summaryrefslogtreecommitdiffstats
path: root/tools/tryselect/selectors/syntax.py
diff options
context:
space:
mode:
Diffstat (limited to 'tools/tryselect/selectors/syntax.py')
-rw-r--r--tools/tryselect/selectors/syntax.py708
1 files changed, 708 insertions, 0 deletions
diff --git a/tools/tryselect/selectors/syntax.py b/tools/tryselect/selectors/syntax.py
new file mode 100644
index 0000000000..29b80f519a
--- /dev/null
+++ b/tools/tryselect/selectors/syntax.py
@@ -0,0 +1,708 @@
+# This Source Code Form is subject to the terms of the Mozilla Public
+# License, v. 2.0. If a copy of the MPL was not distributed with this
+# file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+
+import os
+import re
+import sys
+from collections import defaultdict
+
+import mozpack.path as mozpath
+from moztest.resolve import TestResolver
+
+from ..cli import BaseTryParser
+from ..push import build, push_to_try
+
+here = os.path.abspath(os.path.dirname(__file__))
+
+
+class SyntaxParser(BaseTryParser):
+ name = "syntax"
+ arguments = [
+ [
+ ["paths"],
+ {
+ "nargs": "*",
+ "default": [],
+ "help": "Paths to search for tests to run on try.",
+ },
+ ],
+ [
+ ["-b", "--build"],
+ {
+ "dest": "builds",
+ "default": "do",
+ "help": "Build types to run (d for debug, o for optimized).",
+ },
+ ],
+ [
+ ["-p", "--platform"],
+ {
+ "dest": "platforms",
+ "action": "append",
+ "help": "Platforms to run (required if not found in the environment as "
+ "AUTOTRY_PLATFORM_HINT).",
+ },
+ ],
+ [
+ ["-u", "--unittests"],
+ {
+ "dest": "tests",
+ "action": "append",
+ "help": "Test suites to run in their entirety.",
+ },
+ ],
+ [
+ ["-t", "--talos"],
+ {
+ "action": "append",
+ "help": "Talos suites to run.",
+ },
+ ],
+ [
+ ["-j", "--jobs"],
+ {
+ "action": "append",
+ "help": "Job tasks to run.",
+ },
+ ],
+ [
+ ["--tag"],
+ {
+ "dest": "tags",
+ "action": "append",
+ "help": "Restrict tests to the given tag (may be specified multiple times).",
+ },
+ ],
+ [
+ ["--and"],
+ {
+ "action": "store_true",
+ "dest": "intersection",
+ "help": "When -u and paths are supplied run only the intersection of the "
+ "tests specified by the two arguments.",
+ },
+ ],
+ [
+ ["--no-artifact"],
+ {
+ "action": "store_true",
+ "help": "Disable artifact builds even if --enable-artifact-builds is set "
+ "in the mozconfig.",
+ },
+ ],
+ [
+ ["-v", "--verbose"],
+ {
+ "dest": "verbose",
+ "action": "store_true",
+ "default": False,
+ "help": "Print detailed information about the resulting test selection "
+ "and commands performed.",
+ },
+ ],
+ ]
+
+ # Arguments we will accept on the command line and pass through to try
+ # syntax with no further intervention. The set is taken from
+ # http://trychooser.pub.build.mozilla.org with a few additions.
+ #
+ # Note that the meaning of store_false and store_true arguments is
+ # not preserved here, as we're only using these to echo the literal
+ # arguments to another consumer. Specifying either store_false or
+ # store_true here will have an equivalent effect.
+ pass_through_arguments = {
+ "--rebuild": {
+ "action": "store",
+ "dest": "rebuild",
+ "help": "Re-trigger all test jobs (up to 20 times)",
+ },
+ "--rebuild-talos": {
+ "action": "store",
+ "dest": "rebuild_talos",
+ "help": "Re-trigger all talos jobs",
+ },
+ "--interactive": {
+ "action": "store_true",
+ "dest": "interactive",
+ "help": "Allow ssh-like access to running test containers",
+ },
+ "--no-retry": {
+ "action": "store_true",
+ "dest": "no_retry",
+ "help": "Do not retrigger failed tests",
+ },
+ "--setenv": {
+ "action": "append",
+ "dest": "setenv",
+ "help": "Set the corresponding variable in the test environment for "
+ "applicable harnesses.",
+ },
+ "-f": {
+ "action": "store_true",
+ "dest": "failure_emails",
+ "help": "Request failure emails only",
+ },
+ "--failure-emails": {
+ "action": "store_true",
+ "dest": "failure_emails",
+ "help": "Request failure emails only",
+ },
+ "-e": {
+ "action": "store_true",
+ "dest": "all_emails",
+ "help": "Request all emails",
+ },
+ "--all-emails": {
+ "action": "store_true",
+ "dest": "all_emails",
+ "help": "Request all emails",
+ },
+ "--artifact": {
+ "action": "store_true",
+ "dest": "artifact",
+ "help": "Force artifact builds where possible.",
+ },
+ "--upload-xdbs": {
+ "action": "store_true",
+ "dest": "upload_xdbs",
+ "help": "Upload XDB compilation db files generated by hazard build",
+ },
+ }
+ task_configs = []
+
+ def __init__(self, *args, **kwargs):
+ BaseTryParser.__init__(self, *args, **kwargs)
+
+ group = self.add_argument_group("pass-through arguments")
+ for arg, opts in self.pass_through_arguments.items():
+ group.add_argument(arg, **opts)
+
+
+class TryArgumentTokenizer:
+ symbols = [
+ ("separator", ","),
+ ("list_start", r"\["),
+ ("list_end", r"\]"),
+ ("item", r"([^,\[\]\s][^,\[\]]+)"),
+ ("space", r"\s+"),
+ ]
+ token_re = re.compile("|".join("(?P<%s>%s)" % item for item in symbols))
+
+ def tokenize(self, data):
+ for match in self.token_re.finditer(data):
+ symbol = match.lastgroup
+ data = match.group(symbol)
+ if symbol == "space":
+ pass
+ else:
+ yield symbol, data
+
+
+class TryArgumentParser:
+ """Simple three-state parser for handling expressions
+ of the from "foo[sub item, another], bar,baz". This takes
+ input from the TryArgumentTokenizer and runs through a small
+ state machine, returning a dictionary of {top-level-item:[sub_items]}
+ i.e. the above would result in
+ {"foo":["sub item", "another"], "bar": [], "baz": []}
+ In the case of invalid input a ValueError is raised."""
+
+ EOF = object()
+
+ def __init__(self):
+ self.reset()
+
+ def reset(self):
+ self.tokens = None
+ self.current_item = None
+ self.data = {}
+ self.token = None
+ self.state = None
+
+ def parse(self, tokens):
+ self.reset()
+ self.tokens = tokens
+ self.consume()
+ self.state = self.item_state
+ while self.token[0] != self.EOF:
+ self.state()
+ return self.data
+
+ def consume(self):
+ try:
+ self.token = next(self.tokens)
+ except StopIteration:
+ self.token = (self.EOF, None)
+
+ def expect(self, *types):
+ if self.token[0] not in types:
+ raise ValueError(
+ "Error parsing try string, unexpected %s" % (self.token[0])
+ )
+
+ def item_state(self):
+ self.expect("item")
+ value = self.token[1].strip()
+ if value not in self.data:
+ self.data[value] = []
+ self.current_item = value
+ self.consume()
+ if self.token[0] == "separator":
+ self.consume()
+ elif self.token[0] == "list_start":
+ self.consume()
+ self.state = self.subitem_state
+ elif self.token[0] == self.EOF:
+ pass
+ else:
+ raise ValueError
+
+ def subitem_state(self):
+ self.expect("item")
+ value = self.token[1].strip()
+ self.data[self.current_item].append(value)
+ self.consume()
+ if self.token[0] == "separator":
+ self.consume()
+ elif self.token[0] == "list_end":
+ self.consume()
+ self.state = self.after_list_end_state
+ else:
+ raise ValueError
+
+ def after_list_end_state(self):
+ self.expect("separator")
+ self.consume()
+ self.state = self.item_state
+
+
+def parse_arg(arg):
+ tokenizer = TryArgumentTokenizer()
+ parser = TryArgumentParser()
+ return parser.parse(tokenizer.tokenize(arg))
+
+
+class AutoTry:
+ # Maps from flavors to the job names needed to run that flavour
+ flavor_jobs = {
+ "mochitest": ["mochitest-1", "mochitest-e10s-1"],
+ "xpcshell": ["xpcshell"],
+ "chrome": ["mochitest-o"],
+ "browser-a11y": ["mochitest-ba"],
+ "browser-media": ["mochitest-bmda"],
+ "browser-chrome": [
+ "mochitest-browser-chrome-1",
+ "mochitest-e10s-browser-chrome-1",
+ "mochitest-browser-chrome-e10s-1",
+ ],
+ "devtools-chrome": [
+ "mochitest-devtools-chrome-1",
+ "mochitest-e10s-devtools-chrome-1",
+ "mochitest-devtools-chrome-e10s-1",
+ ],
+ "crashtest": ["crashtest", "crashtest-e10s"],
+ "reftest": ["reftest", "reftest-e10s"],
+ "remote": ["mochitest-remote"],
+ "web-platform-tests": ["web-platform-tests-1"],
+ }
+
+ flavor_suites = {
+ "mochitest": "mochitests",
+ "xpcshell": "xpcshell",
+ "chrome": "mochitest-o",
+ "browser-chrome": "mochitest-bc",
+ "browser-a11y": "mochitest-ba",
+ "browser-media": "mochitest-bmda",
+ "devtools-chrome": "mochitest-dt",
+ "crashtest": "crashtest",
+ "reftest": "reftest",
+ "web-platform-tests": "web-platform-tests",
+ }
+
+ compiled_suites = [
+ "cppunit",
+ "gtest",
+ "jittest",
+ ]
+
+ common_suites = [
+ "cppunit",
+ "crashtest",
+ "firefox-ui-functional",
+ "geckoview",
+ "geckoview-junit",
+ "gtest",
+ "jittest",
+ "jsreftest",
+ "marionette",
+ "marionette-e10s",
+ "mochitests",
+ "reftest",
+ "robocop",
+ "web-platform-tests",
+ "xpcshell",
+ ]
+
+ def __init__(self):
+ self.topsrcdir = build.topsrcdir
+ self._resolver = None
+
+ @property
+ def resolver(self):
+ if self._resolver is None:
+ self._resolver = TestResolver.from_environment(cwd=here)
+ return self._resolver
+
+ @classmethod
+ def split_try_string(cls, data):
+ return re.findall(r"(?:\[.*?\]|\S)+", data)
+
+ def paths_by_flavor(self, paths=None, tags=None):
+ paths_by_flavor = defaultdict(set)
+
+ if not (paths or tags):
+ return dict(paths_by_flavor)
+
+ tests = list(self.resolver.resolve_tests(paths=paths, tags=tags))
+
+ for t in tests:
+ if t["flavor"] in self.flavor_suites:
+ flavor = t["flavor"]
+ if "subsuite" in t and t["subsuite"] == "devtools":
+ flavor = "devtools-chrome"
+
+ if "subsuite" in t and t["subsuite"] == "a11y":
+ flavor = "browser-a11y"
+
+ if "subsuite" in t and t["subsuite"] == "media-bc":
+ flavor = "browser-media"
+
+ if flavor in ["crashtest", "reftest"]:
+ manifest_relpath = os.path.relpath(t["manifest"], self.topsrcdir)
+ paths_by_flavor[flavor].add(os.path.dirname(manifest_relpath))
+ elif "dir_relpath" in t:
+ paths_by_flavor[flavor].add(t["dir_relpath"])
+ else:
+ file_relpath = os.path.relpath(t["path"], self.topsrcdir)
+ dir_relpath = os.path.dirname(file_relpath)
+ paths_by_flavor[flavor].add(dir_relpath)
+
+ for flavor, path_set in paths_by_flavor.items():
+ paths_by_flavor[flavor] = self.deduplicate_prefixes(path_set, paths)
+
+ return dict(paths_by_flavor)
+
+ def deduplicate_prefixes(self, path_set, input_paths):
+ # Removes paths redundant to test selection in the given path set.
+ # If a path was passed on the commandline that is the prefix of a
+ # path in our set, we only need to include the specified prefix to
+ # run the intended tests (every test in "layout/base" will run if
+ # "layout" is passed to the reftest harness).
+ removals = set()
+ additions = set()
+
+ for path in path_set:
+ full_path = path
+ while path:
+ path, _ = os.path.split(path)
+ if path in input_paths:
+ removals.add(full_path)
+ additions.add(path)
+
+ return additions | (path_set - removals)
+
+ def remove_duplicates(self, paths_by_flavor, tests):
+ rv = {}
+ for item in paths_by_flavor:
+ if self.flavor_suites[item] not in tests:
+ rv[item] = paths_by_flavor[item].copy()
+ return rv
+
+ def calc_try_syntax(
+ self,
+ platforms,
+ tests,
+ talos,
+ jobs,
+ builds,
+ paths_by_flavor,
+ tags,
+ extras,
+ intersection,
+ ):
+ parts = ["try:"]
+
+ if platforms:
+ parts.extend(["-b", builds, "-p", ",".join(platforms)])
+
+ suites = tests if not intersection else {}
+ paths = set()
+ for flavor, flavor_tests in paths_by_flavor.items():
+ suite = self.flavor_suites[flavor]
+ if suite not in suites and (not intersection or suite in tests):
+ for job_name in self.flavor_jobs[flavor]:
+ for test in flavor_tests:
+ paths.add("{}:{}".format(flavor, test))
+ suites[job_name] = tests.get(suite, [])
+
+ # intersection implies tests are expected
+ if intersection and not suites:
+ raise ValueError("No tests found matching filters")
+
+ if extras.get("artifact") and any([p.endswith("-nightly") for p in platforms]):
+ print(
+ 'You asked for |--artifact| but "-nightly" platforms don\'t have artifacts. '
+ "Running without |--artifact| instead."
+ )
+ del extras["artifact"]
+
+ if extras.get("artifact"):
+ rejected = []
+ for suite in suites.keys():
+ if any([suite.startswith(c) for c in self.compiled_suites]):
+ rejected.append(suite)
+ if rejected:
+ raise ValueError(
+ "You can't run {} with "
+ "--artifact option.".format(", ".join(rejected))
+ )
+
+ if extras.get("artifact") and "all" in suites.keys():
+ non_compiled_suites = set(self.common_suites) - set(self.compiled_suites)
+ message = (
+ "You asked for |-u all| with |--artifact| but compiled-code tests ({tests})"
+ " can't run against an artifact build. Running (-u {non_compiled_suites}) "
+ "instead."
+ )
+ string_format = {
+ "tests": ",".join(self.compiled_suites),
+ "non_compiled_suites": ",".join(non_compiled_suites),
+ }
+ print(message.format(**string_format))
+ del suites["all"]
+ suites.update({suite_name: None for suite_name in non_compiled_suites})
+
+ if suites:
+ parts.append("-u")
+ parts.append(
+ ",".join(
+ "{}{}".format(k, "[%s]" % ",".join(v) if v else "")
+ for k, v in sorted(suites.items())
+ )
+ )
+
+ if talos:
+ parts.append("-t")
+ parts.append(
+ ",".join(
+ "{}{}".format(k, "[%s]" % ",".join(v) if v else "")
+ for k, v in sorted(talos.items())
+ )
+ )
+
+ if jobs:
+ parts.append("-j")
+ parts.append(",".join(jobs))
+
+ if tags:
+ parts.append(" ".join("--tag %s" % t for t in tags))
+
+ if paths:
+ parts.append("--try-test-paths %s" % " ".join(sorted(paths)))
+
+ args_by_dest = {
+ v["dest"]: k for k, v in SyntaxParser.pass_through_arguments.items()
+ }
+ for dest, value in extras.items():
+ assert dest in args_by_dest
+ arg = args_by_dest[dest]
+ action = SyntaxParser.pass_through_arguments[arg]["action"]
+ if action == "store":
+ parts.append(arg)
+ parts.append(value)
+ if action == "append":
+ for e in value:
+ parts.append(arg)
+ parts.append(e)
+ if action in ("store_true", "store_false"):
+ parts.append(arg)
+
+ return " ".join(parts)
+
+ def normalise_list(self, items, allow_subitems=False):
+ rv = defaultdict(list)
+ for item in items:
+ parsed = parse_arg(item)
+ for key, values in parsed.items():
+ rv[key].extend(values)
+
+ if not allow_subitems:
+ if not all(item == [] for item in rv.values()):
+ raise ValueError("Unexpected subitems in argument")
+ return rv.keys()
+ else:
+ return rv
+
+ def validate_args(self, **kwargs):
+ tests_selected = kwargs["tests"] or kwargs["paths"] or kwargs["tags"]
+ if kwargs["platforms"] is None and (kwargs["jobs"] is None or tests_selected):
+ if "AUTOTRY_PLATFORM_HINT" in os.environ:
+ kwargs["platforms"] = [os.environ["AUTOTRY_PLATFORM_HINT"]]
+ elif tests_selected:
+ print("Must specify platform when selecting tests.")
+ sys.exit(1)
+ else:
+ print(
+ "Either platforms or jobs must be specified as an argument to autotry."
+ )
+ sys.exit(1)
+
+ try:
+ platforms = (
+ self.normalise_list(kwargs["platforms"]) if kwargs["platforms"] else {}
+ )
+ except ValueError as e:
+ print("Error parsing -p argument:\n%s" % e)
+ sys.exit(1)
+
+ try:
+ tests = (
+ self.normalise_list(kwargs["tests"], allow_subitems=True)
+ if kwargs["tests"]
+ else {}
+ )
+ except ValueError as e:
+ print("Error parsing -u argument ({}):\n{}".format(kwargs["tests"], e))
+ sys.exit(1)
+
+ try:
+ talos = (
+ self.normalise_list(kwargs["talos"], allow_subitems=True)
+ if kwargs["talos"]
+ else []
+ )
+ except ValueError as e:
+ print("Error parsing -t argument:\n%s" % e)
+ sys.exit(1)
+
+ try:
+ jobs = self.normalise_list(kwargs["jobs"]) if kwargs["jobs"] else {}
+ except ValueError as e:
+ print("Error parsing -j argument:\n%s" % e)
+ sys.exit(1)
+
+ paths = []
+ for p in kwargs["paths"]:
+ p = mozpath.normpath(os.path.abspath(p))
+ if not (os.path.isdir(p) and p.startswith(self.topsrcdir)):
+ print(
+ 'Specified path "%s" is not a directory under the srcdir,'
+ " unable to specify tests outside of the srcdir" % p
+ )
+ sys.exit(1)
+ if len(p) <= len(self.topsrcdir):
+ print(
+ 'Specified path "%s" is at the top of the srcdir and would'
+ " select all tests." % p
+ )
+ sys.exit(1)
+ paths.append(os.path.relpath(p, self.topsrcdir))
+
+ try:
+ tags = self.normalise_list(kwargs["tags"]) if kwargs["tags"] else []
+ except ValueError as e:
+ print("Error parsing --tags argument:\n%s" % e)
+ sys.exit(1)
+
+ extra_values = {k["dest"] for k in SyntaxParser.pass_through_arguments.values()}
+ extra_args = {k: v for k, v in kwargs.items() if k in extra_values and v}
+
+ return kwargs["builds"], platforms, tests, talos, jobs, paths, tags, extra_args
+
+ def run(self, **kwargs):
+ if not any(kwargs[item] for item in ("paths", "tests", "tags")):
+ kwargs["paths"] = set()
+ kwargs["tags"] = set()
+
+ builds, platforms, tests, talos, jobs, paths, tags, extra = self.validate_args(
+ **kwargs
+ )
+
+ if paths or tags:
+ paths = [
+ os.path.relpath(os.path.normpath(os.path.abspath(item)), self.topsrcdir)
+ for item in paths
+ ]
+ paths_by_flavor = self.paths_by_flavor(paths=paths, tags=tags)
+
+ if not paths_by_flavor and not tests:
+ print(
+ "No tests were found when attempting to resolve paths:\n\n\t%s"
+ % paths
+ )
+ sys.exit(1)
+
+ if not kwargs["intersection"]:
+ paths_by_flavor = self.remove_duplicates(paths_by_flavor, tests)
+ else:
+ paths_by_flavor = {}
+
+ # No point in dealing with artifacts if we aren't running any builds
+ local_artifact_build = False
+ if platforms:
+ local_artifact_build = kwargs.get("local_artifact_build", False)
+
+ # Add --artifact if --enable-artifact-builds is set ...
+ if local_artifact_build:
+ extra["artifact"] = True
+ # ... unless --no-artifact is explicitly given.
+ if kwargs["no_artifact"]:
+ if "artifact" in extra:
+ del extra["artifact"]
+
+ try:
+ msg = self.calc_try_syntax(
+ platforms,
+ tests,
+ talos,
+ jobs,
+ builds,
+ paths_by_flavor,
+ tags,
+ extra,
+ kwargs["intersection"],
+ )
+ except ValueError as e:
+ print(e)
+ sys.exit(1)
+
+ if local_artifact_build and not kwargs["no_artifact"]:
+ print(
+ "mozconfig has --enable-artifact-builds; including "
+ "--artifact flag in try syntax (use --no-artifact "
+ "to override)"
+ )
+
+ if kwargs["verbose"] and paths_by_flavor:
+ print("The following tests will be selected: ")
+ for flavor, paths in paths_by_flavor.items():
+ print("{}: {}".format(flavor, ",".join(paths)))
+
+ if kwargs["verbose"]:
+ print("The following try syntax was calculated:\n%s" % msg)
+
+ push_to_try(
+ "syntax",
+ kwargs["message"].format(msg=msg),
+ stage_changes=kwargs["stage_changes"],
+ dry_run=kwargs["dry_run"],
+ closed_tree=kwargs["closed_tree"],
+ push_to_lando=kwargs["push_to_lando"],
+ )
+
+
+def run(**kwargs):
+ at = AutoTry()
+ return at.run(**kwargs)