Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (05caf07) 94.99% compared to head (1f980ed) 95.06%.
Report is 1 commits behind head on master.

@@            Coverage Diff             @@
##           master    #1520      +/-   ##
==========================================
+ Coverage   94.99%   95.06%   +0.07%     
==========================================
  Files          45       44       -1     
  Lines        7551     7564      +13     
==========================================
+ Hits         7173     7191      +18     
+ Misses        378      373       -5     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@ogrisel the PR now proposes a fix to the coverage pipeline. The issue seems to have been that codecov and/or github integration do not support updating the github annotations dynamically with successive upload from parallel pipelines, but we had each pipeline individually upload their report. The first pipeline to reach codecov would trigger codecov annotations and those would not be cleared by subsequent updates, despite annotated code being actually covered by later pipelines. (I also suspect consecutive uploads do not merge correctly for python to begin with so this might have affected the final reports too ?)

I changed the script so that every pipeline creates a local artifact with its coverage report, and a single pipeline is triggered after all other are done that combines all coverage reports before issuing a single codecov upload.

That will also prevent receiving early e-mails from codecov because the codecov pipeline is temporarily failing because incomplete reports, despite later turning green with subsequent updates. It also fixes some unelegant codecov error messages in the pipelines.

While debugging I also found that the 3.8 distributed pipeline was not working because the pinned versions were uninstallable and caused pytest.importorskip("distributed") to just skip.

There were also issues when combining coverage reports from different pipelines because coverage reports stored absolute paths, and those diverged accross pipelines (macos / windows / linux absolute paths are different ?). Fixed now.