Re: [Chicken-users] More on aliasing.

[felix]

>Basically, if a C_word is an int, things are probably ok, but if a C_word
is a
> long long, you're in trouble.
>

This sounds promising. On the other hand the code Chicken generates casts
really
furiously, and the code is quite demanding to the C compiler. Not
necessarily
hard to compile, but it combines some seldom-used features (noreturn
attributes,
some regparm attributes here and there, etc.) This can trigger compiler bugs
(in gcc)
which would normally not appear in human-written code. Especially the
`noreturn'
attr is probably not used as much in any other machine-generated code.

I think what we should do is enable `-fstrict-aliasing' only on platforms
where it definitely works (x86, ppc) and explicitly use
`-fno-strict-aliasing'.
I heard somewhere that GCC 3.0 enables it by default. Is this true?
(Brad Lucier reported in e-mail that this option is needed to compile
Gambit-code correctly).


cheers,
felix