On Wed, Oct 15, 2014 at 11:52:59PM +0200, Jakub Jelinek wrote:
> On Wed, Oct 15, 2014 at 03:48:20PM -0600, Jeff Law wrote:
> > Isn't slamming error_mark_node well established at this point? I fact I
> > recall seeing it documented to be used in this kind of way to prevent future
> > errors.
>
> I think in this case it is way too risky to put error_mark_node there,
> it will affect also all the places where the object with that type isn't
> dereferenced.
> If there are spare bits on the type, either using one for this error, or
> to represent some error has been diagnosed for the particular type and
> hash table lookup could be used to look up which exactly, or just hash table
> without any bit...
I thought sticking error_mark_node into the type would be feasible
here, because we do it elsewhere too, and the program is invalid
anyway.
Well - maybe the new patch is better.
Thanks,
Marek
