Re: Add forward declarations

On 10/01/2018 05:37 PM, Christophe Lyon wrote:
> Hi,
>
> While building newlib for Aarch64, I noticed several warnings because
> of missing prototypes. I am not familiar enough to know why the same
> warnings do not appear when building for Arm.
>
> This patch adds the missing prototypes, tested by rebuilding for
> Aarch64 and Arm, it removed the warnings and didn't generate any
> error.
>
> OK?
>
> Christophe
A primary reason for prototypes is that they are for checking, and fully-proper 
checking is a single header file that checks both the source providing the 
function as well as places that call the function.  That is, adding prototypes 
in the function definition file purely for the purposes of avoiding missing 
prototype warnings is missing the real usefulness behind the warning and giving 
a false sense of security.  (You'd be better off just taking 
-Wmissing-prototypes out of your options to avoid those particular warnings.)  
The prototypes need to be added to a header file, not the functions' own source.

Put another way, as a general goal I'd think we ought to be fixing the cause of 
a warning to remove the warning, rather than doing things to "hide" or "mask" 
the warning (which could be considered as kludges).  Putting protoypes into the 
function source file is clearly masking the deficiency that there is not a 
header file with prototypes, rather than fixing the actual deficiency.  While 
there certainly are times when addressing the root cause of a warning is not 
possible, this does not seem to be one where it is appropriate. Probably 
something like sys/_syscall.h (or some such name) should be created and used.  
We should be cleaning up bad practice when it exists, not furthering it.  So, 
definitely a good thing to get rid of the warnings, but there is a much better 
approach to take in this case.

Corrina or Jeff, any suggestions for a good place for syscall prototypes?  (My 
first thought was sys/syscalls.h, but the Linux syscall(2) manpage mentions that 
related to syscall() and "For SYS_xxx definitions."  Perhaps it would be OK, but 
perhaps not, too, which is why I ended up suggesting _syscall.h.)  Does someone 
know about a precedent in BSD or the like?

Craig