Re: Deferencing type-punned pointers, and how to stop gcc giving youa warning

[Chris Sherlock <csherlock optusnet com au>]

I understand... some of the patches I've put in bugzilla are kind of 
ugly, especially the gdk x11 stuff! I was even seriously considering 
making a *macro* that did it for me... but discarded this as somewhat 
stupid. I only want to say one last thing, before I ever hold my piece: 
I'm just a tad worried that we're sticking our heads in the sand. I'd 
always thought that it's best to get rid of compiler warnings, if only 
to potentially stop things from breaking further down the track.

I do notice that Owen has accepted/created patches for pango that do 
much the same as I'm doing. Should we be? And if we have, shouldn't we 
be reversing them then as we're making the code harder to maintain?

The question is... would it be better to set the following compiler flags:

-Wno-strict-aliasing (from memory)
-fno-strict-aliasing
-ansi

Or even compile at only -O. Nah. Not that one.

Here's another thing I'd like to know: why did C99 even come up with 
strict aliasing? was it to improve performance? If so, maybe a small 
amount of reorganisation of code could be in order (perhaps - I'm more 
musing out loud and wouldn't really know how to go about doing it).

One last thing: half these warnings wouldn't happen if we knew what 
standard we were compiling to: C89 or C99. I asked on #gnome-hackers the 
other night and I got two responses. jdub said it was undefined (lovely 
word that) but then suggested we compile to C89 to try to keep max 
compatibility. In which case the -ansi tag would be the go. Of course, 
if we compile to C89, what features and advantages of C99 are we missing 
out on?

Anyway, please try to remember I'm kind of new to gnome hacking, and 
mainly tried to resolve the warnings as my first foray into contributing 
to the GNOME community. So go easy on the flames! Try to remember I am a 
precious and unique snowflake and I can't take the heat.

Chris

Morten Welinder wrote:
> Glib and GTK+ violate the C standards in many different ways.  It's
> probably not useful at this point to expect compliance.
>
> For example, I believe that are quite a few places that will be
> unhappy if a void* cannot hold a function type pointer.  They
> are, after all, not required to even have the same size.
>
> Consequently I believe that these warnings should be eliminated
> using as few casts as possible.  Casts are truly a maintenance
> nightmare.  If there is a way to simply make gcc shut up, then
> I think that is the way to go.
>
> Morten