Re: current mainline, port status

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Benjamin Kosnik <benjamin.kosnik@gmail.com> writes:

| > >FAIL: 26_numerics/complex/13450.cc (test for excess errors)
| > >FAIL: 26_numerics/complex/complex_value.cc (test for excess errors)
| > >FAIL: 26_numerics/complex/pow.cc (test for excess errors)
| > >
| > >all fail with various forms of (this linker error):
| > >
| > >...libstdc++-v3/include/complex:593: undefined reference to `carg'
| > >[and cargf, cexpf, cexp, cabsl]
| > >
| > >How are other ports looking these days?
| > >  
| > >
| > I think the single most importante open issue is the above, which also
| > affects hpux, and Darwin, for instance, but I haven't really studied it
| > in detail: I'm afraid, for 4.0 at least, we can only "fix" it by a 
| > combination
| > of configury bits and testsuite tweaks.
| 
| Yo dudes.
| 
| I seem to remember this dimly: Gaby asked for help with some of the
| complex bits, where __builtin_carg wasn't available or couldn't be used,
| and so a reference to carg is used instead.
| 
| I can't find the reference right now on the list. Gaby?

I don't have the references handy neither.

As a temporary fix, we may consider adding more stubs to
libmath/stubs.c. 

But for 4.1 we should aim at having GCC come with a library (libgcc?)
where all those fallback routines are common to front-ends and
libraries.  It does not make sense that every front-end or library
goes on reinventing the wheels with different colors. 

-- Gaby