Repository navigation
v0.3.85 — Expected test errors no longer raise console error pressure, and editor quit is confirmed by process exit
Why this matters
A PlayMode suite where every test passed could still report
console_error_pressure_detected: true, the flag meaning "Unity logged console
errors during this run", because a test deliberately triggered an error and
asserted it with LogAssert.Expect. Suites longer than about 5 minutes could end
in request_lifecycle_reset after the Play Mode domain reload, even with a
larger timeoutMs. And when the Unity editor was in the background in Edit
Mode, unity.editor.quit replied "acknowledged" while the editor stayed open.
By default, request-editor-quit returned at that acknowledgement. With
--wait-for-exit, it waited 20-30 s and then reported
editor_quit_ack_without_exit.
What changed
- Long test runs keep the caller's timeout. Entering Play Mode reloads the
script domain, which drops the request inside the editor. The host (the
launcher and MCP server that run outside Unity) then waits for the run to
finish and recovers its result; this is the post-reset recovery wait.
unity.tests.run_editmodeandunity.tests.run_playmodeno longer stop that
wait at a fixed 300 s cap. The longest live run in this release took 250 s;
waits longer than 300 s are covered by host unit tests only. A runtime config
post_reset_recovery_cap_msfor these two operations is ignored and listed in
runtime-config-showignored_overrides. Recovery now reports
post_reset_recovery_mode,post_reset_recovery_budget_ms,
post_reset_recovery_elapsed_msandpost_reset_recovery_stop_reason. - Console errors are attributed per test. Through Unity Test Framework's log
scope, an error counts asexpected_by_testonly when the test passed and
either matched it withLogAssert.Expector had
LogAssert.ignoreFailingMessagesset. Those errors no longer set
console_error_pressure_detected. Errors without that evidence (off the main
thread, outside a test, during an interrupted test, before a domain reload, or
on a Test Framework without a readable log scope) still raise pressure.
test_verdictis unchanged. New fields:
console_error_expected_by_test_count,console_error_not_expected_count,
console_error_classification_basisand up to 3console_error_sampleswith
the active test name.console_error_count_since_request_startkeeps its raw
meaning and still includes expected errors. - Final status of a finished run uses that run's counts. Counts are stored
with the result (console_error_count_at_completionand related fields)
instead of being read from the live editor counter that later runs change. A
result stored without them reportslower_bound_from_recorded_attributionor
unavailable. - Test results record the test setup. A new
test_environmentobject holds
the Game View render size, preset label,preset_kind, pixel size,
render_matches_preset, size group, build target andcapture_status. It is
captured without opening, focusing or waiting on a window. - The compact CLI summary keeps the verdict. When the compact CLI envelope
(the short JSON summary the launcher prints) exceeds 8 KB, it drops
console_error_samples, thentest_environment, then shortens
first_failuretext before falling back to the reduced form. It lists what it
dropped incompact_omitted_fields. The reduced form keepstest_verdict,
the pass/fail/skip totals,failure_countand the console error pressure
fields. - Editor quit works when the editor is unfocused. On Unity
6000.0.58f2on
macOS, Unity skippedEditorApplication.delayCallin Edit Mode while the
editor was not the active app. The quit now fires from a one-shot
EditorApplication.updatetick. The quit intent is kept for 120 s so that it
survives a domain reload; that part is covered by EditMode and unit tests
only. The acknowledgement (the bridge's reply that it accepted the quit) adds
quit_mechanism: editor_update_one_shot,editor_pid,
editor_application_focused,is_playingandscene_dirty. Older packages
reportquit_mechanism: legacy_delay_call. - Behaviour change:
request-editor-quitwaits for the process to exit by
default (--exit-timeout-msdefault 30000). If the editor survives, the
command exits non-zero witheditor_quit_ack_without_exit, the live editor
PIDs and a recovery command; it sends no signal.--no-wait-for-exitrestores
the old ack-only result ascloseout_classification: quit_ack_unverified
withprocess_exit_verified: false. A quit the bridge rejects now fails at
once witheditor_quit_request_rejected(closeout_classification: quit_request_rejected). unity.game_view.configureselects only fixed-resolution sizes.9x16
used to select Unity's built-in16:9 Portraitaspect-ratio preset and report
resolution_applied. It now returns an error that explains the match is an
aspect-ratio preset, or creates a fixed custom size when
allowCreateCustomSizeis set. The resolvedgame_viewaddspreset_kind
(fixed_resolution,aspect_ratio,free,unknown).- Printed recovery commands use the absolute launcher path. Recovery,
cleanup and refresh commands now start with the quoted absolute launcher path
(PowerShell gets a&prefix) and include a quoted project root where
needed. An unknown, missing or mismatched launcher path falls back to the
bare launcher name, which does not run from an arbitrary directory. ensure-readyreports a declared package correctly.
package_import_state.dependency_declarednow reads the manifest, so a
git-pinned dependency no longer showsfalsewhilestatus-summaryreports
it ready.- Upgrade together. Update the Unity package pin and the host launcher
together. A project still pinned to v0.3.84 keeps the old quit path. With the
editor unfocused in Edit Mode, that quit does not run, so
request-editor-quitfails after about 30 s with
editor_quit_ack_without_exit; a focused editor still quits. Older packages
also keep raw-count console error pressure. - Unsaved changes on quit.
EditorApplication.Exitis forced and discards
unsaved scene and Prefab Mode changes without a prompt. That is unchanged, but
the quit now actually happens while the editor is unfocused. Check
scene_dirtyin the acknowledgement. - Restart clients. Restart MCP clients after updating. After upgrading an
installed launcher, restart twice, because the first start still runs the
previous refresh launcher.
What this gives developers
- Fewer false alarms from expected errors. With the new package and a Test
Framework that has a readable log scope, main-thread errors that a passing
test expected no longer set pressure. Errors logged off the main thread,
outside a test, during an interrupted test or before a domain reload still
set it. When pressure is true, the samples name the test that was running. - Long PlayMode suites are no longer cut off by a fixed 5-minute recovery cap;
they wait up to thetimeoutMsyou gave (live runs in this release reached
250 s, so the longer case rests on host unit tests). - Final status you read later describes the run you asked about, even after
other runs. - A failing layout test shows the resolution it ran at, and
preset_kindtells
you whether the Game View used a fixed size or an aspect-ratio preset. - By default, a successful
request-editor-quitmeans the editor process
exited. If the editor is still running after the timeout, the command names
the live PIDs and prints a recovery command. With--no-wait-for-exit, a zero
exit code only means the quit was acknowledged (quit_ack_unverified). - Large test results keep their verdict and totals in compact output.
Validation
- Host Python suite: 1279 tests OK, 14 skipped on macOS. Release safety, docs
freshness and version consistency checks passed. - Remote CI (Integration Tests on Windows, Ubuntu and macOS, and Discovery
Checks) passed for the release commit, and the Release Tag Gate passed. - Live checks on macOS with Unity
6000.0.58f2in two consumer projects:- Package EditMode self-tests:
186/186passed,console_error_pressure_detected
false. - A 428-test PlayMode suite: 426 passed, 2 skipped, 0 failed in 237 s,
pressure false. Before this release it reported pressure true from errors
the tests expected. - A 39-test PlayMode fixture whose test expects one exception:
39/39,
pressure false,expected_by_test1. Before this release it reported
pressure true. test_environmentwas correct at the1080x1920and1440x3200fixed
presets and under an aspect-ratio preset (render_matches_presetempty). A
layout test that fails at1440x3200shows the 20:9 render size in its own
result.- Request final status for a completed run returned that run's own counts
after later runs. - Printed recovery commands ran unchanged from the home directory.
- Editor quit while unfocused in Edit Mode: on v0.3.84 the quit was
acknowledged but the editor did not exit (20-30 s, then ack-without-exit).
With this release the editor exited in about 1.4-1.8 s without SIGTERM on a
fresh editor, after a PlayMode run, after a refresh and after a forced
domain reload (6 runs). In the domain reload case the quit was sent after
the reload finished. - One more quit run used
--no-wait-for-exit. That mode does not check
process exit, so it is not part of the exit times above. unity.game_view.configurewith9x16returns the explanatory error;
1080x1920and1440x3200still select fixed presets.
- Package EditMode self-tests:
Known limitations
- Remote Unity Package CI remains waived because its runner has no Unity
license. - A post-reset recovery wait longer than 300 s was not verified live; the
longest live suite took 250 s, and the longer case is covered by host unit
tests only. - A quit armed before a domain reload and fired after it was not shown live; it
is covered by EditMode and unit tests only. - The
EditorApplication.delayCallbehaviour behind the quit fix was observed
only on Unity6000.0.58f2on macOS. - Unity 2021.3 was only compiled offline at C# 9 against 2021.3 assemblies, not
run. - Windows and Ubuntu editors were not tested live.
- Printed commands were run live only on macOS from the home directory; cmd and
PowerShell quoting on Windows hosts were not run live. allowCreateCustomSize: trueafter an aspect-ratio-only match and quitting
with a dirty scene were not verified live.