Reapply renderer stress fixes without terminal sizing regression

apps/desktop/docs/RENDERER_STRESS_QA.md

@@ -0,0 +1,230 @@
+# Renderer Stress QA Fixtures and Instrumentation
+
+This is the repeatable process for local renderer QA with seeded V2
+workspaces, large dirty diffs, terminal stress, WebGL context-loss coverage,
+and renderer instrumentation. Use it before broad manual QA when a change may
+affect workspace switching, changes/diff panels, terminal rendering, tab/pane
+state, or renderer responsiveness.
+
+## Safety
+
+- Run this only against a local dev `SUPERSET_HOME_DIR`. The fixture command
+  writes local TanStack and host-service SQLite rows.
+- Do not point this at production or shared data.
+- The fixture worktrees are disposable. By default they live at
+  `~/workplace/playground/superset-renderer-stress-fixtures`.
+- If the desktop app is already open when fixtures are created, restart it
+  before running stress so the main process sees the seeded host DB state.
+
+## What The Fixture Command Seeds
+
+`bun --cwd apps/desktop stress:renderer:fixtures` creates:
+
+- A disposable git repo with many tracked files.
+- Multiple git worktrees with large dirty changes, deletes, new files, and
+  renames.
+- A V2 project row in the local persisted collections DB.
+- V2 workspace rows for the generated worktrees.
+- Matching host-service `projects` and `workspaces` rows in
+  `host/<organization-id>/host.db`.
+
+The stress harness then drives those workspaces through the renderer using the
+Chrome DevTools Protocol (CDP) and the in-app renderer stress bridge.
+
+## Prepare Fixture Workspaces
+
+For a normal large-diff run:
+
+```bash
+bun --cwd apps/desktop stress:renderer:fixtures -- --json
+```
+
+For a smaller smoke fixture while iterating:
+
+```bash
+bun --cwd apps/desktop stress:renderer:fixtures -- \
+  --workspace-count 2 \
+  --changed-files 20 \
+  --lines-per-file 20 \
+  --base-files 80 \
+  --json
+```
+
+If you are running with `SKIP_ENV_VALIDATION=1`, the renderer may use the
+mock auth organization id (`mock-org-id`). In that case, seed the fixture into
+that org and pass a host id from an existing local host DB:
+
+```bash
+bun --cwd apps/desktop stress:renderer:fixtures -- \
+  --organization-id mock-org-id \
+  --host-id <local-host-id> \
+  --workspace-count 2 \
+  --changed-files 20 \
+  --lines-per-file 20 \
+  --base-files 80 \
+  --json
+```
+
+If the fixture command reports `Host DB not found` for the mock org, copy an
+existing local host DB into the mock org path before seeding:
+
+```bash
+mkdir -p superset-dev-data/host/mock-org-id
+sqlite3 superset-dev-data/host/<existing-org-id>/host.db \
+  ".backup 'superset-dev-data/host/mock-org-id/host.db'"
+```
+
+Keep the `workspaceIds` from the JSON output. Most targeted QA runs should pass
+those ids explicitly so the harness exercises real seeded workspaces instead
+of falling back to hash navigation.
+
+## Start The Desktop App With CDP
+
+Start the dev app in one shell:
+
+```bash
+SKIP_ENV_VALIDATION=1 \
+SUPERSET_RENDERER_STRESS_CDP_PORT=9333 \
+bun --cwd apps/desktop dev
+```
+
+`SUPERSET_RENDERER_STRESS_CDP_PORT` opens the Electron renderer for CDP-based
+stress automation. The main process also disables extension loading while this
+variable is set so extension noise does not affect results.
+
+## Run The Stress Harness
+
+Run stress from another shell after the app has loaded:
+
+```bash
+bun --cwd apps/desktop stress:renderer -- \
+  --port 9333 \
+  --scenario all \
+  --workspace-ids <workspace-id-1>,<workspace-id-2> \
+  --iterations 1000 \
+  --route-iterations 240 \
+  --heavy-iterations 500 \
+  --timeout-ms 300000
+```
+
+For the terminal-heavy flow we used in this PR:
+
+```bash
+bun --cwd apps/desktop stress:renderer -- \
+  --port 9333 \
+  --scenario terminal-heavy \
+  --workspace-ids <workspace-id-1>,<workspace-id-2> \
+  --terminal-iterations 200 \
+  --terminal-tab-count 32 \
+  --terminal-panes-per-tab 4 \
+  --terminal-lines 80 \
+  --terminal-payload-bytes 4096 \
+  --interval-ms 0 \
+  --settle-ms 1500 \
+  --timeout-ms 300000 \
+  --max-heartbeat-delay-ms 10000 \
+  --max-long-task-ms 10000 \
+  --progress-every 10
+```
+
+The terminal-heavy scenario forces terminal WebGL context loss by default. Add
+`--no-terminal-webgl-loss` when you need a control run without forced context
+loss.
+
+## Useful Scenarios
+
+- `workspace-switch`: repeatedly activates workspaces.
+- `workspace-switch-heavy`: activates workspaces while generating synthetic
+  tabs and panes.
+- `workspace-heavy`: runs mixed workspace, pane, browser, and diff actions.
+- `route-sweep`: navigates renderer routes.
+- `terminal-heavy`: creates synthetic terminal tabs/panes, writes large ANSI
+  payloads, switches tabs, and forces terminal WebGL context loss.
+- `all`: combines route, workspace, heavy, and terminal coverage.
+
+## Instrumentation Knobs
+
+Use these options when a run fails or feels slow:
+
+```bash
+--profile-cpu
+--react-probe
+--json
+--progress-every 10
+--max-heartbeat-delay-ms <ms>
+--max-long-task-ms <ms>
+--timeout-ms <ms>
+```
+
+What to look at:
+
+- `errorCount` and `errors`: uncaught renderer errors and unhandled rejections.
+- `maxHeartbeatDelayMs`: event-loop stalls detected by the renderer heartbeat.
+- `maxLongTaskDurationMs` and `longTasks`: browser long-task evidence.
+- `terminalWebglContextLosses`: terminal canvas/WebGL context-loss samples.
+- CPU profile output: hottest JS frames during the run.
+- React probe output: commit/component counts when the React DevTools hook is
+  available.
+
+For strict automated runs, keep the default heartbeat and long-task thresholds.
+For crash reproduction or exploratory QA, raise them so the harness can keep
+running long enough to collect the actual renderer error.
+
+## Reproducing A Suspected Regression
+
+Use paired runs so the result identifies the failing subsystem instead of just
+the commit range:
+
+1. Run the fixture command and record the generated workspace ids.
+2. Start the app with CDP enabled.
+3. Run the target stress scenario on the current branch.
+4. Locally revert only the suspected fix with `git revert --no-commit <sha>`
+   or restore only the relevant files.
+5. Restart the app with CDP enabled.
+6. Run the exact same stress command with the same workspace ids.
+7. Restore the branch and rerun the command once more.
+
+In this PR, that process showed the renderer did not need a full process crash
+to fail. The failing signal was an unhandled rejection:
+
+```text
+RangeError: WebAssembly.instantiate(): Out of memory
+```
+
+The stack came from `@xterm/addon-image`, and it reproduced both with the
+WebGL context-loss fix reverted and restored. That isolated the issue to
+per-terminal image decoder memory, not daemon update/attach behavior and not
+terminal sizing.
+
+## Manual QA After Stress
+
+After an automated run passes, keep the app open and manually check:
+
+- Workspace switching between the seeded large-diff workspaces.
+- Changes panel status, file list, file selection, diff render, and staging.
+- Terminal tab switching, pane splitting, clear, search, and resize.
+- New terminal creation immediately after attach. The terminal should size
+  correctly without waiting for a later layout event.
+- Returning to an existing terminal. It should not perform a full replay just
+  because the user switched away and back.
+- Navigation away from and back to the workspace route.
+- Opening and closing enough terminal tabs/panes to verify WebGL fallback does
+  not change terminal geometry.
+
+## Cleanup
+
+The fixture command removes previous rows for the generated
+`renderer-stress-large-diff` project before inserting fresh rows. To remove the
+fixture files manually:
+
+```bash
+rm -rf ~/workplace/playground/superset-renderer-stress-fixtures
+```
+
+If you copied a host DB into `mock-org-id` only for stress testing, remove it
+when you want to return to the normal local auth org:
+
+```bash
+rm -rf superset-dev-data/host/mock-org-id
+```
+

apps/desktop/docs/V2_LAUNCH_TEST_PLAN.md

@@ -6,6 +6,9 @@ Pair with `V2_LAUNCH_CONTEXT.md` for architectural background and
 
 ## Setup
 
+For renderer stress fixtures, CDP instrumentation, terminal-heavy runs, and
+large-diff workspace QA, see `RENDERER_STRESS_QA.md`.
+
 1. `bun dev` at the repo root.
 2. Ensure your active org has V2 cloud enabled (or you're testing a V2
    project).

apps/desktop/package.json

@@ -34,7 +34,9 @@
 		"generate:routes": "tsr generate",
 		"pretypecheck": "bun run generate:icons && bun run generate:routes",
 		"typecheck": "tsc --noEmit",
-		"test": "bun test"
+		"test": "bun test",
+		"stress:renderer": "bun run scripts/stress-renderer.ts",
+		"stress:renderer:fixtures": "bun run scripts/prepare-renderer-stress-fixtures.ts"
 	},
 	"dependencies": {
 		"@ai-sdk/anthropic": "3.0.64",