compiletest: Test execution time-stamping should take into account debugger pretty printers.

[michaelwoerister]

Modifying one of the pretty-printer python files in src/etc does "un-ignore" debuginfo tests. The relevant files should be added to the list of inputs in compiletest::up_to_date():

rust/src/tools/compiletest/src/main.rs

Lines 491 to 507 in fd8099f

	fn up_to_date(config: &Config, testpaths: &TestPaths, props: &EarlyProps) -> bool {
	let stamp = mtime(&stamp(config, testpaths));
	let mut inputs = vec![
	mtime(&testpaths.file),
	mtime(&config.rustc_path),
	];
	for aux in props.aux.iter() {
	inputs.push(mtime(&testpaths.file.parent().unwrap()
	.join("auxiliary")
	.join(aux)));
	}
	for lib in config.run_lib_path.read_dir().unwrap() {
	let lib = lib.unwrap();
	inputs.push(mtime(&lib.path()));
	}
	inputs.iter().any(|input| *input > stamp)
	}

Bonus points for exiting from that function immediately when finding something out-of-date.

[k0pernicus]

Maybe I can help here as a first contribution to the programming language.
I will try to propose a solution for tomorrow evening.

[k0pernicus]

So, I have read the code and I tried to understand where is the problem, and what kind of solution I can propose.

This is my thought:

1. The problem is pretty-printer Python files in src/etc are not considered in up_to_date; so, if you modified one of those files, you will not ignoring them (am I right?).
2. You propose to add the relevant files (debugger_pretty_printers_common.py, gdb_load_rust_pretty_printers.py, and gdb_rust_pretty_printers.py I think) as inputs to up_to_date, to consider them as "important files to not ignore" (am I right?).
3. In order to include the pretty printer files, I have to include them in TestPaths, right?

Thanks for those responses.

[michaelwoerister]

@k0pernicus You're on the right track:

1. Almost right. When the compiletest tool executes a test and that test passes, it will not re-execute it until something relevant has changed. "Not re-executing" in practice means that the test will show up as ignored in the commandline output. The problem here is that modifications to the pretty printer files are not picked up as "relevant changes". As a consequence the tests stay "ignored" even though they should actually be re-run because the changes to the pretty printer scripts can influence their result.
2. Correct. inputs is a list of all things that must not have changed since the last test execution if that test is to stay ignored. If any file from inputs has a modification time that is newer than the last test execution, the test has to be re-executed.
3. Not quite. TestPaths contains the location of the test that is being executed. What we want here is to just directly add the pretty-printer script paths to the inputs list. Something like

let rust_src_dir = ...; 
inputs.push(rust_src_dir.join("src/etc/debugger_pretty_printers_common.py"));
inputs.push(rust_src_dir.join("src/etc/gdb_load_rust_pretty_printers.py"));
...

Unfortunately the rust_src_dir is not readily available. But you could move the runtest::TestCx::find_rust_src_root() to common::Config and then use that for finding it.

[k0pernicus]

Ok, I understand.
Thanks a lot for these explanations.

I will work on that issue after work, in order to submit a PR + code review for tomorrow morning.

[k0pernicus]

@michaelwoerister I just created the PR to share the code.
I compiled the project and I ran the test (using ./x.py test).
Unfortunately, I don't know how to test my changes (which is why I put a "WiP")...

Is it possible to obtain a simple example of "procedure", in order to test my changes please?
Thanks a lot.