On Jul 20, 2015 5:56 PM, "Ludovic Courtès" <l...@gnu.org> wrote: > "Claes Wallin (韋嘉誠)" <g...@clacke.user.lysator.liu.se> skribis:
> >> > + (license gpl2+))) > >> > >> R being GPLv3+, this should be the same. > > > > I understand the reasoning that a package is more user-oriented than > > developer-oriented and should reflect the license of the whole, but there's > > an argument for reflecting the original license as well. Has this been > > discussed? > > This has been mentioned in past reviews. Basically the intent is for > ‘license’ to reflect the license of the whole, but we often end up > leaving a comment in cases where there’s some ambiguity. > > I think it’s hard to do better without maintaining ‘copyright’ files > à la Debian. I suddenly had an idea. There is a directional compatibility graph between the most common licenses. That means calculating the license of a package can often be trivial, at least a conservative guess. You could just "guix license python-rpy2" and it would tell you that the code itself is GPLv2+ (because package definition says so), but because of dependencies the package as installed is effectively GPLv3+. Except when you can't, so there should be an "it's complicated" state as well, possibly resolvable through manual hints in package defs. This could be a check in lint as well, to make sure e.g. no GPLv2 package relies on a GPLv3 package without an explanation. Defining exactly in what way a package depends on another (i.e. if it makes it a derivative) could be a later excercise.
