Re: (fyi) SEGV in jc1 for java.util.Date (from jikes bytecode only)

[Anthony Green <green@cygnus.com>]

Godmar wrote:
> FWIW, jikes is known to put nops in certain places where other
> compilers don't.

Yes, I just noticed that.  It also appears to leave in non-nop
dead bytecode.

> That's probably not your problem: I thought you might find this
> pecularity interesting anyway.

Actually, yes it was the cause of one problem.  I've been testing
jikes with some parts of libgcj.  Gcj doesn't process `dead code', and
this was screwing up the EH ranges (which appear to include dead code
now and then).

I have patch for this that I'm testing right now.  The idea is to
replace the dead bytecode ranges with nops before generating trees.
The EH ranges get set correctly and the backend removes the nops.

Thanks,

AG

-- 
Anthony Green                                               Cygnus Solutions
                                                       Sunnyvale, California