Re: [RFC] libstdc++/6720 and libstdc++/6671

[Mark Mitchell <mark at codesourcery dot com>]

>| If the user put a bits/std_cmath.h in the -I path
>| ahead of the one we provided, he or she must have intended us to use
>| it.  If it doesn't work that's the user's problem.
>
> That line of reasoning misses the point that a bits/std_cmath.h may
> get in the way not because the user intended it or that that is the
> intended use (as for the library) but just by accident.  The compiler
> must deliver the expected semantics.

OK, I think the lines are drawn. :-)

Gaby's claim is that saying "-I..." doesn't mean that you don't want
conforming behavior; you're just defining some of the
implementation-defined places you look for header files.

Zack's claim is that saying "-I..." is like saying "-fno-const-strings";
you've now tweaked the language around.

I wish these CodeSourcery guys could ever agree on something. :-)

Both positions seem defensible to me.

I think the right thing to do is find out what people on the standards
reflector think.  If most people think the committee meant one thing,
we should probably do that thing.  Gaby, would you ask on the reflector
and report back?

Thanks,

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com