Re: [PATCH] Fix pb in nested function decomposition pass

[Richard Henderson <rth at redhat dot com>]

On Wed, Sep 14, 2005 at 11:23:37PM +0200, Eric Botcazou wrote:
> I personally think that this would be less clean than what we currently have, 
> i.e. a self-contained pass that runs very early so that all functions are 
> equal when the cgraph machinery kicks in.  We would need to delay the pass 
> until after the cgraph reachability stuff has been run, which IMHO is asking 
> for future troubles.  But I do see the merits of your position.

Or push forward some of the reachability stuff for the purpose of nested
functions.  I'll be the first to admit that cgraph is a twisty mess at
the moment, and that it calls out for a serious looking-at.

It may be that things get squiggly and we need to invoke the reachability
code from within tree-nested.c, but even in that case I'd hope that we can
re-use the actual scanning code from cgraph.

But that's all for later examination.


>     * tree-nested.c (get_frame_type): Mark the "non-local frame structure"
>     as addressable.

Ok.



r~