Sync up error.c with gnulib

  hi,

The following patch should sync up the glibc version of error.c with
gnulib.  There are a few changes that are pending inclusion into
gnulib; I have posted them to bug-gnulib but the email is probably
held in moderation.

Summary of changes:

- Use of !_LIBC instead of HAVE_CONFIG_H
- Code changes in [!_LIBC] that don't affect us
- Minor formatting changes
- Use __builtin_expect in shared code
- Define some macros in [_LIBC] that are used in gnulib but never
  defined in glibc (change proposed in gnulib)
- Flip macro check for STRERROR_R_CHAR_P so that it does not throw a
  warning (change proposed in gnulib)

The only change in generated code on x86_64 is due to change in line
numbers for error_at_line.  OK to commit?

Siddhesh

	Sync up with gnulib.
	* misc/error.c: Use !_LIBC instead of HAVE_CONFIG_H.
	[!_LIBC && ENABLE_NLS]: Include gettext.h.
	[_LIBC]: Define USE_UNLOCKED_IO, _GL_ATTRIBUTE_FORMAT_PRINTF
	and _GL_ARG_NONNULL.
	[USE_UNLOCKED_IO]: Include unlocked-io.h.
	[!_LIBC]: Include code for Windows and Cygwin.
	[!_LIBC && !HAVE_DECL_STRERROR_R && !STRERROR_R_CHAR_P]:
	Include prototype for int strerror_r.
	[!_LIBC] (is_open): New function.
	(flush_stdout): New function.
	(print_errno_message): Use it.
	(error): Likewise.
	(error_at_line): Likewise.
	(error_tail) Add function attribute macros.  Use
	__builtin_expect.
  

On 9 July 2014 11:33, Siddhesh Poyarekar <siddhesh@redhat.com> wrote:
> hi,
>
> The following patch should sync up the glibc version of error.c with
> gnulib.  There are a few changes that are pending inclusion into
> gnulib; I have posted them to bug-gnulib but the email is probably
> held in moderation.
>
> Summary of changes:
>
> - Use of !_LIBC instead of HAVE_CONFIG_H
> - Code changes in [!_LIBC] that don't affect us
> - Minor formatting changes
> - Use __builtin_expect in shared code
> - Define some macros in [_LIBC] that are used in gnulib but never
>   defined in glibc (change proposed in gnulib)
> - Flip macro check for STRERROR_R_CHAR_P so that it does not throw a
>   warning (change proposed in gnulib)

AFAICT STRERROR_R_CHAR_P is not defined in gnulib any more so it
should probably be removed altogether.

> The only change in generated code on x86_64 is due to change in line
> numbers for error_at_line.  OK to commit?
>
> Siddhesh
>
>         Sync up with gnulib.
>         * misc/error.c: Use !_LIBC instead of HAVE_CONFIG_H.
>         [!_LIBC && ENABLE_NLS]: Include gettext.h.
>         [_LIBC]: Define USE_UNLOCKED_IO, _GL_ATTRIBUTE_FORMAT_PRINTF
>         and _GL_ARG_NONNULL.
>         [USE_UNLOCKED_IO]: Include unlocked-io.h.
>         [!_LIBC]: Include code for Windows and Cygwin.
>         [!_LIBC && !HAVE_DECL_STRERROR_R && !STRERROR_R_CHAR_P]:
>         Include prototype for int strerror_r.
>         [!_LIBC] (is_open): New function.
>         (flush_stdout): New function.
>         (print_errno_message): Use it.
>         (error): Likewise.
>         (error_at_line): Likewise.
>         (error_tail) Add function attribute macros.  Use
>         __builtin_expect.
>
> diff --git a/misc/error.c b/misc/error.c
> index 63bd309..ee0cf86 100644
> --- a/misc/error.c
> +++ b/misc/error.c
> @@ -18,24 +18,36 @@
>
>  /* Written by David MacKenzie <djm@gnu.ai.mit.edu>.  */
>
> -#ifdef HAVE_CONFIG_H
> +#if !_LIBC
>  # include <config.h>
>  #endif
>
> +#include "error.h"
> +
>  #include <stdarg.h>
>  #include <stdio.h>
>  #include <stdlib.h>
>  #include <string.h>
>
> +#if !_LIBC && ENABLE_NLS
> +# include "gettext.h"
> +# define _(msgid) gettext (msgid)
> +#endif
> +
>  #ifdef _LIBC
>  # include <libintl.h>
>  # include <stdbool.h>
>  # include <stdint.h>
>  # include <wchar.h>
>  # define mbsrtowcs __mbsrtowcs
> +# define USE_UNLOCKED_IO 0
> +# define _GL_ATTRIBUTE_FORMAT_PRINTF(a, b)
> +# define _GL_ARG_NONNULL(a)
>  #endif
>
> -#include "error.h"
> +#if USE_UNLOCKED_IO
> +# include "unlocked-io.h"
> +#endif
>
>  #ifndef _
>  # define _(String) String
> @@ -46,7 +58,7 @@
>     function without parameters instead.  */
>  void (*error_print_progname) (void);
>
> -/* This variable is incremented each time `error' is called.  */
> +/* This variable is incremented each time 'error' is called.  */
>  unsigned int error_message_count;
>
>  #ifdef _LIBC
> @@ -57,7 +69,7 @@ unsigned int error_message_count;
>  # include <limits.h>
>  # include <libio/libioP.h>
>
> -/* In GNU libc we want do not want to use the common name `error' directly.
> +/* In GNU libc we want do not want to use the common name 'error' directly.
>     Instead make it a weak alias.  */
>  extern void __error (int status, int errnum, const char *message, ...)
>       __attribute__ ((__format__ (__printf__, 3, 4)));
> @@ -77,11 +89,29 @@ extern void __error_at_line (int status, int errnum, const char *file_name,
>
>  #else /* not _LIBC */
>
> -# if !HAVE_DECL_STRERROR_R && STRERROR_R_CHAR_P
> +# include <fcntl.h>
> +# include <unistd.h>
> +
> +# if (defined _WIN32 || defined __WIN32__) && ! defined __CYGWIN__
> +/* Get declarations of the native Windows API functions.  */
> +#  define WIN32_LEAN_AND_MEAN
> +#  include <windows.h>
> +/* Get _get_osfhandle.  */
> +#  include "msvc-nothrow.h"
> +# endif
> +
> +/* The gnulib override of fcntl is not needed in this file.  */
> +# undef fcntl
> +
> +# if !HAVE_DECL_STRERROR_R
>  #  ifndef HAVE_DECL_STRERROR_R
>  "this configure-time declaration test was not run"
>  #  endif
> +#  if STRERROR_R_CHAR_P
>  char *strerror_r ();
> +#  else
> +int strerror_r ();
> +#  endif
>  # endif
>
>  /* The calling program should define program_name and set it to the
> @@ -93,6 +123,51 @@ extern char *program_name;
>  # endif        /* HAVE_STRERROR_R || defined strerror_r */
>  #endif /* not _LIBC */
>
> +#if !_LIBC
> +/* Return non-zero if FD is open.  */
> +static int
> +is_open (int fd)
> +{
> +# if (defined _WIN32 || defined __WIN32__) && ! defined __CYGWIN__
> +  /* On native Windows: The initial state of unassigned standard file
> +     descriptors is that they are open but point to an INVALID_HANDLE_VALUE.
> +     There is no fcntl, and the gnulib replacement fcntl does not support
> +     F_GETFL.  */
> +  return (HANDLE) _get_osfhandle (fd) != INVALID_HANDLE_VALUE;
> +# else
> +#  ifndef F_GETFL
> +#   error Please port fcntl to your platform
> +#  endif
> +  return 0 <= fcntl (fd, F_GETFL);
> +# endif
> +}
> +#endif
> +
> +static void
> +flush_stdout (void)
> +{
> +#if !_LIBC
> +  int stdout_fd;
> +
> +# if GNULIB_FREOPEN_SAFER
> +  /* Use of gnulib's freopen-safer module normally ensures that
> +       fileno (stdout) == 1
> +     whenever stdout is open.  */
> +  stdout_fd = STDOUT_FILENO;
> +# else
> +  /* POSIX states that fileno (stdout) after fclose is unspecified.  But in
> +     practice it is not a problem, because stdout is statically allocated and
> +     the fd of a FILE stream is stored as a field in its allocated memory.  */
> +  stdout_fd = fileno (stdout);
> +# endif
> +  /* POSIX states that fflush (stdout) after fclose is unspecified; it
> +     is safe in glibc, but not on all other platforms.  fflush (NULL)
> +     is always defined, but too draconian.  */
> +  if (0 <= stdout_fd && is_open (stdout_fd))
> +#endif
> +    fflush (stdout);
> +}
> +
>  static void
>  print_errno_message (int errnum)
>  {
> @@ -100,7 +175,7 @@ print_errno_message (int errnum)
>
>  #if defined HAVE_STRERROR_R || _LIBC
>    char errbuf[1024];
> -# if STRERROR_R_CHAR_P || _LIBC
> +# if _LIBC || STRERROR_R_CHAR_P
>    s = __strerror_r (errnum, errbuf, sizeof errbuf);
>  # else
>    if (__strerror_r (errnum, errbuf, sizeof errbuf) == 0)
> @@ -124,7 +199,7 @@ print_errno_message (int errnum)
>  #endif
>  }
>
> -static void
> +static void _GL_ATTRIBUTE_FORMAT_PRINTF (3, 0) _GL_ARG_NONNULL ((3))
>  error_tail (int status, int errnum, const char *message, va_list args)
>  {
>  #if _LIBC
> @@ -165,7 +240,7 @@ error_tail (int status, int errnum, const char *message, va_list args)
>           if (res != len)
>             break;
>
> -         if (__glibc_unlikely (len >= SIZE_MAX / sizeof (wchar_t) / 2))
> +         if (__builtin_expect (len >= SIZE_MAX / sizeof (wchar_t) / 2, 0))
>             {
>               /* This really should not happen if everything is fine.  */
>               res = (size_t) -1;
> @@ -227,7 +302,7 @@ error (int status, int errnum, const char *message, ...)
>                    0);
>  #endif
>
> -  fflush (stdout);
> +  flush_stdout ();
>  #ifdef _LIBC
>    _IO_flockfile (stderr);
>  #endif
> @@ -273,6 +348,7 @@ error_at_line (int status, int errnum, const char *file_name,
>               || (old_file_name != NULL
>                   && file_name != NULL
>                   && strcmp (old_file_name, file_name) == 0)))
> +
>         /* Simply return and print nothing.  */
>         return;
>
> @@ -288,7 +364,7 @@ error_at_line (int status, int errnum, const char *file_name,
>                    0);
>  #endif
>
> -  fflush (stdout);
> +  flush_stdout ();
>  #ifdef _LIBC
>    _IO_flockfile (stderr);
>  #endif
  

On Wed, Jul 09, 2014 at 11:41:09AM +0100, Will Newton wrote:
> On 9 July 2014 11:33, Siddhesh Poyarekar <siddhesh@redhat.com> wrote:
> > hi,
> >
> > The following patch should sync up the glibc version of error.c with
> > gnulib.  There are a few changes that are pending inclusion into
> > gnulib; I have posted them to bug-gnulib but the email is probably
> > held in moderation.
> >
> > Summary of changes:
> >
> > - Use of !_LIBC instead of HAVE_CONFIG_H
> > - Code changes in [!_LIBC] that don't affect us
> > - Minor formatting changes
> > - Use __builtin_expect in shared code
> > - Define some macros in [_LIBC] that are used in gnulib but never
> >   defined in glibc (change proposed in gnulib)
> > - Flip macro check for STRERROR_R_CHAR_P so that it does not throw a
> >   warning (change proposed in gnulib)
> 
> AFAICT STRERROR_R_CHAR_P is not defined in gnulib any more so it
> should probably be removed altogether.

Oh yeah, thanks for pointing that out.  I'll update my patches.

Siddhesh
  

That all seems fine if it gets us to verbatim synch with gnulib.