Re: [PATCH] Fix ipa-pure-const with same_body aliases (take 2)

[Jan Hubicka <hubicka at ucw dot cz>]

> Hi!
> 
> On Mon, Dec 14, 2009 at 12:56:54PM +0100, Richard Guenther wrote:
> > Hm, indeed not very much prettier.  I'll let Honza break the tie here
> > (I'm ok with either variant, some people might argue abstraction is
> > always better - I'm not so strict here).
> 
> Based on IRC feedback from Honza, here is what I'm bootstrapping/regtesting
> ATM:

Thanks, this is OK.
As for the cgraph_set_* api, I think it is good idea from the POV of person reading ipa-*
code that won't need to dive into details of what same_body_alias is about.
For a while I plan to add similar simple helpers for other operations in cgraph to make
code more readable, but didn't get around to do that so far...

It might be interesting to re-investigate when except.c code is effective.  If
I remember well, the original problem was difference of can_throw_external and
can_throw on FP operations in gimple and RTL land that should be fixed, so in
theory, except.c should match only if we miss some gimple optimization.

Honza