[Build System] Remove the use of sub-shells when running test targets

test/CMakeLists.txt

@@ -330,6 +330,7 @@ foreach(SDK ${SWIFT_SDKS})
               ${LIT_ARGS}
               "--param" "swift_test_subset=${test_subset}"
               "--param" "swift_test_mode=${test_mode}"
+              ${SWIFT_LIT_TEST_PATHS}
             DEPENDS ${dependencies}
             COMMENT "Running ${test_subset} Swift tests for ${VARIANT_TRIPLE} from custom test locations"
             USES_TERMINAL)

utils/build-script

@@ -290,16 +290,6 @@ class BuildScriptInvocation(object):
             args.build_subdir = \
                 workspace.compute_build_subdir(args)
 
-        # --test-paths implies --test and/or --validation-test
-        # depending on what directories/files have been specified.
-        if args.test_paths:
-            for path in args.test_paths:
-                if path.startswith('test'):
-                    args.test = True
-                elif path.startswith('validation-test'):
-                    args.test = True
-                    args.validation_test = True
-
         # Add optional stdlib-deployment-targets
         if args.android:
             args.stdlib_deployment_targets.append(

utils/build-script-impl

@@ -1954,6 +1954,21 @@ for host in "${ALL_HOSTS[@]}"; do
         -DSWIFT_RUNTIME_ENABLE_LEAK_CHECKER:BOOL=$(true_false "${SWIFT_RUNTIME_ENABLE_LEAK_CHECKER}")
     )
 
+    if [[ "${TEST_PATHS}" ]]; then
+        build_dir=$(build_directory ${host} "swift")
+
+        test_paths=()
+        for path in ${TEST_PATHS}; do
+            test_paths+=(
+                "${build_dir}/$(echo ${path} | sed -E "s/^(validation-test|test)/\1-${host}/")"
+            )
+        done
+
+        swift_cmake_options=(
+            "${swift_cmake_options[@]}"
+            -DSWIFT_LIT_TEST_PATHS="${test_paths}")
+    fi
+
     if [[ "${SKIP_TEST_SOURCEKIT}" ]] ; then
         swift_cmake_options=(
             "${swift_cmake_options[@]}"
@@ -3174,43 +3189,12 @@ for host in "${ALL_HOSTS[@]}"; do
                 trap "tests_busted ${product} '(${target})'" ERR
 
                 test_target="$target"
-                test_paths=()
-
-                if [[ ${test_target} == check-swift* ]]; then
-                    if [[ "${TEST_PATHS}" ]]; then
-                        test_target="${test_target}-custom"
-                        for path in ${TEST_PATHS}; do
-                            test_paths+=(
-                                "${build_dir}/$(echo ${path} | sed -E "s/^(validation-test|test)/\1-${host}/")"
-                            )
-                        done
-                    fi
+                if [[ ${test_target} == check-swift* ]] && [[ "${TEST_PATHS}" ]]; then
+                    test_target="${test_target}-custom"
                 fi
 
-                # NOTE: In dry-run mode, build_dir might not exist yet. In that
-                #       case, -n query will fail. So, in dry-run mode,
-                #       we don't expand test script.
-                if [[ ! "${DRY_RUN}" && "${CMAKE_GENERATOR}" == Ninja ]] && !( "${build_cmd[@]}" --version 2>&1 | grep -i -q llbuild ) && [[ -z "${DISTCC_PUMP}" ]]; then
-                    # Ninja buffers command output to avoid scrambling the output
-                    # of parallel jobs, which is awesome... except that it
-                    # interferes with the progress meter when testing.  Instead of
-                    # executing ninja directly, have it dump the commands it would
-                    # run, strip Ninja's progress prefix with sed, and tell the
-                    # shell to execute that.
-                    # However, if we do this in a subshell in the ``sh -e -x -c`` line,
-                    # errors in the command will not stop the script as they should.
-                    echo "Generating dry run test command from: ${build_cmd[@]} -n -v ${test_target}"
-                    dry_run_command_output="$(${build_cmd[@]} -n -v ${test_target} | sed -e 's/[^]]*] //')"
-                    echo "Test command: ${dry_run_command_output}"
-
-                    if [[ ! "${test_paths}" ]]; then
-                        env bash -ex <(echo -e "${dry_run_command_output}")
-                    else
-                        env bash -ex <(echo -e "${dry_run_command_output}" "${test_paths[@]}")
-                    fi
-                else
-                    call "${build_cmd[@]}" ${BUILD_TARGET_FLAG} ${test_target}
-                fi
+                call "${build_cmd[@]}" ${BUILD_TARGET_FLAG} ${test_target}
+
                 echo "-- ${target} finished --"
             fi
         done

utils/build_swift/driver_arguments.py

@@ -166,6 +166,16 @@ def _apply_default_arguments(args):
     if not args.android or not args.build_android:
         args.build_android = False
 
+    # --test-paths implies --test and/or --validation-test
+    # depending on what directories/files have been specified.
+    if args.test_paths:
+        for path in args.test_paths:
+            if path.startswith('test'):
+                args.test = True
+            elif path.startswith('validation-test'):
+                args.test = True
+                args.validation_test = True
+
     # --validation-test implies --test.
     if args.validation_test:
         args.test = True