[Patch] Fix ac_c99_complex configury

[Mark Mitchell]

Paolo Carlini wrote:
> Mark Mitchell wrote:
> 
> 
>>Heh.  Do you actually know the answer to my question, or are you
>>confused like me?
> 
> 
> I don't, really, on the spot: otherwise I would have insisted with
> Benjamin (right now I'm absorbed by move semantics, by the way): I can
> *see* that configure.ac is clearly split in two parts, and a working
> linker is *not* assumed for crosses. Honestly, I have *no* idea why: I
> don't build cross-compilers very often myself, admittedly. Can you
> figure out something more by looking at configure.ac?!? I would really
> like to learn more about all that.

I looked; it doesn't make sense to me.  We certainly use link-tests for 
other aspects of cross configuration.  All I see is:

   # Here's why GLIBCXX_CHECK_MATH_SUPPORT, and other autoconf macros 

   # that involve linking, can't be used: 

   #    "cannot open sim-crt0.o" 

   #    "cannot open crt0.o" 


I actually think that this means that someone's got a botched toolchain 
build, but, hey, I could certainly be wrong.

> Speaking of questions: what about my question regarding
> include/backward? Are your customers still including <iostream.h>? Tell
> me not ;)

I didn't realize that was directed at me.

I'm sure some our customers are still using <iostream.h>...  It's not a 
terribly hard thing to change, but it can mean updating all of your 
(thousands of) source files.  I don't have an opinion on whether V3 
should rip this out or not.

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304