CMake: Run finalizers for test-like executables on all platforms

Previously we ran some finalizer functions for tests and manual
tests, only for specific platforms, and via different code paths.

This change introduces a new unified way of running all finalizers for
all test-like executables, including benchmarks.

To ensure a smoother transition, the new way is opt-out, and the old
way can be enabled by setting the QT_INTERNAL_SKIP_TEST_FINALIZERS_V2
variable to true in case we encounter some issues in the CI.

The finalizers are only run for test-like executables, and not all
internal executables, because there are some unsolved issues there.

One particular case is in qtdeclarative where that will create a cycle
for qmlimportscanner to depend on itself.
A proper solution here would be to have some kind of mapping or
mechanism to exclude finalizers for targets where they would try to
run themselves.

Pick-to: 6.8
Task-number: QTBUG-93625
Task-number: QTBUG-112212
Change-Id: I52b3a1c02c298c4a18ce2c75d7e491ae79d191a0
Reviewed-by: Alexey Edelev <[email protected]>
Reviewed-by: Joerg Bornemann <[email protected]>

cmake/QtExecutableHelpers.cmake

@@ -38,11 +38,37 @@ function(qt_internal_add_executable name)
         _qt_is_benchmark_test ${arg_QT_BENCHMARK_TEST}
     )
 
-    if(ANDROID)
-        _qt_internal_android_executable_finalizer(${name})
-    endif()
-    if(WASM)
-        qt_internal_wasm_add_finalizers(${name})
+    # Opt out to skip the new way of running test finalizers, and instead use the old way for
+    # specific platforms.
+    # TODO: Remove once we confirm that the new way of running test finalizers for all platforms
+    # doesn't cause any issues.
+    if(NOT QT_INTERNAL_SKIP_TEST_FINALIZERS_V2)
+        # We don't run finalizers for all executables on all platforms, because there are still
+        # some unsolved issues there. One of them is trying to run finalizers for the
+        # qmlimportscanner executable would create a circular depenendecy trying to run
+        # qmlimportscanner on itself.
+        #
+        # For now, we only run finalizers for test-like executables on all platforms, and all
+        # android and wasm internal executables.
+        # For android and wasm all executables, to be behavior compatible with the old way of
+        # running finalizers.
+        if(ANDROID
+            OR WASM
+            OR arg_QT_TEST
+            OR arg_QT_MANUAL_TEST
+            OR arg_QT_BENCHMARK_TEST)
+            set(QT_INTERNAL_USE_POOR_MANS_SCOPE_FINALIZER TRUE)
+            _qt_internal_finalize_target_defer("${name}")
+        endif()
+    else()
+        if(ANDROID)
+            # This direct calls the finalizer, which in the v2 way is deferred.
+            _qt_internal_android_executable_finalizer(${name})
+        endif()
+        if(WASM)
+            # This defer calls the finalizer.
+            qt_internal_wasm_add_finalizers(${name})
+        endif()
     endif()
 
     if(arg_QT_APP AND QT_FEATURE_debug_and_release AND CMAKE_VERSION VERSION_GREATER_EQUAL "3.19.0")

cmake/QtTestHelpers.cmake

@@ -1126,6 +1126,14 @@ function(qt_internal_collect_command_environment out_path out_plugin_path)
 endfunction()
 
 function(qt_internal_add_test_finalizers target)
+    # Opt out to skip the new way of running test finalizers, and instead use the old way for
+    # specific platforms.
+    # TODO: Remove once we confirm that the new way of running test finalizers for all platforms
+    # doesn't cause any issues.
+    if(QT_INTERNAL_SKIP_TEST_FINALIZERS_V2)
+        return()
+    endif()
+
     # It might not be safe to run all the finalizers of _qt_internal_finalize_executable
     # within the context of a Qt build (not a user project) when targeting a host build.
     # At least one issue is missing qmlimportscanner at configure time.