fix(python): launch phy and suite2p through uv instead of .BAT wrappers

[tabedzki]

🤖 Human guided, AI assisted PR (using this skill). AI text below. 🤖

🤖 AI text below 🤖

Stacked on #2 — targets feat/uv-python-env, not main.

The bug

Clicking Open Phy on macOS failed with:

zsh:1: permission denied: PythonScripts/open_phy.BAT

open_phy.BAT and open_suite2p.BAT are cmd.exe batch files, so MATLAB's system() handed them to zsh, which tried to execute a non-executable Windows file. Setting the execute bit would not have helped either — @echo off and call conda activate are not shell commands.

Both wrappers also assumed phy and suite2p lived in a conda env named on the command line (iblenv). That does not hold in practice: on the reporting machine phy is installed in a separate phy2 env, and iblenv has no phy at all.

The fix

Replace both .BAT wrappers with standalone launchers that declare their dependencies inline (PEP 723) and run under uv — which the app already locates at startup for the parameter helper scripts:

uv run --no-project PythonScripts/open_phy.py <sorting_output_dir>

• PythonScripts/open_phy.py — phy==2.1.0
• PythonScripts/open_suite2p.py — suite2p[gui]==1.1.0

Both are pinned from PyPI, so neither tool needs conda now. They are deliberately kept out of the root pyproject.toml: phy pins numpy<2 and pulls its own Qt stack, which cannot coexist with the >=3.14 helper-script environment. uv resolves each into its own cached environment on first use.

OpenExtGUI.m now builds the uv command from a new app.py_uv, set alongside app.py_env in getPythonEnv.m, and reports a clear message when uv is unavailable rather than launching nothing.

Cross-platform uv discovery

findUv never assumed a fixed path, but MATLAB's system() inherits a minimal PATH, which was the real failure mode. On unix it now probes $SHELL -l -c first, sourcing the user's profile:

minimal PATH, plain probe:        NOT FOUND
minimal PATH, login-shell probe:  /opt/homebrew/bin/uv

Fallback install locations now also cover linuxbrew, /usr/bin and /snap/bin.

Drive-by bug fix

In OpenExtGUI, when system() threw, out was never assigned, so the following if out ~= 0 raised its own error and masked the real failure. The exit-status check moved inside the try block.

Verification

• phy 2.1.0 and suite2p 1.1.0 both install and import under uv on macOS; phy --help lists template-gui
• The exact MATLAB-style command runs under a stripped env -i environment
• Paths containing spaces parse correctly (phy gets past params.py into array loading)
• find_stat resolves all three suite2p nesting levels; an empty dir opens the GUI empty
• checkcode on all three modified .m files reports 0 syntax errors (the one remaining lint is a pre-existing unused event callback argument)

suite2p backwards compatibility

Confirmed, since it decided the pin. A legacy-format plane0 directory (ops/stat/F/Fneu/spks/iscell written with allow_pickle) was loaded through suite2p 1.1.0's real GUI entry point, suite2p.gui.io.load_files. All arrays returned correctly, and a missing optional redcell.npy degraded to "no channel 2" rather than erroring. Pinning the current release does not strand existing results.

Not included

• OpenExtGUI2 (IBL atlas GUI) still uses the iblenv conda env — migrating iblapps-master is a larger change. Note it is currently broken on macOS for a related reason: getCondaEnvPython matches the Name column of conda env list, which is blank when envs live outside the standard envs dir, so app.py_ibl_env resolves to [].
• Neither GUI window was launched to confirm it paints, as no real sorting output was available. The launch path is verified up to the point where phy loads data.

Assisted-by: ClaudeCode:claude-opus-5

🤖 Generated with Claude Code