Re: New warnings on emacs-26 branch with gcc 8.2.0

[Andy Moreton]

On Sat 11 Aug 2018, Eli Zaretskii wrote:

>> From: Andy Moreton <address@hidden>
>> Date: Sat, 11 Aug 2018 16:02:03 +0100
>> 
>> On Sat 11 Aug 2018, Eli Zaretskii wrote:
>> 
>> >> From: Andy Moreton <address@hidden>
>> >> Date: Sat, 11 Aug 2018 11:41:20 +0100
>> >> 
>> >> > OK, I've now done so.  Andrew, please see if this fixes the original
>> >> > problem with this warning.
>> >> 
>> >> Yes, that ha removed that batch of warnings.
>> >
>> > Thanks for testing.
>> >
>> >> There are still other warnings: one from -Wformat-overflow and 78 from
>> >> -Wcast-function-type.
>> >
>> > Can you show the warnings from -Wcast-function-type?
>> 
>> Here are the warnings from commit ec6f588940e5, built with gcc 8.2.0.
>> Mostly this seems to be GetProcAddress results, where it complains that
>> FARPROC and the desired fuinction type don't match. The gcc manual says
>> that casting via "void (*)(void)" can be used to pacify the warning.
>> 
>> 
>>   CC       dynlib.o
>> C:/emacs/git/emacs/emacs-26/src/dynlib.c: In function 'dynlib_addr':
>> C:/emacs/git/emacs/emacs-26/src/dynlib.c:160:6: warning: cast between 
>> incompatible function types from 'FARPROC' {aka 'long long int (*)()'} to 
>> 'BOOL (*)(DWORD,  const CHAR *, struct HINSTANCE__ **)' {aka 'int (*)(long 
>> unsigned int,  const char *, struct HINSTANCE__ **)'} [-Wcast-function-type]
>>       (GetModuleHandleExA_Proc) GetProcAddress (hm_kernel32,
>>       ^
>
> Does it help to take the GetProcAddress call in parentheses, like
> this:
>
>         s_pfn_Get_Module_HandleExA =
>           (GetModuleHandleExA_Proc) (GetProcAddress (hm_kernel32,
>                                                      "GetModuleHandleExA"));

Doesn't help - gcc still warns.

> If this doesn't help, what about removing the cast entirely?

Also still warns. This does work:

          s_pfn_Get_Module_HandleExA =
            (GetModuleHandleExA_Proc) (void (*)(void))
            GetProcAddress (hm_kernel32, "GetModuleHandleExA");

This is sliughtly less ugly and also pacifies the warning:

#define FN_PTR_CAST(fnptrtype, fnptr)                   \
          ((fnptrtype) (void (*)(void)) (fnptr))

          s_pfn_Get_Module_HandleExA =
            FN_PTR_CAST(GetModuleHandleExA_Proc,
                        GetProcAddress (hm_kernel32, "GetModuleHandleExA"));


> In any case, I think it's a GCC bug: it thinks we are type-casting the
> function being called, which is GetProcAddress, whereas what we really
> want to do is cast the _value_ the function returns.

I disagree. GetProcAddress returns FARPROC, and GetModuleHandleExA_Proc
has a different signature. It is valid to warn about this.

    AndyM