Am 16.10.2016 um 22:22 schrieb Marius Bakke: > Can we leverage this mechanism instead of propagating everything? Are > there any drawbacks to doing that?
propagating all inputs is the correct way for two reasons: a) In Python only *one* version of a module can but used within an environment [*] So you will end up in troubles if some packages list different (incompatible) versions of a package in a .pth file. You will have a lot of fun debugging this. (It took me one or two hours to understand what's going on there, and being one of the core developers of PyInstaller I'm quite skilled regarding all these import mechanisms in Python.) b) guix will not be able to tack these dependencies AFAIK, see https://debbugs.gnu.org/cgi/bugreport.cgi?bug=20765 [*] One can install different version of a module, but this is rarely used and not supported when installing via pip. > I'll have to do some more investigation around how those .pth files are > created, but food for thought. Perhaps it only works with eggs? They are very archaic and work with normal module, too. https://docs.python.org/3/library/site.html You may want to compare the results of (I suggest using a virtual environment for the tests): easy_install py==1.4.31 python ./setup.py easy_install py==1.4.31 python ./setup.py install py==1.4.31 python ./setup.py install --root=/ py==1.4.31 BTW: Just today I discovered nasty problems with namespace packages under python2. So I still need to do some recherché. -- Regards Hartmut Goebel | Hartmut Goebel | h.goe...@crazy-compilers.com | | www.crazy-compilers.com | compilers which you thought are impossible |