JuliaGaussianProcesses / KernelFunctions.jl

Julia package for kernel functions for machine learning
https://juliagaussianprocesses.github.io/KernelFunctions.jl/stable/
MIT License
266 stars 32 forks source link

Stop testing on 1.3 #420

Closed willtebbutt closed 2 years ago

willtebbutt commented 2 years ago

IIRC we agreed to stop testing on 1.3 once 1.6 became the LTS.

codecov[bot] commented 2 years ago

Codecov Report

Merging #420 (0f538d5) into master (04fa7f7) will not change coverage. The diff coverage is n/a.

Impacted file tree graph

@@           Coverage Diff           @@
##           master     #420   +/-   ##
=======================================
  Coverage   92.74%   92.74%           
=======================================
  Files          52       52           
  Lines        1199     1199           
=======================================
  Hits         1112     1112           
  Misses         87       87           

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 04fa7f7...0f538d5. Read the comment docs.

devmotion commented 2 years ago

Just my two cents: I think dropping tests but not upper bounding the Julia version will lead to silent breakage (had to fix too many of these in other packages...). Regarding compat bounds, I think there should be a clear motivation for dropping old Julia versions, and personally I don't think it's particularly useful to drop < 1.6 just because 1.6 is the new LTS if the package works fine with older Julia versions. However, e.g. I don't think we should spend time on tracking down and fixing bugs that are specific to older Julia versions or on making PRs compatible with these versions if it requires significant effort.

st-- commented 2 years ago

Whilst we proclaim compatibility with 1.3, we should keep running tests on 1.3. I think at this point we shouldn't do anything extra to maintain 1.3 compatibility, but as long as it's at no extra cost, we might as well keep it going. The change I'd like to see is more on the process level: once we've got a PR that fails on 1.3 (but works on 1.6+), then we should remove 1.3 compatibility (and drop the test).

willtebbutt commented 2 years ago

Sounds good to me.