refactor: enhance test framework with automated resource tracking and scripted error capture capabilities

This commit is contained in:
2026-06-26 09:40:17 +08:00
parent 948a99cf90
commit 00f9d98f4b
58 changed files with 3594 additions and 1289 deletions
@@ -0,0 +1 @@
@@ -0,0 +1,51 @@
@tool
extends Logger
## Captures GDScript runtime errors emitted while a test is running.
##
## Deliberately no class_name: this script is compiled dynamically by
## script_error_capture_loader.gd only when Logger exists. It lives in a
## `.gdignore`d folder so older Godot editor scans never parse `extends Logger`.
##
## Only ERROR_TYPE_SCRIPT is captured. push_error(), push_warning(), and
## engine-internal ERR_FAIL_* checks are often valid negative-path assertions and
## should not abort the test.
var _mutex := Mutex.new()
var _capturing := false
var _errors := PackedStringArray()
func begin_capture() -> void:
_mutex.lock()
_capturing = true
_errors.clear()
_mutex.unlock()
func end_capture() -> PackedStringArray:
_mutex.lock()
var captured := _errors.duplicate()
_capturing = false
_errors.clear()
_mutex.unlock()
return captured
func _log_error(
function: String,
file: String,
line: int,
code: String,
rationale: String,
_editor_notify: bool,
error_type: int,
_script_backtraces: Array,
) -> void:
if error_type != ERROR_TYPE_SCRIPT:
return
_mutex.lock()
if _capturing:
var text := rationale if not rationale.is_empty() else code
_errors.append("%s (%s:%d in %s)" % [text, file, line, function])
_mutex.unlock()
@@ -0,0 +1,26 @@
@tool
extends RefCounted
## Builds the Logger-based test script-error capture lazily.
##
## Logger exists only on newer Godot versions. The capture implementation lives
## under a `.gdignore`d folder and is compiled from source only after the runner
## verifies the Logger API is present, so older editor scans do not parse an
## `extends Logger` file and emit red startup errors.
const SCRIPT_ERROR_CAPTURE_PATH := "res://addons/godot_ai/testing/loggers/script_error_capture.gd"
static func build() -> Object:
if not ClassDB.class_exists("Logger") or not OS.has_method("add_logger"):
return null
if not FileAccess.file_exists(SCRIPT_ERROR_CAPTURE_PATH):
return null
var source := FileAccess.get_file_as_string(SCRIPT_ERROR_CAPTURE_PATH)
if source.is_empty():
return null
var script := GDScript.new()
script.source_code = source
if script.reload() != OK:
return null
return script.new()
@@ -0,0 +1 @@
uid://pmtc07go8ty4
+80 -2
View File
@@ -5,11 +5,25 @@ extends RefCounted
## Lightweight test runner for MCP plugin tests. Discovers test_* methods
## on McpTestSuite instances, runs them, and collects structured results.
const ScriptErrorCaptureLoader := preload("res://addons/godot_ai/testing/script_error_capture_loader.gd")
var _results: Array[Dictionary] = []
var _last_run_ms: int = 0
var _script_error_capture: Object = null
var _capture_registered := false
func _notification(what: int) -> void:
if what == NOTIFICATION_PREDELETE and _capture_registered and _script_error_capture != null:
OS.call("remove_logger", _script_error_capture)
_capture_registered = false
func run_suite(suite: McpTestSuite, test_filter: String = "", exclude_test_filter: String = "") -> void:
var owns_capture := not _capture_registered
if owns_capture:
_register_capture()
var name := suite.suite_name()
var methods := _get_test_methods(suite)
var exclusions := _parse_exclusions(exclude_test_filter)
@@ -29,9 +43,12 @@ func run_suite(suite: McpTestSuite, test_filter: String = "", exclude_test_filte
continue
suite._reset()
_begin_script_error_capture()
suite.setup()
suite.call(method_name)
suite.teardown()
var script_errors := suite._unexpected_script_errors(_end_script_error_capture())
suite._free_tracked()
## Issue #19 defence: free any `_McpTest*` nodes the test created, even
## nested ones. If the scene gets auto-saved mid-test while one of these
@@ -39,10 +56,23 @@ func run_suite(suite: McpTestSuite, test_filter: String = "", exclude_test_filte
## with a "missing dependency" error. Runs after every test, not just at
## suite boundaries, so a test that fails mid-flow can't leave a trap
## for the next test or for scene autosave.
var scene_root_for_cleanup := EditorInterface.get_edited_scene_root()
var scene_root_for_cleanup := _edited_scene_root()
if scene_root_for_cleanup != null and scene_root_for_cleanup.is_inside_tree():
_free_mcp_test_nodes_recursive(scene_root_for_cleanup)
if not script_errors.is_empty():
var abort_message := "Aborted by SCRIPT ERROR: %s" % "; ".join(script_errors)
if suite._failed:
abort_message += " (after assertion failure: %s)" % suite._message
_results.append({
"suite": name,
"test": method_name,
"passed": false,
"message": abort_message,
"assertion_count": suite._assertion_count,
})
continue
if suite._skipped:
_results.append({
"suite": name,
@@ -70,6 +100,9 @@ func run_suite(suite: McpTestSuite, test_filter: String = "", exclude_test_filte
"assertion_count": suite._assertion_count,
})
if owns_capture:
_unregister_capture()
func run_suites(suites: Array, suite_filter: String = "", test_filter: String = "", ctx: Dictionary = {}, verbose: bool = false, exclude_test_filter: String = "") -> Dictionary:
_results.clear()
@@ -83,12 +116,17 @@ func run_suites(suites: Array, suite_filter: String = "", test_filter: String =
var _prev_console_echo := McpLogBuffer.console_echo
McpLogBuffer.console_echo = false
## If a prior run was interrupted after registering the logger but before
## normal teardown, remove that stale registration before starting fresh.
_unregister_capture()
_register_capture()
for suite: McpTestSuite in suites:
if not suite_filter.is_empty() and suite.suite_name() != suite_filter:
continue
## Snapshot scene children before the suite so we can clean up leaks.
var scene_root := EditorInterface.get_edited_scene_root()
var scene_root := _edited_scene_root()
var before_children: Array[Node] = []
if scene_root != null:
before_children = _get_children_snapshot(scene_root)
@@ -119,6 +157,7 @@ func run_suites(suites: Array, suite_filter: String = "", test_filter: String =
else:
run_suite(suite, test_filter, exclude_test_filter)
suite.suite_teardown()
suite._free_tracked()
## Remove any nodes the suite left behind (failed undo, missing cleanup).
if scene_root != null and scene_root.is_inside_tree():
@@ -126,9 +165,48 @@ func run_suites(suites: Array, suite_filter: String = "", test_filter: String =
_last_run_ms = Time.get_ticks_msec() - start
McpLogBuffer.console_echo = _prev_console_echo
_unregister_capture()
return get_results(verbose)
func _register_capture() -> void:
if _capture_registered:
return
if _script_error_capture == null:
_script_error_capture = ScriptErrorCaptureLoader.build()
if _script_error_capture == null:
return
OS.call("add_logger", _script_error_capture)
_capture_registered = true
func _unregister_capture() -> void:
if not _capture_registered:
return
if _script_error_capture == null:
_capture_registered = false
return
OS.call("remove_logger", _script_error_capture)
_capture_registered = false
func _begin_script_error_capture() -> void:
if _script_error_capture != null and _capture_registered:
_script_error_capture.call("begin_capture")
func _end_script_error_capture() -> PackedStringArray:
if _script_error_capture == null or not _capture_registered:
return PackedStringArray()
return _script_error_capture.call("end_capture") as PackedStringArray
static func _edited_scene_root() -> Node:
if not Engine.is_editor_hint():
return null
return EditorInterface.get_edited_scene_root()
func get_results(verbose: bool = false) -> Dictionary:
var passed := 0
var failed := 0
+53
View File
@@ -31,6 +31,34 @@ func suite_teardown() -> void:
pass
# ----- tracked allocations (freed by the runner after each test) -----
var _tracked_objects: Array[Object] = []
## Register a manually-managed Object (plain Object or out-of-tree Node) so
## the runner frees it after the current test. RefCounted instances are
## accepted but ignored because they manage their own lifetime.
func track(obj: Object) -> Object:
if obj != null and not obj is RefCounted:
_tracked_objects.append(obj)
return obj
## Free everything registered via track(). Called by the runner after each
## test's teardown() and again after suite_teardown().
func _free_tracked() -> void:
for obj in _tracked_objects:
if not is_instance_valid(obj) or (obj is Node and obj.is_queued_for_deletion()):
continue
if obj is Node:
var parent := (obj as Node).get_parent()
if parent != null:
parent.remove_child(obj)
obj.free()
_tracked_objects.clear()
# ----- assertion state (managed by McpTestRunner) -----
var _failed: bool = false
@@ -38,6 +66,7 @@ var _message: String = ""
var _assertion_count: int = 0
var _skipped: bool = false
var _skip_reason: String = ""
var _expected_script_error_substrings: Array[String] = []
# ----- suite-level state (managed by McpTestRunner) -----
@@ -53,6 +82,7 @@ func _reset() -> void:
_assertion_count = 0
_skipped = false
_skip_reason = ""
_expected_script_error_substrings.clear()
func _reset_suite_state() -> void:
@@ -121,6 +151,29 @@ func skip_on_godot_lt(min_version: String, reason: String = "") -> bool:
return false
## Allow one captured SCRIPT ERROR whose text contains `substring`.
## Use only for negative-path tests that intentionally compile or execute
## invalid GDScript and assert on the resulting diagnostics.
func expect_script_error_containing(substring: String) -> void:
_expected_script_error_substrings.append(substring)
func _unexpected_script_errors(captured: PackedStringArray) -> PackedStringArray:
var unexpected := PackedStringArray()
var remaining := _expected_script_error_substrings.duplicate()
for error in captured:
var matched_index := -1
for i in range(remaining.size()):
if error.find(remaining[i]) != -1:
matched_index = i
break
if matched_index == -1:
unexpected.append(error)
else:
remaining.remove_at(matched_index)
return unexpected
## Trigger an undo against whichever history (scene or global) holds the most
## recent action. `EditorUndoRedoManager` in Godot 4.x doesn't expose `.undo()`
## directly — you resolve the history's underlying UndoRedo and call it there.