If tests pass with this (and Plots are able to build in makedocs) I think it's a great fix. If the second doesn't work we could probably just add the xvfb to the makedocs part.

This is an automated message.
Plots were compared to references. No changes were detected.

Codecov Report

Merging #401 (077e75f) into master (2fd813d) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #401   +/-   ##
=======================================
  Coverage   82.40%   82.40%           
=======================================
  Files          31       31           
  Lines        2824     2824           
=======================================
  Hits         2327     2327           
  Misses        497      497           

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 2fd813d...077e75f. Read the comment docs.

I was also wondering if there is a reason for having these settings spread out over three files? Would it not be enough to set it in runtests.jl and we could skip the rest?

If people want to run make.jl or test_plot.jl by itself I think it is fine to assume they can know enough to set the plotting environment themselves?

And the one in .travis.yml seems like it is not really necessary at all if it is no matter set in the files where it matters anyway?

What is PLOT_TEST even used for? Could only find it in three places in the repo, and that was the three places it was set.

I was also wondering if there is a reason for having these settings spread out over three files? Would it not be enough to set it in runtests.jl and we could skip the rest?

If people want to run make.jl or test_plot.jl by itself I think it is fine to assume they can know enough to set the plotting environment themselves?

And the one in .travis.yml seems like it is not really necessary at all if it is no matter set in the files where it matters anyway?

Even though I know how to do it, it is nice to be able to run the scipts locally whithout editing them. But in general it is needed in test_plots.jl so you can run
]test ControlSystems
locally, even on headless-server. In docs/make.jl so you can run makedocs(modules=[ControlSystems]).
Maybe it can be removed in travis.yml if it is included in those files.

What is PLOT_TEST even used for? Could only find it in three places in the repo, and that was the three places it was set.

This is read by Plots.jl. For example it is used in Plots.jl tests: https://github.com/JuliaPlots/Plots.jl/blob/2774b155b4c2b70db3fefd80bd623c34afed992f/test/runtests.jl#L100
It seems to be related to keeping the performance constant when there is no display information: https://github.com/JuliaPlots/Plots.jl/pull/1318/files
Edit: I added it for some reason, but I don't remember exactly.

Even though I know how to do it, it is nice to be able to run the scipts locally whithout editing them. But in general it is needed in test_plots.jl so you can run
]test ControlSystems
locally, even on headless-server. In docs/make.jl so you can run makedocs(modules=[ControlSystems]).
Maybe it can be removed in travis.yml if it is included in those files.

Okay, I was thinking that you might want to run it with other settings if you are not on a headless server but I realise that we probably prefer consistency in test/docs as the default and if you really want to run it in some other way you can dig in there and change it.

This is read by Plots.jl. For example it is used in Plots.jl tests: https://github.com/JuliaPlots/Plots.jl/blob/2774b155b4c2b70db3fefd80bd623c34afed992f/test/runtests.jl#L100
It seems to be related to keeping the performance constant when there is no display information: https://github.com/JuliaPlots/Plots.jl/pull/1318/files
Edit: I added it for some reason, but I don't remember exactly.

Okay, seems good as it is then.

I also tried to run the tests without our travis settings on my headless workstation and it seems to work fine so hopefully this could solve the PkgEval problem.

This was solved in #408