Leo Famulari writes: > On Sat, Jun 18, 2016 at 10:44:16PM -0500, Lukas Gradl wrote: >> Lukas Gradl writes: >> >> > So the flags -I, -L, -l for zlib are all there for both dynamic and >> > static linking. But still no store reference. >> >> I think the reason why there is no reference is that msgpack uses >> zbuffer only for tests. Before compilation, the file only references >> the name "zlib" and does not mention the hash in the path of zlib in the >> store. During compilation (during "check"), this mere name "zlib" gets >> somehow resolved to the path of zlib in the store. The binary file >> resulting from compiling zbuffer should therefore contain a reference to >> zlib, which should be detectable by guix gc --references. I think >> however, that this binary file does not get installed as it is only used >> for tests. All the files that do get installed in the output path of >> msgpack in the store do not contain the hash part of the store-path of >> zlib. They only refer to zlib by name. IIUC this can not be detected >> by guix gc --references since it only searches for the hash part of >> the store-path of zlib. >> >> This is what I grasp from looking at libstore/references.cc and >> libstore/store-api.cc. I am not sure about this though. > > That's my understanding as well. > > Thanks for trying all the things you described in your previous email. > > Since we are still stuck on this, I think we should continue with Ring > and see if this becomes a problem later on. If so, let's contact the > msgpack maintainers and ask for advice. OK, sounds good. > > We should add a caveat in a comment in msgpack's package definition, for > future readers. I added a comment in the attached patch. > > I'm curious — how close are you to a Ring package definition? Yes, packaging Ring is taking longer than I expected (on my side that is). I will send a detailed update within the next two hours. Thank you for all your work on this. I would have never even noticed that issue! Thank you! Best, Lukas