[C++ Patch] PR 37087

[Mark Mitchell]

Paolo Carlini wrote:

> if you have a couple of minutes, I need a clarification about those 
> permerror, used in checks having to do with specializations / 
> instantiations in a different namespace: I heard about an undocumented 
> extension to that effect, which used to be available in old GCCs, but I 
> don't have any accurate, up to date information about that. To give you 
> an idea, I tried changing all those permerror to error (see the 
> attached), and the testsuite still passes... Are there concrete examples 
> which we know we are currently supporting and we want to keep on 
> supporting (with -fpermissive, of course)?

I do remember that we added those checks at some point, so there 
definitely were versions that didn't do them, and there definitely was 
code that broke because I remember running into the error myself.  I 
suspect most such code won't work now anyhow, though; the additional 
strictness we imposed with "typename" and such was far more violent.

I guess I think we should try to leave the permerror in place, unless 
there's a fundamental reason we can't make it work, and add a test (with 
-fpermissive) for this to the testsuite so that we keep it working.

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