'Tcl wasn't installed properly' on python test workflow

[mscheltienne]

I got a couple of workflow running on a matrix of (windows, macOS, ubuntu) with python (3.7, 3.8, 3.9, 3.10). Randomly, some of the windows tests will fail when trying to show matplotlib figures with the error message:

self.tk = _tkinter.create(screenName, baseName, className, interactive, wantobjects, useTk, sync, use)
E       _tkinter.TclError: Can't find a usable init.tcl in the following directories: 
E           {C:\hostedtoolcache\windows\Python\3.7.9\x64\tcl\tcl8.6}
E       
E       C:/hostedtoolcache/windows/Python/3.7.9/x64/tcl/tcl8.6/init.tcl: couldn't read file "C:/hostedtoolcache/windows/Python/3.7.9/x64/tcl/tcl8.6/init.tcl": No error
E       couldn't read file "C:/hostedtoolcache/windows/Python/3.7.9/x64/tcl/tcl8.6/init.tcl": No error
E           while executing
E       "source C:/hostedtoolcache/windows/Python/3.7.9/x64/tcl/tcl8.6/init.tcl"
E           ("uplevel" body line 1)
E           invoked from within
E       "uplevel #0 [list source $tclfile]"
E       
E       
E       This probably means that Tcl wasn't installed properly.

Any idea how to solve this and make it work 100% of the time? It looks to me like Python is not always correctly setup by:

- task: UsePythonVersion@0
  inputs:
    versionSpec: '$(python.version)'
  displayName: 'Use Python $(python.version)'

For x-ref: vferat/pycrostates#41

[hodgestar]

We see the same in QuTiP's builds on Windows, e.g. https://github.com/qutip/qutip/runs/7696687924. It's mysterious to me how its possible that only some workers don't have Python Tcl installed correctly.

[janosh]

Having the same problem over at https://github.com/janosh/pymatviz (CI run). Curious if anyone has a hypothesis why this is flaky.

[janosh]

This discussion suggests setting the matplotlib backend prior to plotting helps:

jobs:
  tests:
    strategy:
      matrix:
        os: [ubuntu-latest, windows-latest]
    runs-on: ${{ matrix.os }}
    env:
      MPLBACKEND: Agg  # https://github.com/orgs/community/discussions/26434
     ...

[wpbonelli]

similar solution worked for me @janosh, resolved Tcl-related flakiness on mac by selecting a non-interactive backend

[kirill-ivlev]

@mscheltienne it seems that it was fixed in the 3rd party library, and was not directly related to the task itself.
Please feel free to re-open the issue if it still exits.

[leoniewgnr]

I encounter the same problem in our test pipeline: https://github.com/ourownstory/neural_prophet/actions/runs/5626902294/job/15248585543?pr=1377
What can I do to resolve this?