Handle multiple scopes in the debugger variables viewer

[jtpio]

References

Up until now the debugger would only display one scope at a time.

However after the switch to debugpy the variables explorer was displaying the same information twice without mentioning the name of the scopes:

This PR fixes the assumptions that variables come from the first scope:

jupyterlab/packages/debugger/src/service.ts

Line 640 in 781c615

const variables = await this._getVariables(scopes[0]);

Which might have been the case until now with ptvsd but does not apply anymore with debugpy.

Code changes

• Update the tree view
• Update the table view
• Handle optional evaluateName and type from the DAP response: https://microsoft.github.io/debug-adapter-protocol/specification#Types_Variable

User-facing changes

Before

Using the latest xeus-python with debugpy:

After

Since both xeus-python and VS code use debugpy, the variables viewer now shows "special variables" and "function variables" too:

EDIT: updated the screencast to show the latest status:

Backwards-incompatible changes

This is mostly a UI change and should still be compatible with the previous versions of xeus-python that rely on ptvsd.

[jupyterlab-dev-mode]

Thanks for making a pull request to JupyterLab!

To try out this branch on binder, follow this link:

[jtpio]

Updated the grid view to show the locals and globals at the same level:

Maybe there could be a dropdown to select the scope, which would update the grid and show variables for one scope at the time.

[jtpio]

Using a dropdown to switch between scopes:

[jtpio]

This should be ready for a first pass of review.

And can be tested on Binder (pre-built for 27b2246): https://mybinder.org/v2/gh/jtpio/jupyterlab/27b2246b021be3d3b2266b71f6e5d566629f62ef?urlpath=lab-dev

[jasongrout]

@jtpio - is this ready for 3.0? It does look like an important fix.

You asked for a first pass of review. Do you consider this PR fairly done, or do you see a lot of iteration and further work based on review comments?

[jtpio]

is this ready for 3.0? It does look like an important fix.

Yes, and the Binder should be pre-built to make it easier to try out: https://mybinder.org/v2/gh/jtpio/jupyterlab/debugger-scopes?urlpath=lab-dev

I think this solves the main issue of handling multiple scopes. Then more iteration could be done in follow-up PRs if we think the UX should be different or tweaked.

[afshin]

I am having trouble getting this branch to work. This is what I am seeing:

[jtpio]

@afshin this looks like the compatibility issue mentioned in #9414 (comment).

Which version of xeus-python are you using? On Binder it's 0.9.1.

[afshin]

  debugpy              1.2.0  py37hfbb9781_0  conda-forge/osx-64        2 MB
  ptvsd                4.3.2  py37h60d8a13_3  conda-forge/osx-64        5 MB
  xeus-python          0.9.1  py37ha661077_0  conda-forge/osx-64     1000 KB

These are the things I installed right before making that screencap.

Update: I am using Python 3.7.4

[jtpio]

Update: I am using Python 3.7.4

Would you be able to try with Python 3.8 in a new environment?

This breakpoint issue is probably unrelated to this PR, but might still need to be fixed in the debugger extension.

[afshin]

I am having trouble with this. In a brand new conda env with Python 3.8 I get the same error. And I cannot get the Binder link to load either.

[afshin]

Figures, literally as soon as I sent that message, the Binder finally worked!

[afshin]

@JohanMabille I finally got it working on my system when I installed xeus-python with pip. It does not work if i use mamba or conda.

[afshin]

Thank you @jtpio!

[afshin]

In a follow-on PR, the CSS which handles truncating long paths (and making sure the title "Sources" is not obscured) can probably use some tweaking.

[afshin]

Sorry for the long delay!

[JohanMabille]

@afshin On which platform did you experience the issue? (I bet on OSX, but just to be sure)

[jtpio]

@afshin On which platform did you experience the issue? (I bet on OSX, but just to be sure)

  debugpy              1.2.0  py37hfbb9781_0  conda-forge/osx-64        2 MB
  ptvsd                4.3.2  py37h60d8a13_3  conda-forge/osx-64        5 MB
  xeus-python          0.9.1  py37ha661077_0  conda-forge/osx-64     1000 KB

I think that's the case based on this output ^

[afshin]

Yes I was on macOS (Big Sur, 11.0.1)