On Thu, 25 Feb 2010 13:09:55 -0500, Gaetan Nadon <[email protected]> wrote:

> Can we assume that all modules who did not have -fno-strict-aliasing
> before Jan 2009 have been working correctly and do not need it? That
> gave about 4 years to detect any problems. I am not sure what's the set
> of modules you have in mind that should have -fno-strict-aliasing.

My main concern is in code dealing with either the wire protocol or
image data, with a minor concern for the rest of the X server which does
DrawablePtr/PixmapPtr/WindowPtr hacks. The protocol code could
presumably be fixed to use void * everywhere, which (iirc) bypasses the
aliasing optimization.

Having this be consistent across all protocol libraries and server
modules seems like a good idea, even if there aren't obvious bugs
present now, future changes in GCC may well cause old code to break in
the future and we'll need a hook to make sure that we can apply options
globally across the system.

Another option we may want to consider is -fwrapv which makes signed and
unsigned int/long/char operators operate over the finite field of 2ⁿ
instead of some random set.

Thanks, gcc developers. I'm sure you can read the ANSI C spec in a way
which justifies your optimizations. Please give me back my language.

> I am available to write as many patches as it takes to do the right
> thing.

I wouldn't admit that in public :-)

-- 
[email protected]

Attachment: pgpK4rEhk8VXd.pgp
Description: PGP signature

_______________________________________________
xorg-devel mailing list
[email protected]
http://lists.x.org/mailman/listinfo/xorg-devel

Reply via email to