Christopher Li wrote:
> On Wed, Feb 14, 2007 at 11:54:47AM -0800, Christopher Li wrote:
>> On Wed, Feb 14, 2007 at 10:24:09AM -0800, Linus Torvalds wrote:
>>> Actually, it's often better to just explicitly make "ctype" be 
>>> "&bad_ctype" instead.
>>>
>>> It's nice if a NULL type means "type has not been evaluated yet", and then 
>>> using "&bad_ctype" to mean "type evaluated to crap".
>> Good idea. Some of the code already doing that. I am making it as a separate
>> patch because touch a lot of code.
> 
> This is very messy. The current evaluate_* functions are assuming NULL means for
> "some error happen in this expression, I can't get the ctype". It is all over the
> place.

In that case, I will go ahead and apply your patch to eliminate the segfault, and
I can merge a new patch that uses &bad_ctype later.

- Josh Triplett