abi_check has been failing on sparc for a while, looking at the baseline_txt updates it hasn't been updated in a very long time. Therefore I'm pretty sure gcc-4.4 et al. need similar updating. I'll attach the detailed failure logs from make check that lists all the symbol differences.
Created attachment 20281 [details] libstdc++ abi_check failure log on sparc
Sorry if I'm misunderstanding what you are saying, but, first, I should point out that a baseline is a **baseline**, not something you update from time to time at will. Eg, the x86_64-linux baseline is certainly much older, and works perfectly ok. That said, I had a quick look to the failures and seem to me very similar to the failures I obtain on x86_64-linux if I onfigure with --enable-clocale=generic, thus overriding the normal choice of the gnu locale model. Isn't the case that on the sparc machine at issue the required localedata is missing and the configury falls back to generic?
Sorry, I overlooked that I'd been building with --disable-nls, I'll rebuild with --enable-nls and see how things look after that.
I don't think it can be the whole story. Please, double check that you have the required localedata installed, per: http://gcc.gnu.org/onlinedocs/libstdc++/manual/setup.html#manual.intro.setup.prereq and which locale model ends up being selected by [GLIBCXX_ENABLE_CLOCALE]. The ABI baselines have been generated for the GNU locale model, are not suited for the generic locale model.
I've double checked that I have the locales and everything installed. I'm building a fixed setup now, and I validated that "gnu" instead of "generic" is now choosen for the c++locale.h header file by libstdc++'s configure. I'll report the abi_check result when it's ready.
Subject: Re: FAIL: abi_check sparc --enable-nls ought not to affect the choice of locale model; it should be solely about translations of GCC's own messages. (If libstdc++'s messages were actually usefully translated - domain registered with the TP and translations handled through the TP, regeneration part of procedures documented at http://gcc.gnu.org/translation.html - then it would also be desirable to be able to configure message translation for host and target separately.)
Ok, once I straightened out all of the locale issues the abi_check failure went away. Closing.