Re: An alias bug

[hjl at lucon dot org (H.J. Lu)]

> 
> 
>   In message <m0yv6P2-000266C@ocean.lucon.org>you write:
>   > That is strange. insn is moved unsafely by egcs in CVS as of 11:00am
>   > PDT on 980711 on my Linux/x86. BTW, I didn't use haifa-sched. I also
>   > use -O2 -g to bootstrap egcs. BTW, have you tried "make check" on my
>   > libg++ addon in egcs?
> I don't use haifa and I use "make bootstrap" to build the compiler.
> 
> I have a million othe things to look at -- so I don't have time to
> pick up your libg++ code, drop it in, test it, debug it, etc.  Especailly
> when I don't know the x86 well enough to debug it very well.
> 
> You have already done most of this work, so it seems to me we're 
> better off with you continuing to investigate the problem.  
> Particularly since you care about libg++.
> 

I will continue working on it. IMHO, it is not just a libg++ issue. It
is a generic problem in the alias code. I will report the loop bug later,
which miscompiles both C++ and C code. Basically, all those bugs boil
down to that we do not provide accurate info to alias code for it to
do its job. It happens in loop optimizer as well as in scheduler. It
shows up more often in C++ than in C. I am afraid if we don't fix it
now, it may become worse down the road as more and more people start
using egcs for serious C++ work.



-- 
H.J. Lu (hjl@gnu.org)