musique/scripts/test.py
2023-02-15 01:18:35 +01:00

354 lines
12 KiB
Python
Executable File

#!/usr/bin/env python3
import argparse
import dacite
import dataclasses
import json
import os
import subprocess
import time
import typing
try:
import alsa_midi as alsa
WITH_MIDI = True
except ModuleNotFoundError as e:
print(f"ERROR: Failed to load module alsa_midi: {e}")
print(" Tests using this module will automatically fail\n"
" if not provided with --skip-midi flag")
exit(1)
TEST_DIR = "regression-tests"
TEST_DB = "test_db.json"
INTERPRETER = "bin/linux/debug/musique"
MIDI_TOLERANCE = 0.005
@dataclasses.dataclass
class MidiEvent:
type: str
args: list[str]
time: float
def connect_to_default_midi_port():
global midi_client
global port # this object keeps alive port, so it needs to live in global space (= live as long as program)
midi_client = alsa.SequencerClient('Musique Tester')
ports = midi_client.list_ports(input = True)
for p in ports:
if p.client_id == 14:
input_port = p
break
else:
assert False, "Linux default MIDI port not found"
port = midi_client.create_port('Musique tester')
port.connect_from(input_port)
def listen_for_midi_events() -> list[MidiEvent] | None:
if not WITH_MIDI:
return None
zero_time = time.monotonic()
events = []
while True:
event = midi_client.event_input(timeout=5)
if event is None:
break
end_time = time.monotonic()
events.append((event, end_time - zero_time))
return events
def normalize_events(events) -> typing.Generator[MidiEvent, None, None]:
for event, time in events:
match event:
case alsa.event.NoteOnEvent():
# TODO Support velocity
yield MidiEvent(type='note_on', args=[str(event.channel), str(event.note)], time=time)
case alsa.event.NoteOffEvent():
yield MidiEvent(type='note_off', args=[str(event.channel), str(event.note)], time=time)
case _:
assert False, f"Unmatched event type: {event.type}"
def compare_midi(xs: list[MidiEvent] | None, ys: list[MidiEvent] | None) -> bool:
if xs is None or ys is None:
return (xs is None) == (ys is None)
# TODO Can we get better performance then O(n^2) algorithm?
# Or at lexst optimize implementation of this one?
if len(xs) != len(ys):
return False
for x in xs:
if not any(x.type == y.type \
and x.args == y.args \
and x.time >= y.time - MIDI_TOLERANCE and x.time <= y.time + MIDI_TOLERANCE
for y in ys):
return False
return True
@dataclasses.dataclass
class Result:
exit_code: int = 0
stdout_lines: list[str] = dataclasses.field(default_factory=list)
stderr_lines: list[str] = dataclasses.field(default_factory=list)
midi_events: list[MidiEvent] | None = None
@dataclasses.dataclass
class TestCase(Result):
name: str = "<unnamed>"
stdin_lines: list[str] = dataclasses.field(default_factory=list)
def run(self, interpreter: str, source: str, cwd: str, *, capture_midi: bool = False) -> Result:
process = subprocess.Popen(
args=[interpreter, source, "-q"],
cwd=cwd,
text=True,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE,
stdin=subprocess.PIPE
)
midi_events = listen_for_midi_events() if capture_midi else None
stdout, stderr = process.communicate()
return Result(
exit_code=process.wait(),
stdout_lines=stdout.splitlines(keepends=False),
stderr_lines=stderr.splitlines(keepends=False),
midi_events = None if midi_events is None else \
list(normalize_events(midi_events))
)
def record(self, interpreter: str, source: str, cwd: str):
print(f"Recording case {self.name}")
result = self.run(interpreter, source, cwd, capture_midi=True)
changes = []
if self.exit_code != result.exit_code: changes.append("exit code")
if self.stderr_lines != result.stderr_lines: changes.append("stderr")
if self.stdout_lines != result.stdout_lines: changes.append("stdout")
if self.midi_events != result.midi_events: changes.append("midi")
if changes:
print(f" changed: {', '.join(changes)}")
self.exit_code = result.exit_code
self.stderr_lines = result.stderr_lines
self.stdout_lines = result.stdout_lines
self.midi_events = result.midi_events
def test(self, interpreter: str, source: str, cwd: str):
print(f" Testing case {self.name} ", end="")
capture_midi = self.midi_events is not None
result = self.run(interpreter, source, cwd, capture_midi=capture_midi)
if self.midi_events is not None:
midi_events = [dacite.from_dict(MidiEvent, event) for event in self.midi_events]
else:
midi_events = None
if self.exit_code == result.exit_code \
and self.stdout_lines == result.stdout_lines \
and self.stderr_lines == result.stderr_lines \
and compare_midi(midi_events, result.midi_events):
print("ok")
return True
print(f"FAILED")
print(f"File: {source}")
if self.exit_code != result.exit_code:
print(f"Different exit code - expected {self.exit_code}, got {result.exit_code}")
for name, expected, actual in [
("standard output", self.stdout_lines, result.stdout_lines),
("standard error", self.stderr_lines, result.stderr_lines)
]:
if expected == actual:
continue
diff_line = None
for i, (exp_line, got_line) in enumerate(zip(expected, actual)):
if exp_line != got_line:
diff_line = i
break
if diff_line is not None:
print(f"First difference at line {diff_line+1} in {name}:")
print(f" Expected: {expected[diff_line]}")
print(f" Got: {actual[diff_line]}")
elif len(expected) > len(actual):
print(f"Expected {name} is {len(expected) - len(actual)} lines longer then actual")
else:
print(f"Actual {name} is {len(actual) - len(expected)} lines longer then expected")
return False
@dataclasses.dataclass
class TestSuite:
name: str
cases: list[TestCase] = dataclasses.field(default_factory=list)
suites = list[TestSuite]()
def suite_case_from_path(path: str) -> tuple[str, str]:
path = os.path.realpath(path)
test_suite, test_case = os.path.split(path)
test_dir, test_suite = os.path.split(test_suite)
_, test_dir = os.path.split(test_dir)
assert test_dir == TEST_DIR, "Provided path doesn't follow required directory structure"
assert test_case.endswith(".mq"), "Test case is not a Musique file"
assert os.path.isfile(path), "Test case is not a file"
return (test_suite, test_case)
def add(path: str) -> list[tuple[TestSuite, TestCase]]:
test_suite, test_case = suite_case_from_path(path)
for suite in suites:
if suite.name == test_suite:
break
else:
print(f"Discovered new test suite: {test_suite}")
suite = TestSuite(name=test_suite)
suites.append(suite)
for case in suite.cases:
if case.name == test_case:
print(f"Test case {test_case} in suite {test_suite} already exists")
return []
case = TestCase(name=test_case)
suite.cases.append(case)
return [(suite, case)]
def update(path: str) -> list[tuple[TestSuite, TestCase]]:
test_suite, test_case = suite_case_from_path(path)
for suite in suites:
if suite.name == test_suite:
break
else:
print(f"Cannot update case {test_case} where suite {test_suite} was not defined yet.")
print("Use --add to add new test case")
return []
for case in suite.cases:
if case.name == test_case:
return [(suite, case)]
print(f"Case {test_case} doesn't exists in suite {test_suite}")
print("Use --add to add new test case")
return []
def testcase(path: str) -> list[tuple[TestSuite, TestCase]]:
test_suite, test_case = suite_case_from_path(path)
for suite in suites:
if suite.name == test_suite:
break
else:
print(f"Cannot test case {test_case} where suite {test_suite} was not defined yet.")
print("Use --add to add new test case")
return []
for case in suite.cases:
if case.name == test_case:
return [(suite, case)]
print(f"Case {test_case} doesn't exists in suite {test_suite}")
print("Use --add to add new test case")
return []
def traverse(discover: bool, update: bool) -> list[tuple[TestSuite, TestCase]]:
to_record = list[tuple[TestSuite, TestCase]]()
if discover:
for suite_name in os.listdir(testing_dir):
if os.path.isdir(os.path.join(testing_dir, suite_name)) and suite_name not in (suite.name for suite in suites):
print(f"Discovered new test suite: {suite_name}")
suites.append(TestSuite(name=suite_name))
for suite in suites:
suite_path = os.path.join(testing_dir, suite.name)
for case_name in os.listdir(suite_path):
if os.path.isfile(os.path.join(suite_path, case_name)) and case_name not in (case.name for case in suite.cases):
print(f"In suite '{suite.name}' discovered new test case: {case_name}")
case = TestCase(name=case_name)
suite.cases.append(case)
to_record.append((suite, case))
if update:
to_record.extend(((suite, case) for suite in suites for case in suite.cases))
return to_record
def test():
successful, total = 0, 0
for suite in suites:
print(f"Testing suite {suite.name}")
for case in suite.cases:
successful += int(case.test(
interpreter=os.path.join(root, INTERPRETER),
source=os.path.join(testing_dir, suite.name, case.name),
cwd=root
))
total += 1
print(f"Passed {successful} out of {total} ({100 * successful // total}%)")
if __name__ == "__main__":
if not os.path.exists(INTERPRETER):
subprocess.run("make debug", shell=True, check=True)
parser = argparse.ArgumentParser(description="Regression test runner for Musique programming language")
parser.add_argument("-d", "--discover", action="store_true", help="Discover all tests that are not in testing database")
parser.add_argument("--update-all", action="store_true", help="Update all tests", dest="update_all")
parser.add_argument("-a", "--add", action="append", help="Add new test to test suite", default=[])
parser.add_argument("-u", "--update", action="append", help="Update test case", default=[])
parser.add_argument('--skip-midi', action="store_true", help="Skip tests expecting MIDI communication", default=False, dest="skip_midi")
args = parser.parse_args()
WITH_MIDI = WITH_MIDI and not args.skip_midi
if WITH_MIDI:
connect_to_default_midi_port()
root = os.path.dirname(os.path.dirname(__file__))
testing_dir = os.path.join(root, TEST_DIR)
test_db_path = os.path.join(testing_dir, TEST_DB)
with open(test_db_path, "r") as f:
for src in json.load(f):
src["cases"] = [TestCase(**case) for case in src["cases"]]
suites.append(TestSuite(**src))
to_record = []
if args.discover or args.update_all:
to_record.extend(traverse(discover=args.discover, update=args.update_all))
elif not (args.add or args.update):
test()
for case in args.add:
to_record.extend(add(case))
for case in args.update:
to_record.extend(update(case))
for (suite, case) in to_record:
case.record(
interpreter=os.path.join(root, INTERPRETER),
source=os.path.join(testing_dir, suite.name, case.name),
cwd=root
)
if to_record:
with open(test_db_path, "w") as f:
json_suites = [dataclasses.asdict(suite) for suite in suites]
json.dump(json_suites, f, indent=2)