Add Restart Server button to dev-mode Setup section by dsarno · Pull Request #406 · hi-godot/godot-ai

dsarno · 2026-05-07T14:56:51Z

Summary

Adds a dedicated Restart Server button to the dock's Setup section, visible whenever ClientConfigurator.is_dev_checkout() is true and Developer mode is on. Sits alongside the existing Start/Stop Dev Server button.
Click routes through new Plugin.force_restart_server_preserving_mode() which picks the managed (_lifecycle.force_restart_server) vs --reload dev (stop_dev_server + _wait_for_port_free + start_dev_server) restart path based on what's currently running, so contributors keep Python auto-reload across the kill+respawn.
While the click is in flight, the button shows Restarting… disabled, mirroring the existing _crash_restart_btn / _version_restart_btn feedback pattern. Tooltip explains "Killing the current server and respawning…".
Disabled with an explanatory tooltip when no server is running on port 8000.

Why

Same-version Python edits get adopted as compatible by server_lifecycle.gd::start_server so neither the version-drift _version_restart_btn nor the spawn-failure _crash_restart_btn surfaces — reloading the plugin or relaunching the editor just re-adopts the stale running server. The existing Start/Stop Dev Server button toggles modes ("Switch to dev mode" / "Exit dev mode" / "Start dev server"), which doesn't match the "kill whatever is running and respawn from current source" semantics contributors actually need when editing Python without bumping the version.

Implementation notes

_update_dev_section_buttons() does a single has_managed_server / is_dev_server_running discovery per _update_status tick and applies state to both the dev-server toggle and the new Restart button. Avoids the duplicate lsof/ps fork Copilot flagged.
force_restart_server_preserving_mode() waits for the port to free between stop_dev_server and start_dev_server — without this, uvicorn's --reload parent reloader can hold the port past OS.kill and the new spawn races the old shutdown.
CLAUDE.md: reworded the "worktree-safety" headline to ban only another session's worktree, matching the nuanced exception two paragraphs below ("use your own worktree, commit frequently").

Test plan

ruff check src/ tests/ — clean
pytest -q — 899 passed
godot --headless --import — no GDScript parse errors
New unit tests for _restart_server_btn_state (3-combination truth table)
New tests in test_dock.gd covering rendering, visibility gating, dispatch, enabled/disabled state, in-flight "Restarting…" state
Live click smoke on the worktree editor — log confirms stop_dev_server → _wait_for_port_free → start_dev_server with --reload (PID changes, new server picks up source edits without a version bump)
CI green on all 12 checks (Python 3.11/3.13, Godot Linux/macOS/Windows, Release smoke ×3, Game capture smoke ×3, close-linked-issues)

🤖 Generated with Claude Code

codecov · 2026-05-07T15:00:20Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

Copilot

Pull request overview

Adds a dev-checkout-only “Restart Server” control to the dock’s Setup section to let contributors force-respawn the currently-running server (managed vs --reload dev server) without toggling dev mode, plus tests to lock in the new button state/visibility/dispatch behavior.

Changes:

Add GodotAiPlugin.force_restart_server_preserving_mode() to restart whichever server is currently holding the HTTP port while preserving managed vs --reload mode.
Render and wire a new “Restart Server” button in mcp_dock.gd (dev checkout + Developer mode), with enable/tooltip state derived from managed/dev-running detection.
Extend unit tests in test_dock.gd and test_dock_dev_server_btn.gd to cover visibility, dispatch, and the _restart_server_btn_state truth table.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
test_project/tests/test_dock.gd	Adds integration-style dock tests for the new Restart Server button (rendering, visibility gating, dispatch, enabled/tooltip behavior).
test_project/tests/test_dock_dev_server_btn.gd	Adds unit tests for `_restart_server_btn_state` truth table.
plugin/addons/godot_ai/plugin.gd	Adds `force_restart_server_preserving_mode()` to restart the active server while preserving managed vs `--reload` mode.
plugin/addons/godot_ai/mcp_dock.gd	Adds the Restart Server button in dev-mode Setup UI plus state/update/pressed handling helpers.
CLAUDE.md	Updates developer documentation around worktree/Godot launch safety guidance.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

dsarno · 2026-05-07T15:17:28Z

+	if is_dev_server_running():
+		stop_dev_server()
+		start_dev_server()
+		return true


Good catch — already addressed in c0df1b8. The dev branch of force_restart_server_preserving_mode now does stop_dev_server() → _wait_for_port_free(ClientConfigurator.http_port(), 5.0) → start_dev_server(), mirroring the managed _lifecycle.force_restart_server path.

Kept the wait inline rather than pushing it into stop_dev_server() so existing callers (the "Exit dev mode" path in _on_dev_server_pressed) keep their non-blocking behavior — only the explicit kill+respawn flow needs to block on port release.

dsarno · 2026-05-07T15:17:30Z

 	_update_dev_server_btn()
+	_update_dev_restart_btn()


Fixed in 6ba9ff8. Added _update_dev_section_buttons() which scans (has_managed, dev_running) once and applies state to both buttons. The per-frame _update_status path now calls only the aggregator; same for the click-deferred refreshes in _on_dev_*_pressed.

The per-button _update_dev_server_btn / _update_dev_restart_btn helpers stayed for test isolation (_seed_dev_restart_btn swaps in a mock plugin and exercises just the restart button without the server-toggle button needing to exist).

Same-version Python edits get adopted as compatible (server_lifecycle.gd adoption arm) so neither the version-drift restart nor the spawn-failure restart surfaces. Contributors editing Python source had no UI affordance to force a respawn — reloading the plugin or relaunching the editor just re-adopted the stale running server. Adds a dedicated Restart Server button next to Start/Stop Dev Server, visible whenever is_dev_checkout() is true and Developer mode is on. Click routes through new force_restart_server_preserving_mode() which picks managed (existing _lifecycle.force_restart_server()) vs --reload dev (stop_dev_server() + start_dev_server()) based on what's running, so the user keeps Python auto-reload across the kill+respawn. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The headline "Always launch Godot from the root repo's test_project/" contradicted the nuance two paragraphs down ("use your own worktree" and "multi-editor is fine when you need the PR's live plugin code"). The intent was always "never point Godot at a vanish-prone worktree", not a blanket ban on worktree launches. Reworded to match the actual intent so the right exception (this session's own worktree, with frequent commits) doesn't read as a violation of the leading rule. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`_dock` is typed as `Node` in this suite, so property chains through `_dock._dev_restart_btn.disabled` / `.tooltip_text` resolve to Variant and `:=` can't infer. CI Linux/macOS/Windows Godot tests all failed to load the suite as a result. Replace `:=` with explicit `: bool` / `: String` annotations on the four locals the parser flagged plus the nearby int local for consistency. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Three actionable findings from the parallel reuse / quality / efficiency agents: 1. Real bug — efficiency review caught a missing wait_for_port_free in force_restart_server_preserving_mode's dev branch. stop_dev_server() issues OS.kill and returns, but the listener can take >500ms to release the port; start_dev_server's fixed 500ms timer then races the old uvicorn shutdown and the --reload spawn fails to bind. Mirrored the managed path's wait via _wait_for_port_free. 2. Trim doc comments restating WHAT per CLAUDE.md "default to no comments" — kept only the WHY (the adoption-arm trap). Drops ~17 lines across mcp_dock.gd and plugin.gd. 3. Extract _seed_dev_restart_btn / _cleanup_dev_restart_btn helpers in test_dock.gd to match the file's existing _seed_server_row / _cleanup_server_row pattern. Cuts boilerplate across the three new dispatch/state tests. Also drops a no-op bool() cast on a value already typed bool. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Per Copilot review on PR #406: _update_dev_server_btn and _update_dev_restart_btn each called _plugin.has_managed_server() and _plugin.is_dev_server_running() independently. The latter shells out via lsof / ps in plugin.gd::is_dev_server_running, so a single _update_status pass forked two subprocesses where one would do. Add _update_dev_section_buttons() that scans once and applies to both buttons. Used from _update_status (per-frame) and the click-deferred refreshes in _on_dev_server_pressed / _on_dev_restart_pressed. The per-button helpers stay so tests that swap in mocks via _seed_dev_restart_btn / _seed_server_row keep their isolated entry points. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

1. Fix Restart/Start label bug. After clicking the primary button, the newly-spawned --reload server's LISTEN holder is uvicorn's `multiprocessing.spawn` worker — that PID's cmdline doesn't contain the godot_ai brand, so `is_dev_server_running()` returned false and the label stayed "Start Dev Server" instead of flipping to "Restart Dev Server" once the server was up. Walk up the parent chain (bounded depth 5) inside `_pid_cmdline_is_godot_ai` so a worker whose parent reloader IS branded gets recognised. Adds `_pid_parent` helper covering POSIX (`ps -o ppid=`) and Windows (PowerShell CIM). 2. Drop the dock's "Mode override" dropdown. The dropdown advertised toggling between the github-repo local plugin and the installed ZIP version, but it didn't actually do that — it just nudged `is_dev_checkout()` for the update banner, which is more usefully driven by the GODOT_AI_MODE env var (still supported, still tested in test_clients.gd). Removes ~50 lines of dead UI plumbing and the stale "fresh check paints over a clean slate" comment. 3. Delete the orphan live_demo test file from a debugging session that never landed cleanly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…parent-walk (#408) * Show "Restarting…" feedback while dev-mode restart is in flight Live smoke uncovered a 5s editor freeze (from _wait_for_port_free) with no acknowledgement of the click — the user wonders if the button did anything. Mirror the existing _crash_restart_btn / _version_restart_btn pattern: set _server_restart_in_progress on click, dispatch via call_deferred + a 0.15s paint window, then clear the flag after the spawn lands. _update_dev_section_buttons checks the flag and renders "Restarting…" + disabled while in flight. Test path is non-tree, so _on_dev_restart_pressed runs synchronously when is_inside_tree() is false — keeps the existing dispatch test green without needing await scaffolding. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Consolidate dev-section to one Restart Dev Server + small ✕ stop Per PR review feedback: two restart-flavored buttons in the same section felt redundant, and the new Restart Server stayed disabled when only an adopted external dev server was running because brand detection misses orphaned multiprocessing.spawn workers. Replaces both _dev_server_btn (3-state Switch/Exit/Start toggle) and _dev_restart_btn (kill+respawn preserving mode) with a single primary button + a small stop affordance: - _dev_primary_btn — full-width "Restart Dev Server" / "Start Dev Server" (label adapts to whether something's running). Always enabled. Click → force_restart_or_start_dev_server which kills any godot-ai server (managed or dev), waits for the port to free, and lands in --reload mode. Same "Restarting…" feedback during the blocking wait. - _dev_stop_btn — compact "✕" beside the primary. Enabled only when a dev server is actually running. Click → stop_dev_server (existing). Intentionally never targets the managed server (lifecycle owns that). Plugin's force_restart_server_preserving_mode is renamed/retargeted as force_restart_or_start_dev_server. The previous "preserve managed mode" semantics drop — clicking Restart now always lands in dev mode, since that's the only mode contributors editing Python source actually want. The lifecycle's force_restart_server stays for the version-drift / crash recovery paths that legitimately preserve managed mode. One verb, one mental model. Adapted tests cover the truth table. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Make force_restart_or_start_dev_server cover orphan workers Previous version branched on has_managed_server / is_dev_server_running, both of which miss orphaned multiprocessing.spawn uvicorn workers — when the parent reloader dies (or the editor is restarted while the worker keeps holding the port), the worker's cmdline is just "spawn_main" without godot_ai branding, so neither detection path fires. Result: clicking Restart skipped the kill and went straight to start_dev_server, which then failed to bind :8000. Switch the kill arm's condition to _is_port_in_use(port). The user clicking Restart is explicit consent to take over the port, so killing whoever's there (regardless of brand) is the right semantics. Lifecycle state still resets when has_managed_server() so the next adoption pass sees a clean slate. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Three follow-ups from live PR #406 review 1. Fix Restart/Start label bug. After clicking the primary button, the newly-spawned --reload server's LISTEN holder is uvicorn's `multiprocessing.spawn` worker — that PID's cmdline doesn't contain the godot_ai brand, so `is_dev_server_running()` returned false and the label stayed "Start Dev Server" instead of flipping to "Restart Dev Server" once the server was up. Walk up the parent chain (bounded depth 5) inside `_pid_cmdline_is_godot_ai` so a worker whose parent reloader IS branded gets recognised. Adds `_pid_parent` helper covering POSIX (`ps -o ppid=`) and Windows (PowerShell CIM). 2. Drop the dock's "Mode override" dropdown. The dropdown advertised toggling between the github-repo local plugin and the installed ZIP version, but it didn't actually do that — it just nudged `is_dev_checkout()` for the update banner, which is more usefully driven by the GODOT_AI_MODE env var (still supported, still tested in test_clients.gd). Removes ~50 lines of dead UI plumbing and the stale "fresh check paints over a clean slate" comment. 3. Delete the orphan live_demo test file from a debugging session that never landed cleanly. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

dsarno requested a review from Copilot May 7, 2026 15:00

Copilot started reviewing on behalf of dsarno May 7, 2026 15:01 View session

Copilot AI reviewed May 7, 2026

View reviewed changes

dsarno and others added 5 commits May 7, 2026 08:15

dsarno force-pushed the dlight/bold-shannon-846b3e branch from 3493e04 to 6ba9ff8 Compare May 7, 2026 15:16

dsarno changed the base branch from main to beta May 7, 2026 15:16

dsarno merged commit ae3f769 into beta May 7, 2026
22 checks passed

dsarno mentioned this pull request May 7, 2026

Dev-section follow-ups: Restarting… feedback + consolidate buttons + parent-walk #408

Merged

5 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add Restart Server button to dev-mode Setup section#406

Add Restart Server button to dev-mode Setup section#406
dsarno merged 5 commits into
betafrom
dlight/bold-shannon-846b3e

dsarno commented May 7, 2026 •

edited

Loading

Uh oh!

codecov Bot commented May 7, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

dsarno May 7, 2026

Uh oh!

dsarno May 7, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

dsarno commented May 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Why

Implementation notes

Test plan

Uh oh!

codecov Bot commented May 7, 2026

Codecov Report

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

dsarno May 7, 2026

Choose a reason for hiding this comment

Uh oh!

dsarno May 7, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dsarno commented May 7, 2026 •

edited

Loading