[v3] revisited QNX 6.1 patch

[Benjamin Kosnik]

> I'd like to follow up on the C header issue as soon as I can clear a
> day or two out of my schedule.  For the time being, I am configuring
> with --enable-cheaders=c.  It mostly works.  I've enclosed the tweaks
> I'm using.  The declaration in new_op.cc conflicts with the system 
> headers, and src/cmath.cc doesn't build with cheaders=c.  There is
> a comment to that effect in that file, so I haven't followed up.

Ok. I'm very intrigued with the idea of getting c headers vs c_std
working for more targets. Although this may sound incredible, I realized
last week that I always experimented with "C" headers on linux (using
glibc headers), when getting something working with a newlib cross would
be much, much easier (and hopefully much less frustrating for me.)

I'll try to build up a hacked version of newlib that emulates what I
suspect is going on with QNX (I suspect that these are relatively
simple, non-nested declarations wrapped with namespace std in addition
to extern "C",  much like solaris 2.8). If I can figure this out with
free tools then getting all the i's dotted and t's crossed for a more
correct "C" header implementation should fall out for all platforms.
That would be cool, and sounds like a fun activity for a friday
afternoon.

> But since our code compiles with and runs well using my version of the
> toolchain, and we haven't found any problems, I'm having a hard time
> getting mgmt to give me the time to finish.

That's encouraging, although I (like you, apparently) would be
interested
in a more formal reporting of results. I'm just curious.

You don't happen to have size comparisons for your gcc toolchain versus
the QNX tools for C++ binaries, do you? I'm expecting the gcc toolchain
to be significantly smaller.

 -benjamin