[patch] Backport of fix for PR c++/27398 to 4.1 branch

[Mark Mitchell]

Jason Merrill wrote:
> It doesn't seem worthwhile to me to backport patches for ICE after error 
> bugs.  The benefit is small (especially since release branches will just 
> say "confused by earlier errors"), so why risk making changes?

I agree.

I wouldn't make this a blanket rule; if there are particularly simple 
changes, I'd let people exercise judgment, but I do think that this is a 
relatively unimportant class of bugs.  If the C++ front end were a 
separable unit -- as I hope some day it will be! -- then there are 
applications (IDEs, for example), where being completely crash-proof is 
very important.  But, for compilation per se, it's not that important, 
especially now that interactive compilation is so fast.  (In the days of 
punch cards, you really wanted to get all your syntax errors on each 
compilation run!)

-- 
Mark Mitchell
CodeSourcery
mark@codesourcery.com
(650) 331-3385 x713