reticulate integration fails with `ModuleNotFoundError: No module named 'positron_ipykernel'` for conda/mamba environments (works for venv)

System details:

Positron and OS details:

Positron Version: 2026.07.1 (user setup) build 5
Code - OSS Version: 1.118.0
Commit: d3cdd608df5afa3378c24c9a9c81b01fe1433a87
Date: 2026-07-08T19:52:09-06:00
Electron: 39.8.8
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.19045

Session details:

R 4.6.1, Python 3.10.12 (system), Python 3.12.13 (conda env)
Reticulate version 1.46.0
Quarto version 1.9.38

Describe the issue:

Positron's reticulate integration fails to start the Python session for conda/mamba environments, exiting immediately with ModuleNotFoundError: No module named 'positron_ipykernel'. The identical flow works for a venv on the same machine. This makes the managed "Python (reticulate)" console unusable with conda environments.

Steps to reproduce the issue:

1. Create a fresh conda/mamba environment:

   mamba create -n testenv python=3.10
   

2. In Positron, open a .qmd file and run:

   library(reticulate)
   use_condaenv("testenv", required = TRUE)
   repl_python()
   

3. Observe that the reticulate Python session fails to start.

For contrast, the same flow against a venv succeeds:

# from a workspace containing a .venv created via `python -m venv .venv`
library(reticulate)
use_virtualenv(".venv", required = TRUE)
repl_python()   # starts normally

Expected or desired behavior:

The "Python (reticulate)" console starts against the conda environment, the same way it does for a venv.

Were there any error messages in the UI, Output panel, or Developer Tools console?

The session exits immediately (exit code 0) with:

Python (reticulate) failed to start up (exit code 0)
Reticulate initialization failed: ModuleNotFoundError: No module named 'positron_ipykernel'
Run `reticulate::py_last_error()` for details.

reticulate::py_last_error() R traceback:

── R Traceback ────────────────────────────────
     ▆
  1. └─.ps.rpc.reticulate_start_kernel(...)
  2.   ├─base::tryCatch(...)
  ...
 11.             └─reticulate::import("positron_ipykernel.positron_ipkernel")
 12.               └─reticulate:::py_module_import(module, convert = convert)

- Open the Developer Tools console by running the `Developer: Toggle Developer Tools` command from the Command Palette
- Open the relevant Output Channel logs by utilizing this guide: https://positron.posit.co/troubleshooting.html

## What I've already tried (all still fail for the conda env)

- Installed `ipykernel` into the conda env (`mamba install -n testenv ipykernel`) — no change; the missing module is `positron_ipykernel`, not `ipykernel`.
- Pinned `ipykernel<7` — no change.
- `use_condaenv("testenv", required = TRUE)` — the original failing path.
- `use_python("/home/<user>/miniforge3/envs/testenv/bin/python", required = TRUE)` — same failure. reticulate appears to still apply conda-style setup because the prefix contains a `conda-meta` folder.
- Confirmed the same machine/WSL works fine with a venv, so this is specifically conda-env vs venv rather than a filesystem/WSL issue (the venv is on `/mnt/d`, the conda env on the native Linux filesystem).

Workaround that does work: setting `positron.reticulate.enabled` to `never` (loses the managed console and Variables pane).

## Hypothesis (may be wrong)

Positron exposes `positron_ipykernel` by adding its bundled kernel directory to the interpreter's search path via `PYTHONPATH`. Conda environment activation seems to rebuild the process environment and drop that `PYTHONPATH` entry, so the bundled module is no longer importable for conda interpreters. venvs don't go through that activation step, so the entry survives — which would explain the venv-vs-conda difference.

Start at .ps.rpc.reticulate_start_kernel and reproduce the conda/mamba and venv flows described in the issue, then inspect how the Python path is preserved during reticulate initialization. Verify the fix by confirming positron_ipykernel imports and the managed "Python (reticulate)" console starts in the conda environment without regressing venv support.