Re: msvc-inval: more options

[Jim Meyering]

Bruno Haible wrote:
>> I noticed that the latest from gnulib's tests gets some new warnings
>> when compiled on Fedora 15:
>>
>>     test-fread.c: In function 'main':
>>     test-fread.c:43:3: error: implicit declaration of function
>> gl_msvc_inval_ensure_handler'
>> [-Werror=implicit-function-declaration]
>
> Oops, yes, I made a mistake. Fixed below.
>
>> Wouldn't it be better to make the commonly-used-in-application-code
>>       gl_msvc_inval_ensure_handler ();
>> a no-op when possible, so that applications don't need to guard
>> uses of it with #if directives like those above?
>
> This would make sense for the cases
>   !HAVE_MSVC_INVALID_PARAMETER_HANDLER
>   HAVE_MSVC_INVALID_PARAMETER_HANDLER && MSVC_INVALID_PARAMETER_HANDLING == 
> DEFAULT_HANDLING
> But in the case
>   HAVE_MSVC_INVALID_PARAMETER_HANDLER && MSVC_INVALID_PARAMETER_HANDLING == 
> HAIRY_LIBRARY_HANDLING
> the handler exists but has, on its own, not the desired effect. Therefore
> I think it's better to be clear about which conditions do benefit from the
> code.
>
> 2011-09-28  Bruno Haible  <address@hidden>
>
>       fgetc, fputc, fread, fwrite tests: Fix link error.
>       * tests/test-fgetc.c (main): Don't invoke gl_msvc_inval_ensure_handler
>       on non-MSVC platforms.
>       * tests/test-fputc.c (main): Likewise.
>       * tests/test-fread.c (main): Likewise.
>       * tests/test-fwrite.c (main): Likewise.
>       Reported by Jim Meyering.

Hi Bruno,

Thank you for the quick fix.
With that, coreutils-updated-to-use-latest-from-gnulib
now passes its build/check tests on Fedora 15 and rawhide.

> --- tests/test-fgetc.c.orig   Wed Sep 28 10:22:49 2011
> +++ tests/test-fgetc.c        Wed Sep 28 10:22:30 2011
> @@ -35,9 +35,10 @@
>  {
>    const char *filename = "test-fgetc.txt";
>
> -  /* We don't have an fread() function that installs an invalid parameter
> +  /* We don't have an fgetc() function that installs an invalid parameter
>       handler so far.  So install that handler here, explicitly.  */
> -#if MSVC_INVALID_PARAMETER_HANDLING == DEFAULT_HANDLING
> +#if HAVE_MSVC_INVALID_PARAMETER_HANDLER \
> +    && MSVC_INVALID_PARAMETER_HANDLING == DEFAULT_HANDLING
>    gl_msvc_inval_ensure_handler ();
>  #endif

That is a detail specific to MSVC.  Why impose the readability and
maintenance hassle of requiring such an expression on all callers?

Wouldn't you prefer to make any MSVC-related code as inconspicuous
as possible?  Not to mention avoiding the duplication of an #if
expression like that...