Paolo Bonzini <bonz...@gnu.org> writes:
>> 2013-11-14 Rainer Orth <r...@cebitec.uni-bielefeld.de>
>>
>> * configure.ac (libcilkrts_cv_have_attribute_visibility): Check
>> for __attribute__((visibility)).
>> * configure: Regenerate.
>> * include/cilk/common.h (CILK_EXPORT, CILK_EXPORT_DATA): Only use
>> __attribute__((visibility)) if HAVE_ATTRIBUTE_VISIBILITY.
>>
>
> Any chance to move the test to config/?
I wondered about this myself. A quick assement of visibility checks in
gcc revealed the following:
* gcc: test needs to be kept separate since it checks assembler and
linker support directly
* libatomic, libcilkrts, libgfortran, libgomp, libitm, libquadmath,
libssp: all define HAVE_ATTRIBUTE_VISIBILITY in either acinclude.m4 or
configure.ac
This set is the least problematic, most libs only use hidden
visibility in their code, libcilkrts differs: protected and default
instead. Perhaps the autoconf macro needs to be parameterized to
select which visibilities to check for?
* libffi: defines HAVE_HIDDEN_VISIBILITY_ATTRIBUTE instead, need to
check if including new config/visibility.m4 is acceptable in upstream
repo
* libgcc: used to define make variable instead, should be adaptable,
uses protected, hidden, default
* libstdc++-v3: used to define _GLIBCXX_HAVE_ATTRIBUTE_VISIBILITY,
prefix might fall from libstdc++-v3 configury automatically
> Otherwise ok.
Since this fixes a bootstrap failure, I've installed the current patch
for the moment.
Rainer
--
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University
