Re: [patch rfc] -Wmissing-prototypes round #3

[Kevin Buettner <kevinb at redhat dot com>]

On Jun 11,  3:24pm, Andrew Cagney wrote:

> > Andrew Cagney <ac131313@redhat.com> writes:
> > 
> > 
> >> - move function declarations to CPU-tdep.h, from config/CPU/tm-CPU.h
> >> Ex, m68k's delta68 functions previously in tm-delta.h.
> >> This ensures that the function declaration is always visible, and that a 
> >> global function in CPU-tdep.c has its declaration in the corresponding 
> >> CPU-tdep.h file.  The config/CPU/tm-CPU.h files were also updated to 
> >> include the corresponding "CPU-tdep.h".
> > 
> > 
> > I'm not sure whether this part of your patch is a good idea.  Quite a
> > number of function declarations in config/CPU/tm-CPU.h are for things
> > that haven't been multi-arched yet.  Keeping the function declaration
> > next to the #define that uses them, makes it easier to remove them
> > once they're no longer needed.
> 
> I thought someone might comment, that's why I tried to make what was 
> happening very clear :-)
> 
> It's a real trade off.
> 
> -- delay -Wmissing-prototypes until multi-arch is finished
> 
> -- move the declarations so that they are always visible
> 
> -- drop -Werror for certain architectures
> 
> Given the problems that -Wmissing-prototypes has thrown up (extern in 
> .c, forgetting to include .h), I'd prefer to see it enabled.
> 
> Also note the number of *-tdep.c functions I make static.  It's pretty 
> clear that people deleting the #define/declaration consistently forget 
> to update the corresponding function definition.
> 
> I think a better way of handling these stray global declarations would 
> be to use some sort of real code analysis tool (which the ari isn't :-).

I happen to agree with Mark on this, but I'm also sympathetic to Andrew's
desire to get -Wmissing-prototypes going.

As a compromise, I suggest moving the function declarations as Andrew
proposes, but also putting a pointer (comment) to the (new) location
of the declaration along with a reminder to remove it and make it
static when the function is question is finally multiarched.

Kevin