PR traget/24188 (-mcmodel=medium ICE)

[Jan Hubicka]

> On Fri, Nov 04, 2005 at 11:14:56PM +0100, Jan Hubicka wrote:
> > > On Fri, Nov 04, 2005 at 03:59:27PM -0500, Jakub Jelinek wrote:
> > > > But I'm pretty certain none of those intrinsic routines would mind
> > > > if NULL has been passed instead.
> > > 
> > > I really don't like zero sized objects at all.  If we can get
> > > away with not creating them in the first place, I think that
> > > would be best.
> > 
> > I can drop the size 0 test part of patch (still we should not ICE on
> > large strings) and change this to fortran frontend bug...
> 
> Can you please do this?
> As shown in PR25254. the strings don't need to be that large with
> -mlarge-data-threshold=1 ...

I am testing the updated patch.
> 
> BTW, is it intentional that such strings aren't merged?  I think we'd
> either need to extract part of mergeable_string_section into a helper
> function that would be also used by i386.c, or duplicate that code.

I originally tought it is not that important to merge the strings, since
the probability of match for large strings is pretty low, but if you are
interested to implement this feature, I won't complain at all ;)

Honza
> 
> 	Jakub