feat(build): add no-terminal flag to omit embedded terminal assets

[xaniexane]

Problem

Fixes #2897.

With the Alpine → Ubuntu terminal migration (#2699), the embedded rootfs payload grows from ~15MB to ~90MB, pushing test APKs to ~113MB. For developers on capped/metered connections, every fork-test build is a heavy download, and the F-Droid build workaround targets legacy API 28 — unsuitable for testing modern behavior.

Change

Adds a no-terminal argument to the build script:

npm run build paid dev apk no-terminal

When set, the build skips the com.foxdebug.acode.rk.exec.proot plugin (which bundles the per-arch alpine.rootfs assets), so the rootfs is never copied into the APK. This follows the exact pattern the fdroid flag already uses for plugin add/remove, and a subsequent normal build re-adds the plugin automatically — no manual cleanup.

Deliberately untouched:

• Runtime $FDROID handling — a no-terminal build behaves like a normal build at runtime; the terminal simply fails fast on open (existing extract-asset error path) since its payload isn't bundled.
• The fdroid flag semantics — combining both flags still works (fdroid wins on runtime behavior).

Testing

• bash -n syntax check passes.
• Exercised the argument/plugin logic in isolation with a stubbed cordova across all four states:
  • no-terminal + plugin present → plugin removed, notice printed
  • normal + plugin present → no plugin churn
  • no-terminal + plugin absent → no-op
  • normal + plugin absent → plugin re-added (self-healing)
• Full APK build not run here (needs Android SDK/Cordova toolchain); the change only gates plugin add/remove, which is the same mechanism fdroid has used in this script.

[greptile-apps]

Confidence Score: 4/5

The PR is not safe to merge until format-on-save prevents a delayed formatter dispatch from modifying a different tab after a same-pane switch.

Findings

1.  Tab switch corrupts content ▶

Summary

This PR adds a no-terminal build argument that removes the proot Cordova plugin before packaging and restores it on a subsequent normal build. It also changes format-on-save to await formatting, propagate formatter results, and avoid initially formatting a file that is no longer active.

• Adds conditional proot plugin removal to produce smaller development APKs without embedded terminal rootfs assets.
• Introduces a true/false/null formatting result contract and propagates it through the format command.
• Makes save operations wait for formatting before reading and writing document content.
• The new format-on-save sequencing does not protect against switching tabs while the formatter itself is awaiting work, leaving a cross-file content corruption race.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Save file A with format-on-save] --> B{A is currently active?}
  B -->|No| F[Skip formatting and save A]
  B -->|Yes| C[Await asynchronous formatter]
  C --> D{User switches pane to file B?}
  D -->|No| E[Apply formatted state to A and save]
  D -->|Yes| G[Pane reuses EditorView for B]
  G --> H[Late formatter dispatch updates B]
  H --> I[Save writes stale content from A]

Loading

_{Reviews (1) · Last reviewed commit: "feat(build): add no-terminal flag to omi..."}

[greptile-apps]

Tab switch corrupts content

If file A starts format-on-save and the user switches the same pane to file B while formatting is still running, the formatter later applies A's formatted output through the pane's reused editor. The document listener assigns that update to file B because B is active by then. This can replace B's in-memory content, mark or autosave B as edited, and save stale unformatted content for A. Re-check the file identity before applying formatter changes, or bind formatting to A's document instead of the pane's mutable active editor.

Knowledge Base Used:

• Shared application services
• Workspace and code editor

[xaniexane]

Thanks for the review! One correction: the "Tab switch corrupts content" finding doesn't apply to this branch as it exists now. It was written against commit 9f01cc3, which still contained the format-on-save work — that has since been moved to its own branch (PR #2915). This branch is now exactly 7d828e9, which only touches utils/scripts/build.sh to add the no-terminal flag. No editor, formatter, or save-path code exists on this branch, so the finding can't apply here. No code change needed on this PR.