On Sat, Nov 26, 2016 at 04:13:45PM +0500, Andrey Rahmatullin wrote: > On Sat, Nov 26, 2016 at 11:55:21AM +0100, Santiago Vila wrote: > > > 'boost::this_thread::hidden::sleep_until(timespec const&)' is defined in > > > libboost_thread.so.1.62.0 but not in libboost_thread.so.1.61.0 so this > > > looks like an artifact of the boost transition. And currently on my sid > > > machine the problem cannot be reproduced because > > > libgnuradio-radar.so.3.7.10 is now indirectly linked to > > > libboost_thread.so.1.62.0 and can find the symbol. OTOH the library is not > > > linked to either of libboost_thread.so directly, so > > > there is some kind of bug here, I'm not sure if it's RC though. > > > > I reported this as "serious" (hence RC) because Release Policy says > > that packages in stretch must be buildable in stretch. > Indeed. OTOH I guess there will be no boost 1.61 in stretch at the point > of the release. > Also, it looks like it's not the fault of the software as it doesn't have > code that calls sleep_until explicitly (and the function exists in 1.61, > just in a different namespase). > > > Maybe some of the build-depends should be versioned? > I don't think so.
Well, you said this was an artifact of the boost transition and in fact, I can't reproduce this anymore in stretch (so you can consider this "fixed" if you like). However, transition artifacts should never propagate to testing, and that's why I think a suitable versioned build-depends could have avoided this. Thanks.