Release 0.1.5: stdout-encoding onboarding fix → production

[vigneshsubbiah16]

Release 0.1.5 → production

Promotes staging to main for the runtime-v0.1.5 release. Pinned release/0.1.5 branch so the production delta is exactly these two commits, nothing in-flight:

• fix(unbound-hook): never let stdout encoding abort onboarding setup #153 — fix(unbound-hook): stdout-encoding crash. unbound-hook setup aborted on Jamf's recurring check-in because the launchd context has no LANG/LC_*, so the interpreter fell back to the ASCII stdout codec and the migration banner's → raised UnicodeEncodeError. main() now reconfigures stdout/stderr to UTF-8 so a diagnostic print can never abort the install. Reproduced on a real Mac under the field environment; fixed, 16/16 setup tests pass.
• chore(release): bump runtime version 0.1.4 -> 0.1.5 #154 — chore(release): bump both binaries' __version__ 0.1.4 → 0.1.5 (required or the release workflow's publish-safety assert fails).

After merge

Push tag runtime-v0.1.5 on main to cut the signed/notarized build → publishes to s3://unbound-release-artifacts/macos/0.1.5/ and repoints the latest/onboard.sh the fleet installs from.

Customer impact

Clears the check-in install failure reported during the Salesloft fleet rollout. The interim env-wrapper workaround in use today can be dropped once 0.1.5 is live.

Greptile Summary

This release promotes the stdout-encoding crash fix to production: _force_utf8_io() is called at the top of main() to reconfigure stdout/stderr to UTF-8 with errors='replace', preventing UnicodeEncodeError from aborting onboarding in Jamf's launchd context where no LANG/LC_* is set. The migration banner's non-ASCII arrow is also replaced with ASCII -> as a belt-and-suspenders measure, and both binaries are version-bumped to 0.1.5.

• _force_utf8_io() added to unbound_hook/main.py: guards all stdout/stderr output for the hook binary against locale-less launchd environments; a regression test verifies the fix end-to-end by monkeypatching an ASCII-only stream.
• setup_cmd.py banner updated: the → character that triggered the original crash is replaced with ->, removing the source of the problem independently of the stream reconfiguration.
• packaging/unbound_discovery_entry.py receives only a version bump; the same _force_utf8_io() guard is not applied despite this binary running in the identical LaunchDaemon/no-locale environment.

Confidence Score: 3/5

The hook binary fix is correct and well-tested, but the discovery binary ships in the same launchd/no-locale environment without the same guard, leaving an unresolved crash path.

The core fix in unbound_hook/main.py is sound and the regression test exercises the exact field scenario. However, unbound_discovery_entry.py runs as a root LaunchDaemon under the identical locale-free environment and calls upstream library code that can print arbitrary Unicode — the same class of crash that this PR fixes in the hook binary remains reachable in the discovery binary.

packaging/unbound_discovery_entry.py — missing the stdout reconfiguration applied to the hook binary, despite running in the same environment.

Important Files Changed

Filename	Overview
binary/src/unbound_hook/main.py	Adds _force_utf8_io() called at main() entry to reconfigure stdout/stderr to UTF-8 with errors='replace'; fix is sound but exception handler silently swallows failures without attempting to log to the other stream.
packaging/unbound_discovery_entry.py	Version bumped to 0.1.5, but the same UTF-8 stdout reconfiguration fix applied to unbound_hook/main.py is absent here despite this binary running in the same launchd/no-locale environment.
binary/src/unbound_hook/setup_cmd.py	Replaces the non-ASCII → arrow with ASCII -> in the migration banner, complementing the _force_utf8_io() fix by eliminating the known crash-triggering character at its source.
binary/tests/test_setup_migration.py	Adds a well-targeted regression test that monkeypatches sys.stdout/stderr to ASCII-only wrappers and verifies main() reconfigures them to UTF-8 and completes successfully.
binary/src/unbound_hook/init.py	Version bump from 0.1.4 to 0.1.5 — straightforward, required by the release workflow's publish-safety assert.

Sequence Diagram

sequenceDiagram
    participant Jamf as Jamf LaunchDaemon (no LANG/LC_*)
    participant main as unbound-hook main()
    participant utf8 as _force_utf8_io()
    participant stdout as sys.stdout (ASCII)
    participant setup as setup_cmd.run()

    Jamf->>main: execv unbound-hook setup --api-key ...
    main->>utf8: call _force_utf8_io()
    utf8->>stdout: "stream.reconfigure(encoding=utf-8, errors=replace)"
    stdout-->>utf8: stream now UTF-8
    utf8-->>main: return
    main->>setup: setup_cmd.run(args)
    setup->>stdout: "print([migration] python->binary sweep)"
    stdout-->>setup: write OK (non-ASCII safe)
    setup-->>main: return 0
    main-->>Jamf: exit 0

Loading

Comments Outside Diff (1)

1. packaging/unbound_discovery_entry.py, line 68-78 (link)

   Discovery binary missing the same UTF-8 reconfiguration

   unbound_discovery_entry.py ships as a separate binary that also runs from a root LaunchDaemon with no LANG/LC_* set — the exact same launchd context that produced the crash in unbound_hook. The upstream library calls (ai_tools_discovery.main(), scan_single_mcp_server.main()) can print arbitrary Unicode (tool names, file paths, server URIs), and if any non-ASCII character reaches stdout before this binary is shipped with the fix, it will raise UnicodeEncodeError and abort discovery on every check-in, just as the hook binary did. The _force_utf8_io() function now exists in the same repo — calling it at the top of main() here would close the gap.

_{Reviews (1): Last reviewed commit: "chore(release): bump runtime version 0.1..." | Re-trigger Greptile}

Greptile also left 1 inline comment on this PR.

[greptile-apps]

Silent pass leaves stream reconfiguration failures unobservable. If stdout.reconfigure() fails but stderr is still functional (or vice versa), a failure silently leaves one stream on the ASCII codec, which could still produce a crash downstream if a non-ASCII character is printed.

Suggested change

	except (ValueError, OSError):
	# Detached/closed stream — nothing we can do, and not worth
	# crashing onboarding over.
	pass
	except (ValueError, OSError) as exc:
	# Detached/closed stream — best-effort: try to warn on the other
	# stream rather than silently leaving it on the ASCII codec.
	other = sys.stderr if stream is sys.stdout else sys.stdout
	try:
	other.write(
	f"[unbound-hook] warning: could not reconfigure stream to UTF-8: {exc}\n"
	)
	except Exception:
	pass

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!