[14:33:06] <ronny> szx: the ideal way for installation on windows is probably wheels
[14:33:20] <ronny> szx: combined with entrypoints of course
[14:47:49] <szx> ronny: it is possible to make it work with a script with hyphens in filename?
[14:48:18] <szx> i'm getting this error: error in x-x-x setup command: ('Invalid module name', 'x-x-x')
[14:48:58] <ronny> szx: the script name can have them, python modules obviously not
[14:50:09] <ronny> szx: its common to put the code for the commandline into something like mypkg.cli anduse my-script = pypkg.cli:main as entrypoint
[16:21:08] <The-Compiler> I'm trying to install a git-package via pip, but I see a PermissionError and then an IndexError inside pip - is this a pip bug? See http://qutebrowser.org:8010/builders/win8/builds/513/steps/unittests/logs/stdio for the full output. (it works on other platforms)
[19:55:23] <dunpeal> Is it normal for Python in a virtualenv to "fall back" on the global stdlib if it can't find a module in the virtualenv?
[20:00:23] <carljm> dunpeal: Only if the virtualenv was created with the --system-site-packages flag.
[20:00:30] <carljm> Or created with a very old version of virtualenv.
[20:00:52] <carljm> Or you're activating the virtualenv using "activate_this.py" or something similar, instead of by running the virtualenv's python binary.
[20:02:40] <dunpeal> carljm: I'm using virtualenv-3.4 12.0.5
[20:03:21] <carljm> That version should create isolated virtualenvs by default.
[20:04:00] <carljm> The other way Python in a virtualenv could appear to fall back on system packages is if you have PYTHONPATH (or maybe PYTHONHOME?) env vars set
[20:04:47] <dunpeal> carljm: I do have PYTHONPATH, but /usr/lib isn't in it.
[20:15:11] <dunpeal> carljm: btw, if virtualenv doesn't just copy the system Python and libs to env, where does it get them?
[20:16:27] <carljm> dunpeal: well, that's certainly interesting. no idea what might be causing that. is there a file lib/python-3.4/site-packages/no-global-site-packages.txt in the env?
[20:16:50] <carljm> dunpeal: it does a complex dance in its custom site.py to get the right paths added to sys.path
[20:17:21] <carljm> dunpeal: wait a second. I'm an idiot. cProfile is in the standard libary.
[20:17:32] <carljm> so the behavior you're seeing is entirely expected.
[20:17:53] <carljm> virtualenvs don't copy the standard library; they _always_ use the base Python's stdlib
[20:18:06] <dunpeal> carljm: why isn't cProfile import from the venv like re is?
[20:19:55] <carljm> the virtualenv's python starts up and finds a key marker file within the venv, making it think that the venv's lib/python-x.x is its stdlib
[20:20:10] <carljm> and adds the base Python's stdlib dir to sys.path
[20:20:30] <carljm> and (only if --system-site-packages was given when the env was created) also the base Python's site-packages dir
[20:20:31] <dunpeal> OK, now I'm confused. I thought the whole point of virtualenv was to isolate the Python executable and library from the system one?
[20:20:47] <carljm> dunpeal: no, the point of virtualenv is to isolate installed packages.
[20:21:03] <carljm> if you want to isolate _everything_, just compile Python with a custom --prefix and install it.
[20:21:27] <carljm> the point of virtualenv is to provide installed-packages isolation while not being as heavyweight as that (no need to copy entire stdlib)
[20:21:49] <dunpeal> carljm: is there a flag to copy the entire stdlib too?
[20:22:17] <carljm> no; that's not a use case virtualenv has ever intended to provide. there's no need - like I said, just install your own Python.
[20:23:12] <carljm> there really wouldn't be any difference between "virtualenv with entire stdlib copied" and "separate installation of Python"
[20:40:44] <_habnabit> so if i do `pip install -r /path/to/requirements.txt`, this seems to be interpreting a requirements.txt of just `-e .` as trying to install from the cwd, not the directory containing the requirements.txt. is there a better thing to put in the requirements.txt, or should i just force the cwd of pip to be the project root?
[20:46:24] <dunpeal> wow, _habnabit is asking a question.
[20:56:10] <carljm> _habnabit: that's too bad, "relative to requirements file" paths would be more useful. I don't know of a workaround; I always do the latter.
[21:28:04] <tos9> ionelmc: oh, no I think I've encountered lots of cases, not just one particular thing.
[21:28:35] <tos9> E.g. one that comes to mind is that I think distutils (and not pip) specifically throws an error message if you pass absolute paths to scripts=.
[21:29:07] <tos9> Saying exactly "don't pass absolute paths, pass them relative to the setup.py" or something