4.4.0-RC bootstrap failed on i686-pc-solaris2.10

[Jonathan Wakely]

2009/4/22 Ian Lance Taylor:
> Jonathan Wakely writes:
>
>> I tried to build the 4.4.0-RC-20090414 on Solaris x86, configured with
>>
>> /var/tmp/build-gcc/gcc-4.4.0-RC-20090414/configure
>> --prefix=/opt/gcc/32-bit/4.4.0-RC-20090414 --enable-languages=c,c++
>> --with-gnu-as --with-as=/var/tmp/build-gcc/binutils_2.18/bin/as
>> --with-ld=/usr/ccs/bin/ld --with-gmp=/var/tmp/build-gcc/stage
>> --with-mpfr=/var/tmp/build-gcc/stage --build=i686-pc-solaris2.10
>> --with-arch-32=pentium4 --with-system-zlib --enable-shared
>> --enable-threads=posix --enable-nls --enable-libstdcxx-debug
>>
>> where the 'stage' dir contains gmp and mpfr installations, built with
>> --disable-shared. The --with-as path is a symlink to the real binutils
>> dir, I realise I don't want to use that /var/tmp path in a real build,
>> but I'm just trying to get my build script working for now.
>
>> ld: fatal: symbol `__i686.get_pc_thunk.bx' is multiply-defined:
>>         (file /var/tmp/build-gcc/gcc/./gcc/crtbegin.o type=FUNC; file
>> _enable_execute_stack_s.o type=FUNC);
>
> This means that the configure script thought that hidden linkonce
> symbols were supported, but for some reason they are not working.  This
> could happen if the configure script used one linker and the build used
> another, but that shouldn't happen with the above configure command.  I
> guess look at config.log to see the test used to set
> USE_HIDDEN_LINKONCE, and compare it to the object files which are
> failing to link.

Further investigation shows that this problem only happens with
--build=i686-pc-solaris2.10, without that the thunks are not emitted,
and so there's no problem.

gcc_cv_as_hidden is set conditionally after checking for a suitable
version of GNU as. For my build (using GNU as from binutils 2.18) that
gets set to yes.

gcc/configure assumes that the linker always supports hidden symbols
for ,*-*-solaris2.9* or *-*-solaris2.1[0-9]*, so sets
gcc_cv_ld_hidden=yes (this is what changed since 4.3.2, which always
set it to no.)

Should I enter a bug or is --build=i686-pc-solaris2.10 not supported?
My configuration used to work with 4.3.2 and seems to fail because
gcc/configure now makes wrong assumptions.

Jonathan