Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1# SPDX-License-Identifier: GPL-2.0
2#
3# Parses KTAP test results from a kernel dmesg log and incrementally prints
4# results with reader-friendly format. Stores and returns test results in a
5# Test object.
6#
7# Copyright (C) 2019, Google LLC.
8# Author: Felix Guo <felixguoxiuping@gmail.com>
9# Author: Brendan Higgins <brendanhiggins@google.com>
10# Author: Rae Moar <rmoar@google.com>
11
12from __future__ import annotations
13import re
14import sys
15
16from enum import Enum, auto
17from typing import Iterable, Iterator, List, Optional, Tuple
18
19from kunit_printer import stdout
20
21class Test:
22 """
23 A class to represent a test parsed from KTAP results. All KTAP
24 results within a test log are stored in a main Test object as
25 subtests.
26
27 Attributes:
28 status : TestStatus - status of the test
29 name : str - name of the test
30 expected_count : int - expected number of subtests (0 if single
31 test case and None if unknown expected number of subtests)
32 subtests : List[Test] - list of subtests
33 log : List[str] - log of KTAP lines that correspond to the test
34 counts : TestCounts - counts of the test statuses and errors of
35 subtests or of the test itself if the test is a single
36 test case.
37 """
38 def __init__(self) -> None:
39 """Creates Test object with default attributes."""
40 self.status = TestStatus.TEST_CRASHED
41 self.name = ''
42 self.expected_count = 0 # type: Optional[int]
43 self.subtests = [] # type: List[Test]
44 self.log = [] # type: List[str]
45 self.counts = TestCounts()
46
47 def __str__(self) -> str:
48 """Returns string representation of a Test class object."""
49 return (f'Test({self.status}, {self.name}, {self.expected_count}, '
50 f'{self.subtests}, {self.log}, {self.counts})')
51
52 def __repr__(self) -> str:
53 """Returns string representation of a Test class object."""
54 return str(self)
55
56 def add_error(self, error_message: str) -> None:
57 """Records an error that occurred while parsing this test."""
58 self.counts.errors += 1
59 stdout.print_with_timestamp(stdout.red('[ERROR]') + f' Test: {self.name}: {error_message}')
60
61class TestStatus(Enum):
62 """An enumeration class to represent the status of a test."""
63 SUCCESS = auto()
64 FAILURE = auto()
65 SKIPPED = auto()
66 TEST_CRASHED = auto()
67 NO_TESTS = auto()
68 FAILURE_TO_PARSE_TESTS = auto()
69
70class TestCounts:
71 """
72 Tracks the counts of statuses of all test cases and any errors within
73 a Test.
74
75 Attributes:
76 passed : int - the number of tests that have passed
77 failed : int - the number of tests that have failed
78 crashed : int - the number of tests that have crashed
79 skipped : int - the number of tests that have skipped
80 errors : int - the number of errors in the test and subtests
81 """
82 def __init__(self):
83 """Creates TestCounts object with counts of all test
84 statuses and test errors set to 0.
85 """
86 self.passed = 0
87 self.failed = 0
88 self.crashed = 0
89 self.skipped = 0
90 self.errors = 0
91
92 def __str__(self) -> str:
93 """Returns the string representation of a TestCounts object."""
94 statuses = [('passed', self.passed), ('failed', self.failed),
95 ('crashed', self.crashed), ('skipped', self.skipped),
96 ('errors', self.errors)]
97 return f'Ran {self.total()} tests: ' + \
98 ', '.join(f'{s}: {n}' for s, n in statuses if n > 0)
99
100 def total(self) -> int:
101 """Returns the total number of test cases within a test
102 object, where a test case is a test with no subtests.
103 """
104 return (self.passed + self.failed + self.crashed +
105 self.skipped)
106
107 def add_subtest_counts(self, counts: TestCounts) -> None:
108 """
109 Adds the counts of another TestCounts object to the current
110 TestCounts object. Used to add the counts of a subtest to the
111 parent test.
112
113 Parameters:
114 counts - a different TestCounts object whose counts
115 will be added to the counts of the TestCounts object
116 """
117 self.passed += counts.passed
118 self.failed += counts.failed
119 self.crashed += counts.crashed
120 self.skipped += counts.skipped
121 self.errors += counts.errors
122
123 def get_status(self) -> TestStatus:
124 """Returns the aggregated status of a Test using test
125 counts.
126 """
127 if self.total() == 0:
128 return TestStatus.NO_TESTS
129 if self.crashed:
130 # Crashes should take priority.
131 return TestStatus.TEST_CRASHED
132 if self.failed:
133 return TestStatus.FAILURE
134 if self.passed:
135 # No failures or crashes, looks good!
136 return TestStatus.SUCCESS
137 # We have only skipped tests.
138 return TestStatus.SKIPPED
139
140 def add_status(self, status: TestStatus) -> None:
141 """Increments the count for `status`."""
142 if status == TestStatus.SUCCESS:
143 self.passed += 1
144 elif status == TestStatus.FAILURE:
145 self.failed += 1
146 elif status == TestStatus.SKIPPED:
147 self.skipped += 1
148 elif status != TestStatus.NO_TESTS:
149 self.crashed += 1
150
151class LineStream:
152 """
153 A class to represent the lines of kernel output.
154 Provides a lazy peek()/pop() interface over an iterator of
155 (line#, text).
156 """
157 _lines: Iterator[Tuple[int, str]]
158 _next: Tuple[int, str]
159 _need_next: bool
160 _done: bool
161
162 def __init__(self, lines: Iterator[Tuple[int, str]]):
163 """Creates a new LineStream that wraps the given iterator."""
164 self._lines = lines
165 self._done = False
166 self._need_next = True
167 self._next = (0, '')
168
169 def _get_next(self) -> None:
170 """Advances the LineSteam to the next line, if necessary."""
171 if not self._need_next:
172 return
173 try:
174 self._next = next(self._lines)
175 except StopIteration:
176 self._done = True
177 finally:
178 self._need_next = False
179
180 def peek(self) -> str:
181 """Returns the current line, without advancing the LineStream.
182 """
183 self._get_next()
184 return self._next[1]
185
186 def pop(self) -> str:
187 """Returns the current line and advances the LineStream to
188 the next line.
189 """
190 s = self.peek()
191 if self._done:
192 raise ValueError(f'LineStream: going past EOF, last line was {s}')
193 self._need_next = True
194 return s
195
196 def __bool__(self) -> bool:
197 """Returns True if stream has more lines."""
198 self._get_next()
199 return not self._done
200
201 # Only used by kunit_tool_test.py.
202 def __iter__(self) -> Iterator[str]:
203 """Empties all lines stored in LineStream object into
204 Iterator object and returns the Iterator object.
205 """
206 while bool(self):
207 yield self.pop()
208
209 def line_number(self) -> int:
210 """Returns the line number of the current line."""
211 self._get_next()
212 return self._next[0]
213
214# Parsing helper methods:
215
216KTAP_START = re.compile(r'KTAP version ([0-9]+)$')
217TAP_START = re.compile(r'TAP version ([0-9]+)$')
218KTAP_END = re.compile('(List of all partitions:|'
219 'Kernel panic - not syncing: VFS:|reboot: System halted)')
220
221def extract_tap_lines(kernel_output: Iterable[str], lstrip=True) -> LineStream:
222 """Extracts KTAP lines from the kernel output."""
223 def isolate_ktap_output(kernel_output: Iterable[str]) \
224 -> Iterator[Tuple[int, str]]:
225 line_num = 0
226 started = False
227 for line in kernel_output:
228 line_num += 1
229 line = line.rstrip() # remove trailing \n
230 if not started and KTAP_START.search(line):
231 # start extracting KTAP lines and set prefix
232 # to number of characters before version line
233 prefix_len = len(
234 line.split('KTAP version')[0])
235 started = True
236 yield line_num, line[prefix_len:]
237 elif not started and TAP_START.search(line):
238 # start extracting KTAP lines and set prefix
239 # to number of characters before version line
240 prefix_len = len(line.split('TAP version')[0])
241 started = True
242 yield line_num, line[prefix_len:]
243 elif started and KTAP_END.search(line):
244 # stop extracting KTAP lines
245 break
246 elif started:
247 # remove the prefix and optionally any leading
248 # whitespace. Our parsing logic relies on this.
249 line = line[prefix_len:]
250 if lstrip:
251 line = line.lstrip()
252 yield line_num, line
253 return LineStream(lines=isolate_ktap_output(kernel_output))
254
255KTAP_VERSIONS = [1]
256TAP_VERSIONS = [13, 14]
257
258def check_version(version_num: int, accepted_versions: List[int],
259 version_type: str, test: Test) -> None:
260 """
261 Adds error to test object if version number is too high or too
262 low.
263
264 Parameters:
265 version_num - The inputted version number from the parsed KTAP or TAP
266 header line
267 accepted_version - List of accepted KTAP or TAP versions
268 version_type - 'KTAP' or 'TAP' depending on the type of
269 version line.
270 test - Test object for current test being parsed
271 """
272 if version_num < min(accepted_versions):
273 test.add_error(f'{version_type} version lower than expected!')
274 elif version_num > max(accepted_versions):
275 test.add_error(f'{version_type} version higer than expected!')
276
277def parse_ktap_header(lines: LineStream, test: Test) -> bool:
278 """
279 Parses KTAP/TAP header line and checks version number.
280 Returns False if fails to parse KTAP/TAP header line.
281
282 Accepted formats:
283 - 'KTAP version [version number]'
284 - 'TAP version [version number]'
285
286 Parameters:
287 lines - LineStream of KTAP output to parse
288 test - Test object for current test being parsed
289
290 Return:
291 True if successfully parsed KTAP/TAP header line
292 """
293 ktap_match = KTAP_START.match(lines.peek())
294 tap_match = TAP_START.match(lines.peek())
295 if ktap_match:
296 version_num = int(ktap_match.group(1))
297 check_version(version_num, KTAP_VERSIONS, 'KTAP', test)
298 elif tap_match:
299 version_num = int(tap_match.group(1))
300 check_version(version_num, TAP_VERSIONS, 'TAP', test)
301 else:
302 return False
303 test.log.append(lines.pop())
304 return True
305
306TEST_HEADER = re.compile(r'^# Subtest: (.*)$')
307
308def parse_test_header(lines: LineStream, test: Test) -> bool:
309 """
310 Parses test header and stores test name in test object.
311 Returns False if fails to parse test header line.
312
313 Accepted format:
314 - '# Subtest: [test name]'
315
316 Parameters:
317 lines - LineStream of KTAP output to parse
318 test - Test object for current test being parsed
319
320 Return:
321 True if successfully parsed test header line
322 """
323 match = TEST_HEADER.match(lines.peek())
324 if not match:
325 return False
326 test.log.append(lines.pop())
327 test.name = match.group(1)
328 return True
329
330TEST_PLAN = re.compile(r'1\.\.([0-9]+)')
331
332def parse_test_plan(lines: LineStream, test: Test) -> bool:
333 """
334 Parses test plan line and stores the expected number of subtests in
335 test object. Reports an error if expected count is 0.
336 Returns False and sets expected_count to None if there is no valid test
337 plan.
338
339 Accepted format:
340 - '1..[number of subtests]'
341
342 Parameters:
343 lines - LineStream of KTAP output to parse
344 test - Test object for current test being parsed
345
346 Return:
347 True if successfully parsed test plan line
348 """
349 match = TEST_PLAN.match(lines.peek())
350 if not match:
351 test.expected_count = None
352 return False
353 test.log.append(lines.pop())
354 expected_count = int(match.group(1))
355 test.expected_count = expected_count
356 return True
357
358TEST_RESULT = re.compile(r'^(ok|not ok) ([0-9]+) (- )?([^#]*)( # .*)?$')
359
360TEST_RESULT_SKIP = re.compile(r'^(ok|not ok) ([0-9]+) (- )?(.*) # SKIP(.*)$')
361
362def peek_test_name_match(lines: LineStream, test: Test) -> bool:
363 """
364 Matches current line with the format of a test result line and checks
365 if the name matches the name of the current test.
366 Returns False if fails to match format or name.
367
368 Accepted format:
369 - '[ok|not ok] [test number] [-] [test name] [optional skip
370 directive]'
371
372 Parameters:
373 lines - LineStream of KTAP output to parse
374 test - Test object for current test being parsed
375
376 Return:
377 True if matched a test result line and the name matching the
378 expected test name
379 """
380 line = lines.peek()
381 match = TEST_RESULT.match(line)
382 if not match:
383 return False
384 name = match.group(4)
385 return name == test.name
386
387def parse_test_result(lines: LineStream, test: Test,
388 expected_num: int) -> bool:
389 """
390 Parses test result line and stores the status and name in the test
391 object. Reports an error if the test number does not match expected
392 test number.
393 Returns False if fails to parse test result line.
394
395 Note that the SKIP directive is the only direction that causes a
396 change in status.
397
398 Accepted format:
399 - '[ok|not ok] [test number] [-] [test name] [optional skip
400 directive]'
401
402 Parameters:
403 lines - LineStream of KTAP output to parse
404 test - Test object for current test being parsed
405 expected_num - expected test number for current test
406
407 Return:
408 True if successfully parsed a test result line.
409 """
410 line = lines.peek()
411 match = TEST_RESULT.match(line)
412 skip_match = TEST_RESULT_SKIP.match(line)
413
414 # Check if line matches test result line format
415 if not match:
416 return False
417 test.log.append(lines.pop())
418
419 # Set name of test object
420 if skip_match:
421 test.name = skip_match.group(4)
422 else:
423 test.name = match.group(4)
424
425 # Check test num
426 num = int(match.group(2))
427 if num != expected_num:
428 test.add_error(f'Expected test number {expected_num} but found {num}')
429
430 # Set status of test object
431 status = match.group(1)
432 if skip_match:
433 test.status = TestStatus.SKIPPED
434 elif status == 'ok':
435 test.status = TestStatus.SUCCESS
436 else:
437 test.status = TestStatus.FAILURE
438 return True
439
440def parse_diagnostic(lines: LineStream) -> List[str]:
441 """
442 Parse lines that do not match the format of a test result line or
443 test header line and returns them in list.
444
445 Line formats that are not parsed:
446 - '# Subtest: [test name]'
447 - '[ok|not ok] [test number] [-] [test name] [optional skip
448 directive]'
449
450 Parameters:
451 lines - LineStream of KTAP output to parse
452
453 Return:
454 Log of diagnostic lines
455 """
456 log = [] # type: List[str]
457 while lines and not TEST_RESULT.match(lines.peek()) and not \
458 TEST_HEADER.match(lines.peek()):
459 log.append(lines.pop())
460 return log
461
462
463# Printing helper methods:
464
465DIVIDER = '=' * 60
466
467def format_test_divider(message: str, len_message: int) -> str:
468 """
469 Returns string with message centered in fixed width divider.
470
471 Example:
472 '===================== message example ====================='
473
474 Parameters:
475 message - message to be centered in divider line
476 len_message - length of the message to be printed such that
477 any characters of the color codes are not counted
478
479 Return:
480 String containing message centered in fixed width divider
481 """
482 default_count = 3 # default number of dashes
483 len_1 = default_count
484 len_2 = default_count
485 difference = len(DIVIDER) - len_message - 2 # 2 spaces added
486 if difference > 0:
487 # calculate number of dashes for each side of the divider
488 len_1 = int(difference / 2)
489 len_2 = difference - len_1
490 return ('=' * len_1) + f' {message} ' + ('=' * len_2)
491
492def print_test_header(test: Test) -> None:
493 """
494 Prints test header with test name and optionally the expected number
495 of subtests.
496
497 Example:
498 '=================== example (2 subtests) ==================='
499
500 Parameters:
501 test - Test object representing current test being printed
502 """
503 message = test.name
504 if test.expected_count:
505 if test.expected_count == 1:
506 message += ' (1 subtest)'
507 else:
508 message += f' ({test.expected_count} subtests)'
509 stdout.print_with_timestamp(format_test_divider(message, len(message)))
510
511def print_log(log: Iterable[str]) -> None:
512 """Prints all strings in saved log for test in yellow."""
513 for m in log:
514 stdout.print_with_timestamp(stdout.yellow(m))
515
516def format_test_result(test: Test) -> str:
517 """
518 Returns string with formatted test result with colored status and test
519 name.
520
521 Example:
522 '[PASSED] example'
523
524 Parameters:
525 test - Test object representing current test being printed
526
527 Return:
528 String containing formatted test result
529 """
530 if test.status == TestStatus.SUCCESS:
531 return stdout.green('[PASSED] ') + test.name
532 if test.status == TestStatus.SKIPPED:
533 return stdout.yellow('[SKIPPED] ') + test.name
534 if test.status == TestStatus.NO_TESTS:
535 return stdout.yellow('[NO TESTS RUN] ') + test.name
536 if test.status == TestStatus.TEST_CRASHED:
537 print_log(test.log)
538 return stdout.red('[CRASHED] ') + test.name
539 print_log(test.log)
540 return stdout.red('[FAILED] ') + test.name
541
542def print_test_result(test: Test) -> None:
543 """
544 Prints result line with status of test.
545
546 Example:
547 '[PASSED] example'
548
549 Parameters:
550 test - Test object representing current test being printed
551 """
552 stdout.print_with_timestamp(format_test_result(test))
553
554def print_test_footer(test: Test) -> None:
555 """
556 Prints test footer with status of test.
557
558 Example:
559 '===================== [PASSED] example ====================='
560
561 Parameters:
562 test - Test object representing current test being printed
563 """
564 message = format_test_result(test)
565 stdout.print_with_timestamp(format_test_divider(message,
566 len(message) - stdout.color_len()))
567
568def print_summary_line(test: Test) -> None:
569 """
570 Prints summary line of test object. Color of line is dependent on
571 status of test. Color is green if test passes, yellow if test is
572 skipped, and red if the test fails or crashes. Summary line contains
573 counts of the statuses of the tests subtests or the test itself if it
574 has no subtests.
575
576 Example:
577 "Testing complete. Passed: 2, Failed: 0, Crashed: 0, Skipped: 0,
578 Errors: 0"
579
580 test - Test object representing current test being printed
581 """
582 if test.status == TestStatus.SUCCESS:
583 color = stdout.green
584 elif test.status in (TestStatus.SKIPPED, TestStatus.NO_TESTS):
585 color = stdout.yellow
586 else:
587 color = stdout.red
588 stdout.print_with_timestamp(color(f'Testing complete. {test.counts}'))
589
590# Other methods:
591
592def bubble_up_test_results(test: Test) -> None:
593 """
594 If the test has subtests, add the test counts of the subtests to the
595 test and check if any of the tests crashed and if so set the test
596 status to crashed. Otherwise if the test has no subtests add the
597 status of the test to the test counts.
598
599 Parameters:
600 test - Test object for current test being parsed
601 """
602 subtests = test.subtests
603 counts = test.counts
604 status = test.status
605 for t in subtests:
606 counts.add_subtest_counts(t.counts)
607 if counts.total() == 0:
608 counts.add_status(status)
609 elif test.counts.get_status() == TestStatus.TEST_CRASHED:
610 test.status = TestStatus.TEST_CRASHED
611
612def parse_test(lines: LineStream, expected_num: int, log: List[str]) -> Test:
613 """
614 Finds next test to parse in LineStream, creates new Test object,
615 parses any subtests of the test, populates Test object with all
616 information (status, name) about the test and the Test objects for
617 any subtests, and then returns the Test object. The method accepts
618 three formats of tests:
619
620 Accepted test formats:
621
622 - Main KTAP/TAP header
623
624 Example:
625
626 KTAP version 1
627 1..4
628 [subtests]
629
630 - Subtest header line
631
632 Example:
633
634 # Subtest: name
635 1..3
636 [subtests]
637 ok 1 name
638
639 - Test result line
640
641 Example:
642
643 ok 1 - test
644
645 Parameters:
646 lines - LineStream of KTAP output to parse
647 expected_num - expected test number for test to be parsed
648 log - list of strings containing any preceding diagnostic lines
649 corresponding to the current test
650
651 Return:
652 Test object populated with characteristics and any subtests
653 """
654 test = Test()
655 test.log.extend(log)
656 parent_test = False
657 main = parse_ktap_header(lines, test)
658 if main:
659 # If KTAP/TAP header is found, attempt to parse
660 # test plan
661 test.name = "main"
662 parse_test_plan(lines, test)
663 parent_test = True
664 else:
665 # If KTAP/TAP header is not found, test must be subtest
666 # header or test result line so parse attempt to parser
667 # subtest header
668 parent_test = parse_test_header(lines, test)
669 if parent_test:
670 # If subtest header is found, attempt to parse
671 # test plan and print header
672 parse_test_plan(lines, test)
673 print_test_header(test)
674 expected_count = test.expected_count
675 subtests = []
676 test_num = 1
677 while parent_test and (expected_count is None or test_num <= expected_count):
678 # Loop to parse any subtests.
679 # Break after parsing expected number of tests or
680 # if expected number of tests is unknown break when test
681 # result line with matching name to subtest header is found
682 # or no more lines in stream.
683 sub_log = parse_diagnostic(lines)
684 sub_test = Test()
685 if not lines or (peek_test_name_match(lines, test) and
686 not main):
687 if expected_count and test_num <= expected_count:
688 # If parser reaches end of test before
689 # parsing expected number of subtests, print
690 # crashed subtest and record error
691 test.add_error('missing expected subtest!')
692 sub_test.log.extend(sub_log)
693 test.counts.add_status(
694 TestStatus.TEST_CRASHED)
695 print_test_result(sub_test)
696 else:
697 test.log.extend(sub_log)
698 break
699 else:
700 sub_test = parse_test(lines, test_num, sub_log)
701 subtests.append(sub_test)
702 test_num += 1
703 test.subtests = subtests
704 if not main:
705 # If not main test, look for test result line
706 test.log.extend(parse_diagnostic(lines))
707 if (parent_test and peek_test_name_match(lines, test)) or \
708 not parent_test:
709 parse_test_result(lines, test, expected_num)
710 else:
711 test.add_error('missing subtest result line!')
712
713 # Check for there being no tests
714 if parent_test and len(subtests) == 0:
715 # Don't override a bad status if this test had one reported.
716 # Assumption: no subtests means CRASHED is from Test.__init__()
717 if test.status in (TestStatus.TEST_CRASHED, TestStatus.SUCCESS):
718 test.status = TestStatus.NO_TESTS
719 test.add_error('0 tests run!')
720
721 # Add statuses to TestCounts attribute in Test object
722 bubble_up_test_results(test)
723 if parent_test and not main:
724 # If test has subtests and is not the main test object, print
725 # footer.
726 print_test_footer(test)
727 elif not main:
728 print_test_result(test)
729 return test
730
731def parse_run_tests(kernel_output: Iterable[str]) -> Test:
732 """
733 Using kernel output, extract KTAP lines, parse the lines for test
734 results and print condensed test results and summary line.
735
736 Parameters:
737 kernel_output - Iterable object contains lines of kernel output
738
739 Return:
740 Test - the main test object with all subtests.
741 """
742 stdout.print_with_timestamp(DIVIDER)
743 lines = extract_tap_lines(kernel_output)
744 test = Test()
745 if not lines:
746 test.name = '<missing>'
747 test.add_error('could not find any KTAP output!')
748 test.status = TestStatus.FAILURE_TO_PARSE_TESTS
749 else:
750 test = parse_test(lines, 0, [])
751 if test.status != TestStatus.NO_TESTS:
752 test.status = test.counts.get_status()
753 stdout.print_with_timestamp(DIVIDER)
754 print_summary_line(test)
755 return test