Skip to content

fix(launcher): detach application output from launcher replies - #22

Merged
sandeepbist merged 1 commit into
mainfrom
codex/launcher-reply-streams
Oct 28, 2026
Merged

sandeepbist merged 1 commit into
mainfrom
codex/launcher-reply-streams

Conversation

@sandeepbist

Copy link
Copy Markdown
Owner

Why

Applications inherited the launcher's reply pipes. A long-running application could keep the JSON response open and leave the launcher waiting.

What changed

  • Use GioUnix's desktop-entry launch API with explicit null-device streams for the application.
  • Verify that the helper finishes before the application exits and that the application can still write afterward.

Scope

  • Covers application output streams and the launcher's JSON response.
  • Preserves desktop-entry quoting, working directories, environment cleanup, and terminal wrapping.

Blast Radius

The change affects application launches through the existing helper. It keeps the native desktop-entry API and does not alter application closing.

Verification

  • python3 tests/launcher.py passes for GUI and terminal entries, quoted arguments, field codes, working directories, environment cleanup, and missing entries.
  • The new reply-pipe regression fails against the original helper and passes with the corrected helper.

@sandeepbist

Copy link
Copy Markdown
Owner Author

Automated independent verification: PASS

This is verification evidence, not a formal approval.

Verified exact head 3b4e593104d8516f7623755514e155daa2d51491 against base main at d9904944e943d33c14fbbfb2976867b03ca67d8c. Stable base-to-head patch-id 5f1ae7b17e30bd03ecfad46cc33efceb26d78496. Both changed paths were reviewed. The supplied isolated worktree remained clean; no production edits were made.

tests/launcher.py passes on exact head. The exact old helper passes its existing GUI/terminal checks, then fails the new regression with Launched app retains launcher reply pipes; helper cannot finish independently. Head returns clean JSON before the waiting fixture exits, and the fixture writes successfully afterward.

tests/launcher-ui.py passes offscreen. Additional isolated fixtures preserve GUI and foot/kitty/alacritty/wezterm semantics, PATH lookup, quoting, field codes, working directory, clean environment and stdin EOF. Parent and head both dispatch D-Bus activation to a private registered fixture without executing the desktop fallback. No real applications or terminal emulators were launched.

The installed Gio binding accepts the native call. Upstream GLib 2.90.1 shows both APIs use the same desktop-entry launch dispatcher and SEARCH_PATH defaults; the change supplies explicit child stream descriptors. No unintended desktop-entry semantic change was found. D-Bus activation intentionally ignores spawn descriptors, as documented by Gio.

Changed-comment review found one added external stream-behavior explanation to keep, no deletion recommendations, no MUST KILL or HOLD flags. Actual deletions 0.

The single GitHub snapshot showed 5/5 checks successful, REVIEW_REQUIRED, no reviews or review threads, and merge state BLOCKED. That readiness state is separate from this source PASS. No CI polling, formal approval, merge, watch or compositor/plugin changes occurred.

Scope limits include the installed GLib/Python/Quickshell versions, offscreen UI and synthetic launch fixtures. D-Bus service auto-start and real terminal PTYs were not exercised.

@sandeepbist
sandeepbist force-pushed the codex/launcher-reply-streams branch from 3b4e593 to 1e5c492 Compare October 28, 2026 21:02
@sandeepbist
sandeepbist merged commit 01170de into main Oct 28, 2026
5 checks passed
@sandeepbist
sandeepbist deleted the codex/launcher-reply-streams branch October 28, 2026 21:17
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant