 #  endif
-- 
2.22.0

[PATCH 7/7] spawn-pipe: Fix SIGSEGV on OS/2 kLIBC

[KO Myung-Hun]

* lib/spawn-pipe.c [OS/2 kLIBC]: Check directory correctly if it means
a current directory.
---
 lib/spawn-pipe.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/spawn-pipe.c b/lib/spawn-pipe.c
index 641e9216f..0fe86ffe2 100644
--- a/lib/spawn-pipe.c
+++ b/lib/spawn-pipe.c
@@ -341,7 +341,7 @@ create_pipe (const char *progname,
     }
 
 # else /* __KLIBC__ */
-  if (!(directory == NULL && strcmp (directory, ".") == 0))
+  if (!(directory == NULL || strcmp (directory, ".") == 0))
     {
       /* A directory argument is not supported in this implementation.  */
       saved_errno = EINVAL;
-- 
2.22.0

Re: [PATCH 4/7] stdlib: putenv() needs a cast on OS/2 kLIBC

[Bruno Haible]

Hi KO,

> On OS/2 kLIBC, the first parameter of putenv () is `const char *string'
> not `char *string'.

Can't you fix that in OS/2 kLIBC? Then a workaround in Gnulib would not be
needed.

Bruno

Re: [PATCH 5/7] zerosize-ptr: Fix compilation on OS/2 kLIBC

[Bruno Haible]

Applied with modifications:
> -#if HAVE_SYS_MMAN_H && HAVE_MPROTECT
> +#if HAVE_SYS_MMAN_H && HAVE_MPROTECT && (!defined __KLIBC__ || HAVE_MMAP)

There's a comment two lines above, that explains why HAVE_MMAP is not
the right test here. And since you write "OS/2 kLIBC has <sys/mman.h>
and mprotect() but not mmap()", defined __KLIBC__ implies !HAVE_MMAP.

Bruno

Re: [PATCH 1/7] spawn: Use special invocation for <spawn.h> on OS/2 kLIBC

[Bruno Haible]

Hi KO,

Thanks, I applied patches 1, 2, 3, 5, 6, 7, slightly modified. Some of the
things that you put into the git commit message actually better belong in
the code, as a comment.

Bruno

Re: [PATCH 5/7] zerosize-ptr: Fix compilation on OS/2 kLIBC

[KO Myung-Hun]

Hi/2.

Bruno Haible wrote:
> Applied with modifications:
>> -#if HAVE_SYS_MMAN_H && HAVE_MPROTECT
>> +#if HAVE_SYS_MMAN_H && HAVE_MPROTECT && (!defined __KLIBC__ || HAVE_MMAP)
> 
> There's a comment two lines above, that explains why HAVE_MMAP is not
> the right test here. And since you write "OS/2 kLIBC has <sys/mman.h>
> and mprotect() but not mmap()", defined __KLIBC__ implies !HAVE_MMAP.
> 

I saw that. So I tested HAVE_MMAP only on kLIBC. Anyway I agree that
just defined __KLIBC__ is enough.

Thanks!

-- 
KO Myung-Hun

Using Mozilla SeaMonkey 2.7.2
Under OS/2 Warp 4 for Korean with FixPak #15
In VirtualBox v6.1.10 on Intel Core i7-3615QM 2.30GHz with 8GB RAM

Korean OS/2 User Community : http://www.os2.kr/

Re: [PATCH 4/7] stdlib: putenv() needs a cast on OS/2 kLIBC

[KO Myung-Hun]

Hi/2, Bruno!

Bruno Haible wrote:
> Hi KO,
> 
>> On OS/2 kLIBC, the first parameter of putenv () is `const char *string'
>> not `char *string'.
> 
> Can't you fix that in OS/2 kLIBC? Then a workaround in Gnulib would not be
> needed.
> 

I have no such a right. Instead I'll suggest that to maintainer.
However, I think, it's more proper that Gnulib has a workaround. Because
Gnulib provides such consistent interfaces and functionality.

BTW, do you know the reason why `char *string' not `const char *string'
? I think, `const char *string' is more proper, because putenv() does
not modify the content of `string'.

Thanks as always!

-- 
KO Myung-Hun

Using Mozilla SeaMonkey 2.7.2
Under OS/2 Warp 4 for Korean with FixPak #15
In VirtualBox v6.1.10 on Intel Core i7-3615QM 2.30GHz with 8GB RAM

Korean OS/2 User Community : http://www.os2.kr/

Re: [PATCH 1/7] spawn: Use special invocation for <spawn.h> on OS/2 kLIBC

[KO Myung-Hun]

Hi, Bruno.

Bruno Haible wrote:
> Hi KO,
> 
> Thanks, I applied patches 1, 2, 3, 5, 6, 7, slightly modified. Some of the
> things that you put into the git commit message actually better belong in
> the code, as a comment.
> 

Ok, I'll do so later.

Thanks for your effort!

-- 
KO Myung-Hun

Using Mozilla SeaMonkey 2.7.2
Under OS/2 Warp 4 for Korean with FixPak #15
In VirtualBox v6.1.10 on Intel Core i7-3615QM 2.30GHz with 8GB RAM

Korean OS/2 User Community : http://www.os2.kr/

Re: [PATCH 4/7] stdlib: putenv() needs a cast on OS/2 kLIBC

[KO Myung-Hun]

Hi/2.

Bruno Haible wrote:
> Hi KO,
> 
>> On OS/2 kLIBC, the first parameter of putenv () is `const char *string'
>> not `char *string'.
> 
> Can't you fix that in OS/2 kLIBC? Then a workaround in Gnulib would not be
> needed.
> 

This was rejected by the maintainer of kLIBC. See details here:

http://trac.netlabs.org/libc/ticket/387#comment:2

Would you mind applying this patch ?

-- 
KO Myung-Hun

Using Mozilla SeaMonkey 2.7.2
Under OS/2 Warp 4 for Korean with FixPak #15
In VirtualBox v6.1.10 on Intel Core i7-3615QM 2.30GHz with 8GB RAM

Korean OS/2 User Community : http://www.os2.kr/