Hi Santiago, On So, 2016-09-11 at 11:37 +0200, Santiago Vila wrote: (...) > This is the changelog entry you wrote: > > > * [8ee94bc] gnupg.py: Add extra method to get sec user id. Fixes > doctest > > for GPG 2.1. Thanks to Santiago Vila (Closes: 834683) > > If this is only intended to work with gnupg 2, please add > Build-Depends: gnupg (>= 2) so that autobuilders testing packages in > stretch do not even try to build it with gnupg version 1.
no, this should work for both, 1.4 and 2.1. > OTOH, if this is intended to work with both gnupg 1 and gnupg 2 > (for example, if you intend this to be backported to jessie), > then the problem is still here. Yes, I do ;). And I did test it under jessie, and also the package build under stretch. > The error message: (...) > suggests to me that there is not enough entropy to generate a key. If entropy actually is the problem, it should have always been there for all 1.0.x versions (having that doctest). > (I don't know how to fix that, sorry, maybe an additional > build-depends on some package which wraps accesses to /dev/random to > make them faster, if such package exists). > > The full build log is attached. > > This time I only tried to build it once, but since the problem was > not > supposed to always happen, it is probably correct to say that the > FTBFS-randomness has not been eliminated. Ok, agree, this does add some randomness [which I usually mitigate running something like haveged on the builder host]. I guess this generally means automated tests depending on some entropy must be avoided? Anyway, I don't have a good solution either right now; maybe someone already has? Will look around eventually ;) Thx, S