PR tree-inline/44063 (ice in cgraph_estimate_size_after_inlining)

[Jan Hubicka]

> On Tue, May 11, 2010 at 4:54 PM, Jan Hubicka <hubicka@ucw.cz> wrote:
> >
> > Hi,
> > Richard's always_inline cleanup patch cause us to assume that all always inline functions
> > are of size 0 that breaks when we inline into them.
> 
> Btw, I changed early inlining to never inline regular functions
> into always_inline functions.
> Maybe we should avoid doing so during IPA inlining as well?

This does not work with kernel that is happy to drop always_inline on externally visible
functions.  We should inline into them.

Honza