claude-code-usage-bubble/plans/260516-1730-github-release-auto-update/phase-02-end-to-end-update-test.md

phase, title, status, priority, effort, dependencies

phase	title	status	priority	effort	dependencies
2	End-to-end update test	pending	P1	2h	1

Phase 2: End-to-end update test

Overview

Cut a real test release pair and prove the in-app updater finds it, downloads it, swaps the running .exe, and relaunches into the new version. The updater code already exists; this phase is about flushing out integration bugs between CI output and the updater's expectations (asset name casing, redirect behavior, file lock release timing, version-comparison edge cases).

Requirements

Functional

• Cut v0.1.0 from current main → CI uploads asset.
• Bump Cargo.toml to 0.1.1, commit, push, tag v0.1.1 → CI uploads asset.
• On a Windows test machine, run the v0.1.0 binary downloaded from the v0.1.0 release.
• From the right-click menu, "Check for updates" must transition Idle → Checking → Available.
• Clicking "Update available" must transition Available → Applying and exit the process.
• ~2 seconds later, v0.1.1 must be running (verify via right-click menu showing "Up to date" after re-check, or via file properties on the .exe).
• The version_action apply branch (src/app.rs:1037-1066) must succeed: update::install::begin returns Ok(()) and the process posts WM_QUIT (PostQuitMessage(0) at src/app.rs:1056).

Non-functional

• Test machine has no admin privileges → confirms ensure_writable (src/update/install.rs:81-89) works for %LOCALAPPDATA% install.
• Run from an install path that contains a space (e.g. C:\Users\test user\bin\) to validate spawn_handoff quoting (src/update/install.rs:53-69).

Architecture

Test matrix

Scenario	Where exe lives	Expected
Vanilla user-local install	%LOCALAPPDATA%\ClaudeCodeUsageBubble\	Succeeds
Path with spaces	C:\Users\test user\bin\	Succeeds (cmd /c quoting)
Read-only install dir (e.g. C:\Program Files\…)	C:\Program Files\Bubble\	Failed status surfaces; no crash
Offline	n/a	Failed status, no crash, retries on next 24h timer
Already on latest	n/a	UpToDate status, no download

Observability

• Run with claude-code-usage-bubble.exe --diagnose to capture the log at %TEMP%\claude-code-usage-bubble.log. Look for update apply failed: lines (src/app.rs:1058).
• After update, the new process is started by cmd.exe (detached); Task Manager parent column will show no parent — that's expected.

Related Code Files

• Reference only (no edits expected): src/update/release.rs, src/update/install.rs, src/app.rs (lines 1025-1135), Cargo.toml

Implementation Steps

1. Cut v0.1.0:

   git -C D:/tiennm99/claude-code-usage-bubble tag -a v0.1.0 -m "v0.1.0"
   git -C D:/tiennm99/claude-code-usage-bubble push origin v0.1.0
   

   Wait for the Phase-1 workflow to produce v0.1.0 release with claude-code-usage-bubble.exe. Download the asset locally — this is the "old" binary.

2. Smoke-test the v0.1.0 download on a Windows machine: run it, confirm the bubble appears, right-click → "Check for updates" returns "Up to date" (no v0.1.1 yet).

3. Cut v0.1.1:

   • Bump Cargo.toml version = "0.1.0" → "0.1.1".
   • cargo build --release locally to refresh Cargo.lock.
   • Commit: chore: bump version to 0.1.1.
   • Tag: git tag -a v0.1.1 -m "v0.1.1".
   • Push both: git push origin main && git push origin v0.1.1.

4. Run the v0.1.0 binary (still installed from step 1) and right-click → "Check for updates". Status should transition to "Update available". Click it. The process exits, ~2 s pass, the new v0.1.1 binary should launch automatically.

5. Verify v0.1.1 is running: right-click → "Check for updates" should now return "Up to date". Cross-check claude-code-usage-bubble.exe --diagnose log for the version line, or check file properties in Explorer.

6. Cleanup if it goes wrong:

   • Stuck "Applying" status with no swap → kill the detached cmd.exe in Task Manager, manually copy %LOCALAPPDATA%\ClaudeCodeUsageBubble\updates\update.exe over the running exe location.
   • cmd /c quoting broke → fix in src/update/install.rs:58-60 and retag as v0.1.2.

7. Run negative scenarios (table above): path-with-spaces, read-only install dir, offline. Each must fail-soft without crashing.

Todo List

• v0.1.0 release cut and asset downloaded
• v0.1.0 binary verified runnable on Windows
• Cargo.toml bumped to 0.1.1, committed, tagged, pushed
• v0.1.1 release produced by CI
• v0.1.0 binary self-updates to v0.1.1 successfully
• Post-update, "Check for updates" returns "Up to date"
• Negative scenario: install in path with space succeeds
• Negative scenario: read-only install dir surfaces "Failed" status, no crash
• Negative scenario: offline → "Failed", retry timer rearmed

Success Criteria

• A v0.1.0 download → click update → v0.1.1 running with no manual file copying.
• No SmartScreen kill (it will warn on first run; that's expected and documented).
• %TEMP%\claude-code-usage-bubble.log contains no update apply failed lines after the successful run.
• All negative scenarios fail without crashing the bubble.

Risk Assessment

Risk	Likelihood	Mitigation
File lock not released in 2 s window	Low-Medium	The 2 s timeout in spawn_handoff is conservative; if it ever races, bump to 3 s
GitHub CDN redirect not followed by WinHTTP	Very Low	WinHTTP follows redirects by default (no WINHTTP_OPTION_DISABLE_FEATURE set); will surface in step 4 if broken
Antivirus quarantines the freshly-written staging exe	Medium	Document the workaround (allowlist the install dir); future signing fixes this
Test pollutes real release feed	Low	If you must test with throwaway tags, use workflow_dispatch (creates draft) instead of pushing the tag

Security Considerations

• The downloaded asset is fetched over HTTPS from a *.githubusercontent.com CDN; WinHTTP validates certs against the system root store.
• No checksum verification yet — accepted risk (HTTPS + cert pinning is the floor). Future enhancement: ship SHA256SUMS.txt and verify in install::download.
• The cmd /c command string is composed only from current_exe() and stage_path(); neither is user-controlled. No shell injection vector.