Matthias Klose wrote: > Thiemo Seufer writes: > > Seems like the BTS hides away patches buried in controls messages. > > Appended and resent again. > > > > > > Thiemo > > > Thiemo, this is your call as a mips maintainer. The patch doesn't > change the method how the lock works, so it is compatible with code > compiled with the old version of atomicity.h, isn't it?
This is correct. It doesn't even change an internal interface, purely the implementation is slightly different. > The code is found in a header file, so which packages need to be > rebuilt to actually fix the bug? libstdc++v3 is the only _direct_ user. When installed, atomicity.h is hidden away in /usr/include/c++/3.3/mips-linux/bits/, this makes it unlikely some other package uses it by accident. OTOH, it gets included in std::string, and the version in gcc-3.3 is inline, so in worst case a recompilation of all standard template library users is needed to be on the safe side. For gcc-3.4 the atomic functions aren't inlined, so only libstdc++v3 itself and users of a static libstdc++v3 would be affected. Probably things aren't as bad as the worst case suggests, at least I was able to replace on a R10000 system the current libstdc++v3 with a fixed one, and haven't seen any problems related to this since then. Thiemo
