PR opt/10056

[Jan Hubicka]

> On Mon, 2003-03-24 at 09:27, Jan Hubicka wrote:
> > Hi,
> > here is the fix.  It is somewhat ugly - only way I found around is to
> > temporarily turn the ugly jump into normal condjump and then uglify it
> > back.  All my other attempts failed because
> > force_nonfallthru_and_redirect is called on invalid CFG and the code
> > works only because we know what force_nonfallthru_and_redirect is doing.
> 
> Thanks for working on this.  Couldn't we just treat e_taken as being the
> same as e_fall in this case?
no, things are more complicated - even if you patch cfglayout.c to get
round, it will end up calling force_nonfallthru that is not designed
to work in the cfglayout.c scheme of CFG (cfg is temporarily invalid
until the fixing finishes) and gets confused entirely.

Honza
> 
> -- 
> Mark Mitchell
> CodeSourcery, LLC
> mark@codesourcery.com