PR: Pass new environment variable to the kernel and remove some benign errors thanks to debugpy 1.6.0 (IPython console)

[mrclary]

Description of Changes

• Removed relevant errors from the benign error filter in ipythonconsole plugin
• Added the environment variable PYDEVD_DISABLE_FILE_VALIDATION to IPython Console

The advent of debugpy=1.6.0 provided an environment variable to bypass the errors encountered in #16927, #16828, and #17552.

Currently, only 1.5.2 is available on conda-forge; when 1.6.0 is available, I'll mark as ready for review.

Issue(s) Resolved

Fixes #17552

Affirmation

By submitting this Pull Request or typing my (user)name below,
I affirm the Developer Certificate of Origin
with respect to all commits and content included in this PR,
and understand I am releasing the same under Spyder's MIT (Expat) license.

I certify the above statement is true and correct: @mrclary

[mrclary]

@ccordoba12 @dalthviz should we also require debugpy>=1.6.0? We don't have an explicit requirement on this right now, so I'm not sure what we want to do if users still have 1.5.1.

[mrclary]

@ccordoba12, is there a reason why we are not testing Python 3.9 and conda for macOS? Just curious.

[ccordoba12]

Because Github only offers 5 Mac Github Action slots per organization on the free plan, so we can't spend them all in Spyder.

I think an option could be to run our tests only in Python 3.9 (instead of 3.7 and 3.9), so we can reduce the amount of slots we use per pull request. We'll take a look at that after 5.3.0 is released.

[mrclary]

Because Github only offers 5 Mac Github Action slots per organization on the free plan, so we can spend them all in Spyder.

I think an option could be to run our tests only in Python 3.9 (instead of 3.7 and 3.9), so we can reduce the amount of slots we use per pull request. We'll take a look at that after 5.3.0 is released.

I see, thanks for the explanation. So that means we use 4 right now, correct? 3.7 pip slow, 3.7 pip fast, lite build, and full build? I have been considering for a while whether we really needed to make Lite builds for every pull request. If we limited pull requests to just the full builds, that may help; on releases, we can have both full and lite. But as you say, we can discuss after 5.3.0.

[ccordoba12]

So that means we use 4 right now, correct? 3.7 pip slow, 3.7 pip fast, lite build, and full build?

Right.

I have been considering for a while whether we really needed to make Lite builds for every pull request

Well, the nice thing about the Lite builds is that they are like half-size, so faster to download, install and test. I use the Windows Lite one all the time.

[ccordoba12]

Thanks @mrclary for your help with this!

[dalthviz]

@ccordoba12 @dalthviz should we also require debugpy>=1.6.0?

If the env var is available since debugpy 1.6.0 then I would say so. Probably the requirement should be done over spyder-kernels, right? and then we change the minimum version for spyder-kernels over Spyder before doing the release.

[mrclary]

Upon further consideration, I don't think we should require debugpy>=1.6.0. First, 1.6.0 is not yet available on conda-forge, so we cannot set the requirement until it becomes available. Second, while it is true that the environment variable is not recognized by debugpy<1.6.0, it will do no harm there either. Third, the particular benign errors addressed are exclusive to the standalone applications for which 1.6.0 will be the default version anyway. In short, it doesn't matter which version users have installed because the only contexts that matter (standalone apps) will use 1.6.0, so I don't see a need to add another requirement to either spyder or spyder-kernels.

[ccordoba12]

Third, the particular benign errors addressed are exclusive to the standalone applications for which 1.6.0 will be the default version anyway. In short, it doesn't matter which version users have installed because the only contexts that matter (standalone apps) will use 1.6.0, so I don't see a need to add another requirement to either spyder or spyder-kernels.

Right, that's what I thought too (and sorry for letting that unanswered, I just missed the question).