Use pytest hooks for test execution instead of "scraping" the JUnit XML results.

[karrtikr]

Couldn't find an open issue, if we have one feel free to close this one against it

Note we already use hooks for test discovery, this issue for test execution. The output on executing tests is not stable, so the assumptions we make when scraping output can break things. (For eg. #12403, pytest-dev/pytest#7463, #7265)

API details: pytest-dev/pytest#7463 (comment)

Similar issue for unittest framework: #10119

[DonJayamanne]

Here you go #1132
I thought we are doing that today

[karrtikr]

@DonJayamanne This one is for execution, not discovery. We already do that for discovery.

[luabud]

@karrtikr I believe we don't scrape out the output from execution, do we?
cc @ericsnowcurrently

[karrtikr]

We do scrape out the output from execution, which is something we probably shouldn't.

[ericsnowcurrently]

Are you talking about what goes on in TestRunner.run() and it's use of execObservable()?

perhaps particularly relevant:

vscode-python/src/client/testing/common/runner.ts

Lines 75 to 103 in c37c79d

	result.out.subscribe(
	(output) => {
	stdOut += output.out;
	// If the test runner python module is not installed we'll have something in stderr.
	// Hence track that separately and check at the end.
	if (output.source === 'stderr') {
	stdErr += output.out;
	}
	if (options.outChannel) {
	options.outChannel.append(output.out);
	}
	},
	reject,
	async () => {
	// If the test runner python module is not installed we'll have something in stderr.
	if (
	moduleName &&
	pythonExecutionServicePromise &&
	ErrorUtils.outputHasModuleNotInstalledError(moduleName, stdErr)
	) {
	const pythonExecutionService = await pythonExecutionServicePromise;
	const isInstalled = await pythonExecutionService.isModuleInstalled(moduleName);
	if (!isInstalled) {
	return reject(new ModuleNotInstalledError(moduleName));
	}
	}
	resolve(stdOut);
	}
	);

Regardless, it sounds like you have specific code in mind. Where is it?

[karrtikr]

I had initially encountered a very particular place where I found the code was scraping the testFunction.testFunction.traceback to figure out the test filePath:

vscode-python/src/client/testing/pytest/services/testMessageService.ts

Lines 90 to 97 in a98be63

	if (testFunction.testFunction.traceback) {
	const fileMatches =
	testFunction.testFunction.traceback.match(/^((\.\.[\\\/])*.+\.py)\:(\d+)\:.*$/gim) || [];
	for (const fileDetailsMatch of fileMatches) {
	const fileDetails = fileDetailsMatch.split(':');
	let filePath = fileDetails[0];
	filePath = path.isAbsolute(filePath) ? filePath : path.resolve(rootDirectory, filePath);
	const fileUri = Uri.file(filePath);

And secondly, yes, I saw what goes in TestRunner.run() and saw we execute module pytest. We collect results in a JUnit XML file, which is then parsed here:

vscode-python/src/client/testing/common/xUnitParser.ts

Lines 82 to 100 in 61b179b

	function getJunitResults(parserResult: any): TestSuiteResult | undefined {
	// This is the newer JUnit XML format (e.g. pytest 5.1 and later).
	const fullResults = parserResult as { testsuites: { testsuite: TestSuiteResult[] } };
	if (!fullResults.testsuites) {
	return (parserResult as { testsuite: TestSuiteResult }).testsuite;
	}
	const junitSuites = fullResults.testsuites.testsuite;
	if (!Array.isArray(junitSuites)) {
	throw Error('bad JUnit XML data');
	}
	if (junitSuites.length === 0) {
	return;
	}
	if (junitSuites.length > 1) {
	throw Error('got multiple XML results');
	}
	return junitSuites[0];
	}

We don't scrape terminal output so it's not as bad, but JUnit XML format can change as well, happened once to us here: #6990 and broke running tests for everybody.