This is the mail archive of the libstdc++@gcc.gnu.org mailing list for the libstdc++ project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Couple of new issues in mainline


Hi Benjamin,

which ultimately makes this kind of problem possible, I'm wondering whether something went slightly, unexpectedly, wrong with Benjamin's doxygen changes around the same time, or, more generally, whether we should tackle the problem in another way... Thanks in advance for any feedback!


Hey Paolo! I noticed this as well.

However, I don't think it is the doxygen patch, as that is now on the
4.2 branch as well, where it had no ABI testing impact. (My testing
pre-checkin also showed no impact.)


Exactly, sorry for the false alarm.

I'm hoping that this is some middle-end or optimization tweak that is
currently on-going in the compiler.


That's also my hope, we can well wait a bit and see whether that ctype<char>::widen will be inlined again (in the meanwhile, I double checked that marking it always inline indeed removes the problem). Shall we file a PR for that? Worst case we can tweak the linker script.

Anyway, both from the optimization point of view and the present point of view (I would call it "abi stability', or something similar), exploring visibility-inlines-hidded seems a good idea. However, I'm noticing something strange, as I said already: for example the fact that the destructors of those standard exception classes are *out of line*, still end up hidden, not exported from the library, in my experiment. Can you explain that? I'm still confused. Maybe should we ask Jason?

Paolo.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]