<cjwatson> doko: any progress on bug 684703? the last comment on the bug is ScottK asking you for how to provide the information you requested; did you discuss that outside the bug? <doko> cjwatson, ScottK: are these non-virtual thunks referred to in other libs? <doko> sorry, s/libs/binaries/ <ScottK> They are non-virtual thunk. <doko> ScottK: then why add these to the symbols files at all? <ScottK> doko: It's what dpkg-gensymbols generates. <ScottK> c++filt _ZThn16_N11KDcrawIface12RExpanderBoxD0Ev <ScottK> non-virtual thunk to KDcrawIface::RExpanderBox::~RExpanderBox() <ScottK> That's in libkdcraw9.symbols <doko> ScottK: well, maybe ask buxy why he wants these mentioned in the symbols files? <ScottK> doko: They weren't different on different archs before. Why now? <buxy> I don't "pick" symbols, I take everything that's exported. <ScottK> doko: I'm not a build system expert, but right now I have to test build that package on all archs to get a package that builds. That's not supportable. <doko> ScottK, buxy: well, symbols files shouldn't be used unreflected for C++ libs. Debian #605833 is another (unrelated) case <ubottu> Debian bug 605833 in gpsd "symbols file lists symbols from qt libs" [Important,Open] http://bugs.debian.org/605833 <buxy> doko: true, but MoDaX has done such work there that it gets more common... <doko> buxy: which work? <buxy> doko: the possibility to map symbols by their demangled name <buxy> and other similar stuff <doko> ScottK, buxy: so if we agree that these thunks don't belong the interface, then maybe make them optional by default? <buxy> doko: I have no idea whether they belong to the interface or not, and I don't think that I want to have lots of special-cases in dpkg-gensymbols, in particular when these are difficult to identify (I don't want to always run c++filt) <buxy> I rather suggest that the affected package uses the (c++) tag to match this symbol in arch-neutral way <doko> buxy: well, these "special cases" are defined. http://www.codesourcery.com/public/cxx-abi/abi.html#mangling <ScottK> doko: I don't know enough to have an opinion of if they should be optional by default. I just know it's a lot more effort to maintain symbols files if they vary per arch. <doko> ScottK: sure, but you have to live with it <seb128> bcurtiswx, it will be easier <SpamapS> less confusing then to use -8 :) <ScottK> doko: Why? I didn't have to live with it with gcc4.4. <doko> ScottK: because types of parameters are mangled too
** Bug watch added: Debian Bug tracker #605833 http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=605833 -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/684703 Title: Generated symbols different on different archs with gcc-4.5 -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
