Re: problem getting std::string::_Rep::_S_max_size and std::string::_ Rep::_S_terminal under AIX linked

[Benjamin Kosnik <bkoz at redhat dot com>]

> compiling the program worked fine. but when the thing got to linking
> problems occured. the two symbols std::string::_Rep::_S_max_size and
> std::string::_Rep::_S_terminal couldn't be ressolved. so i looked in google

...

> next thing i did was looking in the mailing list archive of gnu. i found
> several threads, but i didn't realize the solution for the problem.

...

> i don't know what else to do. could someboldy please help me?

Sure. I just asked this question to David Edelsohn myself. Here is his
reply to me: 

--------

	In addition to weak symbol support, V3 relies on SVR4 symbol
merging "run-time linking" semantics, especially for C++ streams support.
That is not AIX's default shared object behavior.  AIX 4.2 and later
provided that as an option through -G/-brtl linker flags.  The problem is
that both the library *and* the application linked against the library
need to be linked with the appropriate flags.  This is why I added -Wl,-G
to OPT_LDFLAGS.

	Now libstdc++.a is created "run-time linking enabled" with the -G
flag, but the testcases are not linked with the -brtl flag.  I cannot see
a way to inform the testsuite to use that extra flag.

	Running the testsuite manually with that flag in GCC 3.0, I obtain
results like:

FAIL: 21_strings/capacity.cc (test for excess errors)
WARNING: 21_strings/capacity.cc compilation failed to produce executable
FAIL: 22_locale/ctype_char_members.cc execution test
FAIL: 22_locale/numpunct.cc (test for excess errors)
WARNING: 22_locale/numpunct.cc compilation failed to produce executable
FAIL: 26_numerics/c99_classification_macros_c++.cc (test for excess
errors)
FAIL: 26_numerics/complex_inserters_extractors.cc (test for excess errors)
WARNING: 26_numerics/complex_inserters_extractors.cc compilation failed to
produ
ce executable

                === libstdc++-v3 Summary ===

# of expected passes            239
# of unexpected failures        5
# of expected failures          8

which I think you will agree are acceptable.

	libtool understands "-Wl,-brtl" as a special flag telling it to
build libraries with the -Wl,-brtl flag as well, but it changes other
behaviors such as producing shared libraries named libstdc++.so instead of
libstdc++.a.  AIX only looks for libstdc++.so if an application is linked
with -brtl, so that would break normal linking.  -brtl may adversely
affect AIX applications which do not want SVR4 semantics.  And we do not
want all GCC target libraries linked with -G.  Plus V3 testsuite does not
use it, so *everything* fails to link.  Conclusion: the libtool special
signal is not an option.

	So, I use OPT_LDFLAGS and inform AIX users in the GCC Manual
compatibility page to use -brtl.  I and IBM will just have to teach AIX
GCC users to use this extra option when linking with libstdc++-v3.

	BTW, linking with -brtl in 3.1 does not appear much better because
the test harness reports the linker warnings as failures.  Even with the
-G/-brtl flag, one gets warnings of duplicate symbols for the BSS and DATA
symbols which it is preparing to merge.

--------

It would be nice to fix this for gcc-3.1. Anyway, hope you find it 
helpful. I suggest looking at the AIX parts of the gcc manual.

-benjamin