Add Robot Framework language server (again)

[bollwyvl]

References

• fixes Support/test a Robot Framework language server #332
• supersedes [wip] Add Robot Framework language server #493

Code changes

• refreshes robot dependencies
• adds new robot stuff
  • robotkernel
  • robotframework-lsp
  • robotframework-robocop
  • robotframework-robotidy
  • robotframework-pabot (interested to see the intersection of this with other tools)

    • took this out: couldn't get this to work reliably w/r/t to the language servers... there might be some env-singleton resource issues or something

  • jupyterlab-robotmode
• updates language server docs
  • adds robotframework to the "notebook" section (it's pretty good)
  • generate the shell command for our list of npm ones
    • decide if we want to also list the version we tested?
      • yaml, ts, etc. have bitten us in weird ways in the past, even on point releases
  • some sorting
• adds overrides.json to the binder

User-facing changes

• visitors to the binder will see more robot-related features
• visitors to the binder will see more latex-related features (linting)

Backwards-incompatible changes

• n/a

Chores

• linted
• tested
• documented
• changelog entry

[github-actions]

👈 Launch a binder notebook on branch bollwyvl/jupyterlab-lsp/gh-332-add-rflsp-again

[bollwyvl]

Still need to test in Configuration: once robocop is turned on, underlines don't show up, even though things appear in the diagnostic panel...

[bollwyvl]

underlines don't show up

Ah, all of the diagnostics I've seen have zero width! Really surprised me!

I guess some options (not to be fixed on this issue):

• gutter (we've wanted that anyway, was hoping to wait for CM6)
• offer expanding zero-width diagnostics to the full line

[krassowski]

offer expanding zero-width diagnostics to the full line

or to 1 character? I would like this regardless of gutters.

[bollwyvl]

Sure, perhaps expandZeroWidthDiagnostics with a default of 0, preserving existing behavior, while 1 would be one character, and -1 could go to the end of the line.

[krassowski]

If that becomes a setting, I happy for it to have a non-zero default.
Edit: (zero would be fine too, but it feels like non-zero would be better default)

[krassowski]

Is it still WIP, or can we merge?

[bollwyvl]

I'll merge upstream and take another look, but probably good to go, no substantive further changes planned...

[bollwyvl]

I've

• removed the WIP
• hoisted the diagnostic idea over to Improve visibility of under-described diagnostics #743

If this passes, it's probably ready to go.

[bollwyvl]

Thanks for the binder assist, trying right now. I put the user check in because i test (the rest of) postBuild as part of my local development flow, but having too much fancy stuff configured makes it harder to test... are there any other things we'd want to pre-configure, i guess? Maybe checktex or whatever cuteness it is?

[bollwyvl]

Ugh, the "magic" rstudio stuff on binder is brutal. It looks like our initial solve pulls in 4.1, but then the hard pins bump it back down to 4.0, and then pip installs a bunch of weird stuff for shiny from github tarballs. Gross. I'll have to defer to r users whether this is all worth it, or if we should remove the r stuff from the "main" binder environment, and install just our r-related stuff (without rstudio and shiny server i guess)... conda-forge has made overtures of collaboration with the folk that run that ecosystem and they pretty much are defending their distribution mechanism.

Additionally: it looks like something we're pulling in is forcing us down to jupyterlab 3.1.x, which is good for testing, i guess, but poor for picking up more recent features and bug fixes. It also means that the retrolab we're getting is fairly outdated at this point... I guess that's academic, as we'll want to start testing with lab4 and nb7 soon enough on another branch, and should keep what we've got working today working as long as we can.

[bollwyvl]

Ok, had a go at it on binder. Everything seems mostly working except:

• third-party keyword libraries (e.g. SeleniumLibrary) doesn't seem to resolve, which doesn't happen locally, and potentially something we can investigate at a later date
• completion (both from the kernel and the lsp) seems to be busted

Note that i can't reproduce these issues locally. We could put completion under test, but once again, at some point there's only so much we can control.

conda-lock for reference: jupyter-lsp-binder.conda.lock.txt

[krassowski]

completion (both from the kernel and the lsp) seems to be busted

works for me on binder; the problem might be that it is case sensitive by default; we might want to allow servers specs to override the defaults (and have a default option for caseSensitive set to "auto" which means let server decide). Similarly for SQL we would want case-insensitive completion.

[krassowski]

LGTM, thanks!