Install dependencies through pip

#1

Hi,

I’ve been preparing my package for Julia 1.0. Nearly finished but I am left with one problem: I had a dependency installed through pip via a build.jl script, it was actually working well in Julia 0.6. Now it seems that it does not work anymore… at least if does not build properly on Travis.

I’ve search the forum, there is a old threat from early 2017 that advised doing a conda-forge build. I know this will be a very nice solution because it will actually allow me to get rid of Fortran compilation problems in Python too. But I have absolutely no experience and also no time for doing that…

Does anyone have any insights of how I could make that work again? (or even is experience with conda-forge and willing to give it a try for gcvspline?)

Thanks in advance!
C.

#2

Why not just use Conda.jl to install it via conda-forge?

using Conda
Conda.add_channel("conda-forge")
Conda.add("spectra")
#3

Here’s the relevant Conda issue for installing dependencies with pip, at least.

#4

It looks like gcvspline is not in conda-forge so I guess you need:

using Conda
Conda.add_channel("charlesll")
Conda.add("gcvspline")

(see https://github.com/charlesll/gcvspline#installation)

Not sure why your build script does not work but comments on:

If PyCall.python is installed via conda, maybe it’s safer to…

  • upgrade pip and setuptools using conda
  • not use --user
#5

Thanks!

Your first solution will unfortunately not work on all systems because I only added a Linux 64 bit Conda version a long time ago for gcvspline (I am the maintainer of gcvspline)… I need to work on doing a conda-forge build but I do not have the time to do that now.

I will try the second thing and see if it solves the problem for now.

#6

Because it’s not on conda-forge but on PyPI. (Also the dependency is gcvspline, not Spectra that is the Julia library)

#7

Oops. Sorry, I didn’t notice your username. Don’t mean to rude :slight_smile:

#8

No worries it was not that explicit :slight_smile: