summaryrefslogtreecommitdiffstats
path: root/testing/mochitest/bisection.py
diff options
context:
space:
mode:
Diffstat (limited to 'testing/mochitest/bisection.py')
-rw-r--r--testing/mochitest/bisection.py287
1 files changed, 287 insertions, 0 deletions
diff --git a/testing/mochitest/bisection.py b/testing/mochitest/bisection.py
new file mode 100644
index 0000000000..044be23542
--- /dev/null
+++ b/testing/mochitest/bisection.py
@@ -0,0 +1,287 @@
+import math
+
+import mozinfo
+
+
+class Bisect(object):
+
+ "Class for creating, bisecting and summarizing for --bisect-chunk option."
+
+ def __init__(self, harness):
+ super(Bisect, self).__init__()
+ self.summary = []
+ self.contents = {}
+ self.repeat = 10
+ self.failcount = 0
+ self.max_failures = 3
+
+ def setup(self, tests):
+ """This method is used to initialize various variables that are required
+ for test bisection"""
+ status = 0
+ self.contents.clear()
+ # We need totalTests key in contents for sanity check
+ self.contents["totalTests"] = tests
+ self.contents["tests"] = tests
+ self.contents["loop"] = 0
+ return status
+
+ def reset(self, expectedError, result):
+ """This method is used to initialize self.expectedError and self.result
+ for each loop in runtests."""
+ self.expectedError = expectedError
+ self.result = result
+
+ def get_tests_for_bisection(self, options, tests):
+ """Make a list of tests for bisection from a given list of tests"""
+ bisectlist = []
+ for test in tests:
+ bisectlist.append(test)
+ if test.endswith(options.bisectChunk):
+ break
+
+ return bisectlist
+
+ def pre_test(self, options, tests, status):
+ """This method is used to call other methods for setting up variables and
+ getting the list of tests for bisection."""
+ if options.bisectChunk == "default":
+ return tests
+ # The second condition in 'if' is required to verify that the failing
+ # test is the last one.
+ elif "loop" not in self.contents or not self.contents["tests"][-1].endswith(
+ options.bisectChunk
+ ):
+ tests = self.get_tests_for_bisection(options, tests)
+ status = self.setup(tests)
+
+ return self.next_chunk_binary(options, status)
+
+ def post_test(self, options, expectedError, result):
+ """This method is used to call other methods to summarize results and check whether a
+ sanity check is done or not."""
+ self.reset(expectedError, result)
+ status = self.summarize_chunk(options)
+ # Check whether sanity check has to be done. Also it is necessary to check whether
+ # options.bisectChunk is present in self.expectedError as we do not want to run
+ # if it is "default".
+ if status == -1 and options.bisectChunk in self.expectedError:
+ # In case we have a debug build, we don't want to run a sanity
+ # check, will take too much time.
+ if mozinfo.info["debug"]:
+ return status
+
+ testBleedThrough = self.contents["testsToRun"][0]
+ tests = self.contents["totalTests"]
+ tests.remove(testBleedThrough)
+ # To make sure that the failing test is dependent on some other
+ # test.
+ if options.bisectChunk in testBleedThrough:
+ return status
+
+ status = self.setup(tests)
+ self.summary.append("Sanity Check:")
+
+ return status
+
+ def next_chunk_reverse(self, options, status):
+ "This method is used to bisect the tests in a reverse search fashion."
+
+ # Base Cases.
+ if self.contents["loop"] <= 1:
+ self.contents["testsToRun"] = self.contents["tests"]
+ if self.contents["loop"] == 1:
+ self.contents["testsToRun"] = [self.contents["tests"][-1]]
+ self.contents["loop"] += 1
+ return self.contents["testsToRun"]
+
+ if "result" in self.contents:
+ if self.contents["result"] == "PASS":
+ chunkSize = self.contents["end"] - self.contents["start"]
+ self.contents["end"] = self.contents["start"] - 1
+ self.contents["start"] = self.contents["end"] - chunkSize
+
+ # self.contents['result'] will be expected error only if it fails.
+ elif self.contents["result"] == "FAIL":
+ self.contents["tests"] = self.contents["testsToRun"]
+ status = 1 # for initializing
+
+ # initialize
+ if status:
+ totalTests = len(self.contents["tests"])
+ chunkSize = int(math.ceil(totalTests / 10.0))
+ self.contents["start"] = totalTests - chunkSize - 1
+ self.contents["end"] = totalTests - 2
+
+ start = self.contents["start"]
+ end = self.contents["end"] + 1
+ self.contents["testsToRun"] = self.contents["tests"][start:end]
+ self.contents["testsToRun"].append(self.contents["tests"][-1])
+ self.contents["loop"] += 1
+
+ return self.contents["testsToRun"]
+
+ def next_chunk_binary(self, options, status):
+ "This method is used to bisect the tests in a binary search fashion."
+
+ # Base cases.
+ if self.contents["loop"] <= 1:
+ self.contents["testsToRun"] = self.contents["tests"]
+ if self.contents["loop"] == 1:
+ self.contents["testsToRun"] = [self.contents["tests"][-1]]
+ self.contents["loop"] += 1
+ return self.contents["testsToRun"]
+
+ # Initialize the contents dict.
+ if status:
+ totalTests = len(self.contents["tests"])
+ self.contents["start"] = 0
+ self.contents["end"] = totalTests - 2
+
+ # pylint --py3k W1619
+ mid = (self.contents["start"] + self.contents["end"]) / 2
+ if "result" in self.contents:
+ if self.contents["result"] == "PASS":
+ self.contents["end"] = mid
+
+ elif self.contents["result"] == "FAIL":
+ self.contents["start"] = mid + 1
+
+ mid = (self.contents["start"] + self.contents["end"]) / 2
+ start = mid + 1
+ end = self.contents["end"] + 1
+ self.contents["testsToRun"] = self.contents["tests"][start:end]
+ if not self.contents["testsToRun"]:
+ self.contents["testsToRun"].append(self.contents["tests"][mid])
+ self.contents["testsToRun"].append(self.contents["tests"][-1])
+ self.contents["loop"] += 1
+
+ return self.contents["testsToRun"]
+
+ def summarize_chunk(self, options):
+ "This method is used summarize the results after the list of tests is run."
+ if options.bisectChunk == "default":
+ # if no expectedError that means all the tests have successfully
+ # passed.
+ if len(self.expectedError) == 0:
+ return -1
+ options.bisectChunk = self.expectedError.keys()[0]
+ self.summary.append("\tFound Error in test: %s" % options.bisectChunk)
+ return 0
+
+ # If options.bisectChunk is not in self.result then we need to move to
+ # the next run.
+ if options.bisectChunk not in self.result:
+ return -1
+
+ self.summary.append("\tPass %d:" % self.contents["loop"])
+ if len(self.contents["testsToRun"]) > 1:
+ self.summary.append(
+ "\t\t%d test files(start,end,failing). [%s, %s, %s]"
+ % (
+ len(self.contents["testsToRun"]),
+ self.contents["testsToRun"][0],
+ self.contents["testsToRun"][-2],
+ self.contents["testsToRun"][-1],
+ )
+ )
+ else:
+ self.summary.append("\t\t1 test file [%s]" % self.contents["testsToRun"][0])
+ return self.check_for_intermittent(options)
+
+ if self.result[options.bisectChunk] == "PASS":
+ self.summary.append("\t\tno failures found.")
+ if self.contents["loop"] == 1:
+ status = -1
+ else:
+ self.contents["result"] = "PASS"
+ status = 0
+
+ elif self.result[options.bisectChunk] == "FAIL":
+ if "expectedError" not in self.contents:
+ self.summary.append("\t\t%s failed." % self.contents["testsToRun"][-1])
+ self.contents["expectedError"] = self.expectedError[options.bisectChunk]
+ status = 0
+
+ elif (
+ self.expectedError[options.bisectChunk]
+ == self.contents["expectedError"]
+ ):
+ self.summary.append(
+ "\t\t%s failed with expected error."
+ % self.contents["testsToRun"][-1]
+ )
+ self.contents["result"] = "FAIL"
+ status = 0
+
+ # This code checks for test-bleedthrough. Should work for any
+ # algorithm.
+ numberOfTests = len(self.contents["testsToRun"])
+ if numberOfTests < 3:
+ # This means that only 2 tests are run. Since the last test
+ # is the failing test itself therefore the bleedthrough
+ # test is the first test
+ self.summary.append(
+ "TEST-UNEXPECTED-FAIL | %s | Bleedthrough detected, this test is the "
+ "root cause for many of the above failures"
+ % self.contents["testsToRun"][0]
+ )
+ status = -1
+ else:
+ self.summary.append(
+ "\t\t%s failed with different error."
+ % self.contents["testsToRun"][-1]
+ )
+ status = -1
+
+ return status
+
+ def check_for_intermittent(self, options):
+ "This method is used to check whether a test is an intermittent."
+ if self.result[options.bisectChunk] == "PASS":
+ self.summary.append(
+ "\t\tThe test %s passed." % self.contents["testsToRun"][0]
+ )
+ if self.repeat > 0:
+ # loop is set to 1 to again run the single test.
+ self.contents["loop"] = 1
+ self.repeat -= 1
+ return 0
+ else:
+ if self.failcount > 0:
+ # -1 is being returned as the test is intermittent, so no need to bisect
+ # further.
+ return -1
+ # If the test does not fail even once, then proceed to next chunk for bisection.
+ # loop is set to 2 to proceed on bisection.
+ self.contents["loop"] = 2
+ return 1
+ elif self.result[options.bisectChunk] == "FAIL":
+ self.summary.append(
+ "\t\tThe test %s failed." % self.contents["testsToRun"][0]
+ )
+ self.failcount += 1
+ self.contents["loop"] = 1
+ self.repeat -= 1
+ # self.max_failures is the maximum number of times a test is allowed
+ # to fail to be called an intermittent. If a test fails more than
+ # limit set, it is a perma-fail.
+ if self.failcount < self.max_failures:
+ if self.repeat == 0:
+ # -1 is being returned as the test is intermittent, so no need to bisect
+ # further.
+ return -1
+ return 0
+ else:
+ self.summary.append(
+ "TEST-UNEXPECTED-FAIL | %s | Bleedthrough detected, this test is the "
+ "root cause for many of the above failures"
+ % self.contents["testsToRun"][0]
+ )
+ return -1
+
+ def print_summary(self):
+ "This method is used to print the recorded summary."
+ print("Bisection summary:")
+ for line in self.summary:
+ print(line)