This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: C constant expressions proposals
On Sat, 14 Aug 2004, Zack Weinberg wrote:
> I also don't think this category of bugs ("fine details of what is or
> is not valid C") is worth spending much time or effort on, compared to
> the giant pile of wrong-code-generation, ICE, and poor-optimization
> bugs that are out there.
That sort of argument is why the proposal (or rather the introductory text
around it) specifically lists other benefits in terms of cleaning up
misuses of trees (in advance of constant expressions implementation) and
getting rid of NON_LVALUE_EXPR (afterwards). (This or that feature has a
kludged use to do some bit of constant expressions, so to get rid of that
feature without regressions a better implementation is needed first, after
which that feature can be got rid of.)
Admittedly such cleanups tend to address compile-time performance and
memory consumption - which have systematically regressed far worse than
the other things you list - rather than any of the other things you list,
but Andrew Pinski mentioned (bug 16437, comment 5):
Yes TREE_OVERFLOW should die, there is another PR which causes an
optimization regression because of it.
- which I can well-believe (I think a lot of the middle-end code using
TREE_OVERFLOW is dodgy), and replacing TREE_OVERFLOW and
TREE_CONSTANT_OVERFLOW by more specific front-end information is something
that also logically follows doing constant expressions.
> Since you specifically ask about 20010327-1.c, I think it looks
> awfully like the sort of code that people write down in the guts of an
> operating system, and I think we have broken those people's code a lot
> and maybe we should think about cutting them some slack.
It's only the version with a cast to an integer of different width that's
a problem (and specifically when longs are a different width from
pointers). The case of a cast to an integer of the same width is
20011114-1.c, which I propose to keep valid.
--
Joseph S. Myers http://www.srcf.ucam.org/~jsm28/gcc/
jsm@polyomino.org.uk (personal mail)
jsm28@gcc.gnu.org (Bugzilla assignments and CCs)