This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Introduce abi_word_mode (Richard Kenner) writes:

> > I agree.  I think that it's unfortunate that GCC didn't clearly document 
> > this limitation (perhaps because nobody thought of it); it shows once 
> > again how we can get burned by adding source extensions.  But, to me, it 
> > really does seem the lesser of two evils to preserve the current meaning 
> > (or to remove the feature entirely) rather than to change its meaning.
> Why would you want to remove it?  I don't think there's any other way at the
> source level of stating you want a type whose width is BITS_PER_WORD and
> that's a valuable thing to have for portability reasons.

I think it's only superficially valuable.  I think that when you
really try to write code that uses it, you find that it is actually
not very helpful.  The type of exact width BITS_PER_WORD has very
little meaning or use.  It doesn't help you pin down an ABI, because
of the very issues that we are discussing on this thread.  And it
doesn't help for writing optimal non-ABI code--for that you can use
types like int_fast32_t which are already standardized and supported.
And it doesn't help writing asm code, because you already need to know
the register sizes anyhow.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]