Re: [MiNT] Some comments on GCC 4

[Alan Hourihane <alanh@fairlite.demon.co.uk>]

On Thu, 2008-03-27 at 09:51 +0100, J. F. Lemaire wrote:
> Hello,
> 
> With the help of MiKRO I installed GCC 4.2.3 on my Falcon and compiled 
> the CVS mintlib. I then went along to compile my own project (GEM 
> MiNTnet client) and I would like to share a few thoughts on the 
> process/result.
> 
> - the file 'unistd.h' from CVS mintlib caused literally hundreds of 
> errors, starting from line 461 (I haven't taken note of which ones 
> though). I moved the file and copied over the one from the latest 
> official mintlib and the errors went away;

I would think you may not be including something before unistd which may
be required in your circumstance. Without seeing the errors I can't say,
but CVS mintlib unistd.h is fine.

> - the folder ../include/sys has been replaced by ../include/bits in CVS. 
> Is this normal?

You should still have include/sys, as well as include/bits. Not sure why
you are saying ../

> - the -s option given to GCC does not seem to result in stripping the 
> binaries.

Seems to work fine here, are you sure your application hasn't already 
been stripped with the strip command ?

> - the memory consumption of my program has increased from ca.340 KB to 
> ca.400.

Are you looking at the file size ? as memory consumption shouldn't
change.

If it's file size that's probably because of optimization. You could try
-Os instead of -O2 which would reduce the file size.

> - the only new warning I had (which may or may not mean that my code is 
> clean...) was 'dereferencing type-punned pointer will break 
> strict-aliasing rules' on the following line:
>
> Col = (OBJC_COLORWORD *)&Tree[object].ob_spec.tedinfo->te_color;
> 
> If I understand correctly, GCC 4 does not like the cast. Is there 
> another way to get the colour index of an object than via a cast to 
> OBJC_COLORWORD?

That's a bug in your application, and you can use union's to fix this
type of problem.

Alan.