Add Positron Variables Pane Methods

[t-kalinowski]

This PR adds support for reticulate objects in the Positron Variables Pane.

With an example session:

# needed because positron leaks an internal VIRTUAL_ENV env var into the R process
Sys.unsetenv("VIRTUAL_ENV")

library(reticulate)
library(keras3)

anint <- py_eval("1", FALSE)
adict <- py_eval("{'a':1, 'b':2}", FALSE)
adict['foo'] <- adict # recursive

alist <- py_eval("['a', 1]", FALSE)
atuple <- py_eval("('a', 1)", FALSE)

input <- keras_input(c(24, 24))

output <- input |>
  layer_dense(32, "relu") |>
  layer_dense(10, "softmax")

model <- keras_model(input, output)

We now see:

Screen.Recording.2024-11-11.at.3.12.40.PM.mov

Previously, all python objects presented as:

This uses the ark interface added in posit-dev/ark#560 and posit-dev/ark#561

[dfalbel]

LGTM!

[dfalbel]

I think we should limit the the length of i.get_children(), either by providing an argument from ark to ark_positron_variable_get_children or hardcoded here.

[t-kalinowski]

What do you think the limit should be? Would it be dynamic based on something? Maybe a range of which children are visible in the pane?

If we do something here, we should also include a companion PR to make sure that the Python inspectors.get_children() is in sync. https://github.com/posit-dev/positron/blob/70e96ec21df6dbb9027b1cb96749c4a461f87616/extensions/positron-python/python_files/positron/positron_ipykernel/inspectors.py#L146

[dfalbel]

I think the UI already has a limit of what is actually displayed. This is just a hard limit to avoid building the large object that is never shown anyway. I'd add something like 100. In the Python inspectors, since this is an iterable, I think it might respect whatever the UI really needs to read?

[t-kalinowski]

Do we have a standard way to indicate that not all elements are being shown? E.g., ...truncated or similar?

[dfalbel]

The python inspector does something like:

[t-kalinowski]

It looks like this is handled in the frontend: https://github.com/posit-dev/positron/blob/83d516172b17d2ae02da8b01411de646cb1117c9/src/vs/workbench/contrib/positronVariables/browser/components/variableOverflow.tsx#L87-L91

[t-kalinowski]

Added a python class ChildrenOverflow(), and corresponding methods so that more than 100 items are now presented as an overflow:

E.g.,

x = list(range(1000))

now shows:

[t-kalinowski]

@dfalbel (@lionel-?) Is there a better way to get the path to the Positron installation from R? I'm not sure if this is the most reliable way to resolve the path to the inspectors.py file that ships with Positron.

[t-kalinowski]

@lionel- We'll open a PR to Ark, exporting a function like .ps.positron_ipykernel_path, to return the path to the positron_ipykernel python module that ships with Positron.

[lionel-]

yes that would be reasonable. .ps stands for positron so .ps_ipykernel_path?

This function could query the positron-r extension for that path via UIComm RPC.