Re: stdint and mingw

[Bruno Haible]

Eric Blake wrote:
> mingw has wint_t as unsigned int, but defined WINT_MAX as 0xffff to match 
> WCHAR_MAX.  Without this patch, stdint.m4 claims mingw's header is C99 
> compliant, then crashes the m4 testsuite (now that m4 uses the gnulib unit 
> tests) when encountering the invalid WINT_MAX.  With this patch, 
> test-stdint.c 
> passes once again on mingw.  Is this patch okay to apply?

Well, I prefer to avoid signed integer overflow, as in WINT_MAX + 1, since
we know by now the trouble that it can cause. Can you try this patch instead?

*** m4/stdint.m4        28 Apr 2007 18:05:58 -0000      1.29
--- m4/stdint.m4        30 May 2007 00:31:37 -0000
***************
*** 147,153 ****
    int check_SIG_ATOMIC: SIG_ATOMIC_MIN <= (sig_atomic_t) 0 && (sig_atomic_t) 
0 < SIG_ATOMIC_MAX ? 1 : -1;
    int check_SIZE: (size_t) 0 < SIZE_MAX ? 1 : -1;
    int check_WCHAR: WCHAR_MIN <= (wchar_t) 0 && (wchar_t) 0 < WCHAR_MAX ? 1 : 
-1;
!   int check_WINT: WINT_MIN <= (wint_t) 0 && (wint_t) 0 < WINT_MAX ? 1 : -1;
  
    /* Detect bugs in glibc 2.4 and Solaris 10 stdint.h, among others.  */
    int check_UINT8_C:
--- 147,155 ----
    int check_SIG_ATOMIC: SIG_ATOMIC_MIN <= (sig_atomic_t) 0 && (sig_atomic_t) 
0 < SIG_ATOMIC_MAX ? 1 : -1;
    int check_SIZE: (size_t) 0 < SIZE_MAX ? 1 : -1;
    int check_WCHAR: WCHAR_MIN <= (wchar_t) 0 && (wchar_t) 0 < WCHAR_MAX ? 1 : 
-1;
!   int check_WINT: WINT_MIN <= (wint_t) 0 && (wint_t) 0 < WINT_MAX
!                   && (WINT_MIN < (wint_t) 0 || (wint_t) -1 == (wint_t) 
WINT_MAX)
!                   ? 1 : -1;
  
    /* Detect bugs in glibc 2.4 and Solaris 10 stdint.h, among others.  */
    int check_UINT8_C:

(The copy of mingw that I have doesn't have the problem that you encountered,
therefore I cannot test it.)

> Other holdups in releasing m4 1.4.10:
> - mingw's atof/strtod does not parse infinity or nan, but gnulib's strtod 
> module does not yet provide this capability either.  I hope to work on this 
> soon, if no one beats me to it.
> 
> - gnulib's unistd module does not provide the types gid_t, uid_t, or 
> useconds_t, all of which are undefined in mingw, but Bruno's recent testsuite 
> additions means that test-unistd.c crashes on mingw in trying to use these 
> types.

Two other ones:

  - The recent #include_next patch is broken.

  - test-stdint fails on FreeBSD 6.0 / ia64. It fails in the lines
      verify (TYPE_MINIMUM (sig_atomic_t) == SIG_ATOMIC_MIN);
      verify (TYPE_MAXIMUM (sig_atomic_t) == SIG_ATOMIC_MAX);
    because sig_atomic_t = long but SIG_ATOMIC_MIN and SIG_ATOMIC_MAX are
    defined (in /usr/include/stdint.h) to 32-bit values.

Where can we have a snapshot for testing m4-1.4.10-pre?

Bruno