[parisc-linux] Don't rebuild gcc/glibc yet

[parisc-linux] Don't rebuild gcc/glibc yet

[David Huggins-Daines]

Hi,

glibc is currently a bit borken, probably because I foolishly tried to
enable 'long double' support in gcc:

/home/dhd/src/parisc/glibc-build/libc.a(vfscanf.o): In function `_IO_vfscanf':
/home/dhd/src/parisc/glibc/stdio-common/vfscanf.c:1582: undefined reference to `_U_Qfneg'
/home/dhd/src/parisc/glibc-build/libc.a(strtold.o): In function `__strtold_internal':
/home/dhd/src/parisc/glibc/stdlib/strtod.c:532: undefined reference to `_U_Qfcnvff_sgl_to_quad'

Is the long double stuff in gcc supposed to work at all?  (i.e. are
there definitions of those functions anywhere?)  I get 'sizeof(long
double) == 0' when configuring glibc, but strangely enough it still
wants to use them...

For now I'll back out that bit in gcc.  Sorry about this.

-- 
dhd@linuxcare.com, http://www.linuxcare.com/
Linuxcare. Support for the revolution.

Re: [parisc-linux] Don't rebuild gcc/glibc yet

[Jeffrey A Law]

  In message <87em50btaa.fsf@linuxcare.com>you write:
 > Is the long double stuff in gcc supposed to work at all? 
Yes.

  > (i.e. are there definitions of those functions anywhere?) 
I believe they're in libc.a or milli.a on hpux systems.

  > I get 'sizeof(long double) == 0' when configuring glibc, but
  . strangely enough it still wants to use them...
You should never get sizeof long double == 0 -- if long doubles are
not supported, they should be treated like doubles.
jeff

Re: [parisc-linux] Don't rebuild gcc/glibc yet

[David Huggins-Daines]

Jeffrey A Law <law@cygnus.com> writes:

>   > (i.e. are there definitions of those functions anywhere?) 
> I believe they're in libc.a or milli.a on hpux systems.

Yes, actually, gcc/config/pa/quadlib.c, as I discovered when looking
at the other PA targets.

>   > I get 'sizeof(long double) == 0' when configuring glibc, but
>   . strangely enough it still wants to use them...
> You should never get sizeof long double == 0 -- if long doubles are
> not supported, they should be treated like doubles.

Yeah, I realized why that is, it's because glibc's configure.in does:

dnl check for the size of 'long double'.
AC_CHECK_SIZEOF(long double, 0)

Which makes sense :-)

-- 
dhd@linuxcare.com, http://www.linuxcare.com/
Linuxcare. Support for the revolution.

Re: [parisc-linux] Don't rebuild gcc/glibc yet

[David Huggins-Daines]

Jeffrey A Law <law@cygnus.com> writes:

>   In message <878zv8brly.fsf@linuxcare.com>you write:
>   > Yes, actually, gcc/config/pa/quadlib.c, as I discovered when looking
>   > at the other PA targets.
> Ah.  Actually quadlib.c is just a front-end to routines provided
> elsewhere by the hpux libraries.

Right, I found that out the hard way :(

I suspect that I'll just turn it off for Linux for the time being, as
I don't know how difficult it will be to reimplement those, and
whether there is a good reason to or not.

-- 
dhd@linuxcare.com, http://www.linuxcare.com/
Linuxcare. Support for the revolution.