Control: tags -1 + pending
Hi, 2016-05-20 08:39 David Kalnischkies:
Hi, (disclaimer: apt, not aptitude, maintainer talking) On Fri, May 20, 2016 at 09:54:23AM +1000, James Tocknell wrote:0x00007ffff7b3feed in debVersioningSystem::CheckDep(char const*, int, char const*) () from /usr/lib/x86_64-linux-gnu/libapt-pkg.so.5.0 #0 0x00007ffff7b3feed in debVersioningSystem::CheckDep(char const*, int, char const*) () from /usr/lib/x86_64-linux-gnu/libapt-pkg.so.5.0 #1 0x00005555557767c1 in infer_reason (pkg=..., reasons=std::set with 1 elements = {...}) at ../../../../src/generic/apt/infer_reason.cc:178[It is a bit confusing that the code of frame #1 is actually calling IsSatisfied(), but it is just a wrapper around CheckDep() so lets just ignore that for the moment] 10f0cf4a04b2526e9dea65facde865d993396417 adds this code, which passes the candidate version to IsSatisfied – but not all packages have a candidate! So that code should be adapted to test first if the package has a candidate and only if so check if the candidate satisfies the dependency. Now, we could ask why IsSatisfied requires a valid version and the answer is: nobody (aka me) thought anybody would pass it an invalid version as the soul reason for the one-line IsSatisfied is to hide the call to CheckDep behind a slightly nicer interface (1 vs 3 parameters) and all existing calls in apt passed only valid versions in… I am not sure it is a good idea to add checks to that method as its in various hot paths for apt & it has the potential of hiding bugs (at the expense of opening some like here). Either way, that needs more thought than I can shell out at the moment & adapting the call site should be easy.
Thanks for the dignosis. Since I couldn't find an easy way to reproduce it (the texlive packages are huge for my wrongly-called broadband), I am basing this fix on the diagnosis above. 2016-05-24 05:03 James Tocknell:
I've got a more useful backtrace (had to build apt to get its debug symbols). The issue seems to be line 178 in infer_reason.cc, which got introduced in 0.8. Somehow (I'm not sure, everything seems to have been inlined...), it's calling debVersioningSystem::CheckDep with an invalid address.
It probably happens for the reasons that David explained. I think that it has to do with the reverse dependencies of virtual packages. I tried to reproduce it with csh|tcsh (virtual package c-shell) and some -perl reverse dependency, but I tried to remove them and mark as auto in different combinations for many minutes, and I couldn't trigger it. If you still experience this problem, it would be very nice if you confirm that it's indeed fixed when we release the next version. Cheers. -- Manuel A. Fernandez Montecelo <manuel.montez...@gmail.com>
