[PATCH] Get rid of awkward precision for bitsize types

[Eric Botcazou]

> But it would have the same problem, no?

Truncating or not truncating at run time doesn't change anything since we 
pretend that sizetypes don't overflow.  This worked perfectly before the 
bitfield truncating patch from 2004, it purely pessimized for sizetypes.

> How do we end up with computations in bitsizetype being expanded anyway?  I
> suppose for expansion we want to use the _UNIT variants of everything?

For variable-sized types, both TYPE_SIZE and TYPE_SIZE_UNIT can be big 
expressions which are gimplified; at -O0 nothing cleans up the mess.

-- 
Eric Botcazou