Travis stalling building plots on julia 1.0 but not 1.1

travis
#1

I have a package testing which fails with a weird error on julia 1.0 but not on 1.1. The error is occuring while building Plots.jl:

No output has been received in the last 10m0s, this potentially indicates a stalled build or something wrong with the build itself.

The package build log is here
https://travis-ci.org/baggepinnen/MonteCarloMeasurements.jl/jobs/521297177
and the problem occurs on both linux and osx builds.
One thing I think could have caused this is that the last release of the package declared compatibility with Julia 1.1, but I have now made it compatible with julia 1.0 and declared that.
Anyone familiar with this? Is there something I have to do, like increase version number in a special way, when I “de-bump” the compatibility requirements for the package?

0 Likes

#2

I’m still confused about this. Other packages seem to build the same version of Plots well, e.g.,
https://travis-ci.org/JuliaControl/ControlSystems.jl/jobs/521126813#L593
The lack of any form of error output makes it tricky to debug.

0 Likes

#3

Does this keep happening? I assume you have tried to restart the build?
Building Plots might be a red herring here, are you sure that step is not actually finished and the problem is the next steps? What if you add a print statement first thing in the test/runtests.jl file to see if you get past the build step? You can also poke around with ssh access: https://docs.travis-ci.com/user/running-build-in-debug-mode/ .

0 Likes

#4

It keeps happening, have tried 5 times over two days. I can try with the print statement, but tests pass rapidly on julia v1.1 and nightly and I see no reason as to why they should take so long time on julia v1.0

Edit: You were right. Plots did build, and it is getting stuck at something else. I guess I have to get a version of Julia v1.0 on my local machine and try out. Maybe compilation time is extremely long on v1.0 for my StaticVector operations or something.

0 Likes

#5

So it turns out that compilation time was extremely long for some operations, causing the tests to take more than 10 minutes. Adding the print statements helped as travis received some output making the test go on for longer. The real solution though was to reduce the dimensions on the static vectors, making everything slightly faster, in addition to some status printouts every now and then. Thanks!

0 Likes