Destrictiveness of expansion and tree inlining deadlock?

[Jan Hubicka]

> 
> 
> --On Saturday, February 22, 2003 11:20:31 AM +0100 Jan Hubicka <jh@suse.cz> 
> wrote:
> 
> > Expansion is (to my
> >understanding) destructive and makes later inlining impossible.
> 
> That should not be the case.  It was the case with the RTL inliner, but it 
> should not be the case with the tree inliner.

Hmm, my topological sort in unit-at-a-time manages to assemble most of
inline functions before they are inlined and I am hitting one crash in:
  else if (TREE_CODE (*tp) == UNSAVE_EXPR)
    /* UNSAVE_EXPRs should not be generated until expansion time.  */
    abort ();
This made me believe that expansion creates those trees and thus is
destructive.  In the case this is wrong, what happens here?

Honza
> 
> -- 
> Mark Mitchell                mark@codesourcery.com
> CodeSourcery, LLC            http://www.codesourcery.com