fish-shell/build_tools/littlecheck.py
Fabian Boehm 0f3e58efe7 Update littlecheck to e2f243e80117aad27adfb8da73f4901c88b3af6d
Now prints a message if process was killed by a signal, truncates
output if no more checks were found, has an option to force color.
2022-08-05 18:29:29 +02:00

795 lines
27 KiB
Python
Executable File

#!/usr/bin/env python
""" Command line test driver. """
from __future__ import unicode_literals
from __future__ import print_function
import argparse
import datetime
import io
import re
import shlex
import subprocess
import sys
try:
from itertools import zip_longest
except ImportError:
from itertools import izip_longest as zip_longest
from difflib import SequenceMatcher
# Directives can occur at the beginning of a line, or anywhere in a line that does not start with #.
COMMENT_RE = r"^(?:[^#].*)?#\s*"
# A regex showing how to run the file.
RUN_RE = re.compile(COMMENT_RE + r"RUN:\s+(.*)\n")
REQUIRES_RE = re.compile(COMMENT_RE + r"REQUIRES:\s+(.*)\n")
# A regex capturing lines that should be checked against stdout.
CHECK_STDOUT_RE = re.compile(COMMENT_RE + r"CHECK:\s+(.*)\n")
# A regex capturing lines that should be checked against stderr.
CHECK_STDERR_RE = re.compile(COMMENT_RE + r"CHECKERR:\s+(.*)\n")
SKIP = object()
def find_command(program):
import os
path, name = os.path.split(program)
if path:
return os.path.isfile(program) and os.access(program, os.X_OK)
for path in os.environ["PATH"].split(os.pathsep):
exe = os.path.join(path, program)
if os.path.isfile(exe) and os.access(exe, os.X_OK):
return exe
return None
class Config(object):
def __init__(self):
# Whether to have verbose output.
self.verbose = False
# Whether output gets ANSI colorization.
self.colorize = False
# Whether to show which file was tested.
self.progress = False
def colors(self):
""" Return a dictionary mapping color names to ANSI escapes """
def ansic(n):
return "\033[%dm" % n if self.colorize else ""
return {
"RESET": ansic(0),
"BOLD": ansic(1),
"NORMAL": ansic(39),
"BLACK": ansic(30),
"RED": ansic(31),
"GREEN": ansic(32),
"YELLOW": ansic(33),
"BLUE": ansic(34),
"MAGENTA": ansic(35),
"CYAN": ansic(36),
"LIGHTGRAY": ansic(37),
"DARKGRAY": ansic(90),
"LIGHTRED": ansic(91),
"LIGHTGREEN": ansic(92),
"LIGHTYELLOW": ansic(93),
"LIGHTBLUE": ansic(94),
"LIGHTMAGENTA": ansic(95),
"LIGHTCYAN": ansic(96),
"WHITE": ansic(97),
}
def output(*args):
print("".join(args) + "\n")
import unicodedata
def esc(m):
map = {
"\n": "\\n",
"\\": "\\\\",
"'": "\\'",
'"': '\\"',
"\a": "\\a",
"\b": "\\b",
"\f": "\\f",
"\r": "\\r",
"\t": "\\t",
"\v": "\\v",
}
if m in map:
return map[m]
if unicodedata.category(m)[0] == "C":
return "\\x{:02x}".format(ord(m))
else:
return m
def escape_string(s):
return "".join(esc(ch) for ch in s)
class CheckerError(Exception):
"""Exception subclass for check line parsing.
Attributes:
line: the Line object on which the exception occurred.
"""
def __init__(self, message, line=None):
super(CheckerError, self).__init__(message)
self.line = line
class Line(object):
""" A line that remembers where it came from. """
def __init__(self, text, number, file):
self.text = text
self.number = number
self.file = file
def __hash__(self):
# Chosen by fair diceroll
# No, just kidding.
# HACK: We pass this to the Sequencematcher, which puts the Checks into a dict.
# To force it to match the regexes, we return a hash collision intentionally,
# so it falls back on __eq__().
#
# CheckCmd has the same thing.
return 0
def __eq__(self, other):
if other is None:
return False
if isinstance(other, CheckCmd):
return other.regex.match(self.text)
if isinstance(other, Line):
# We only compare the text here so SequenceMatcher can reshuffle these
return self.text == other.text
raise NotImplementedError
def subline(self, text):
""" Return a substring of our line with the given text, preserving number and file. """
return Line(text, self.number, self.file)
@staticmethod
def readfile(file, name):
return [Line(text, idx + 1, name) for idx, text in enumerate(file)]
def is_empty_space(self):
return not self.text or self.text.isspace()
def escaped_text(self, for_formatting=False):
ret = escape_string(self.text.rstrip("\n"))
if for_formatting:
ret = ret.replace("{", "{{").replace("}", "}}")
return ret
class RunCmd(object):
"""A command to run on a given Checker.
Attributes:
args: Unexpanded shell command as a string.
"""
def __init__(self, args, line):
self.args = args
self.line = line
@staticmethod
def parse(line):
if not shlex.split(line.text):
raise CheckerError("Invalid RUN command", line)
return RunCmd(line.text, line)
class TestFailure(object):
def __init__(self, line, check, testrun, diff=None, lines=[], checks=[]):
self.line = line
self.check = check
self.testrun = testrun
self.error_annotation_lines = None
self.diff = diff
self.lines = lines
self.checks = checks
self.signal = None
def message(self):
fields = self.testrun.config.colors()
fields["name"] = self.testrun.name
fields["subbed_command"] = self.testrun.subbed_command
if self.line:
fields.update(
{
"output_file": self.line.file,
"output_lineno": self.line.number,
"output_line": self.line.escaped_text(),
}
)
if self.check:
fields.update(
{
"input_file": self.check.line.file,
"input_lineno": self.check.line.number,
"input_line": self.check.line.escaped_text(),
"check_type": self.check.type,
}
)
filemsg = "" if self.testrun.config.progress else " in {name}"
fmtstrs = ["{RED}Failure{RESET}" + filemsg + ":", ""]
if self.signal:
fmtstrs += [
" Process was killed by signal {BOLD}" + self.signal + "{RESET}",
""
]
if self.line and self.check:
fmtstrs += [
" The {check_type} on line {input_lineno} wants:",
" {BOLD}{input_line}{RESET}",
"",
" which failed to match line {output_file}:{output_lineno}:",
" {BOLD}{output_line}{RESET}",
"",
]
elif self.check:
fmtstrs += [
" The {check_type} on line {input_lineno} wants:",
" {BOLD}{input_line}{RESET}",
"",
" but there was no remaining output to match.",
"",
]
else:
fmtstrs += [
" There were no remaining checks left to match {output_file}:{output_lineno}:",
" {BOLD}{output_line}{RESET}",
"",
]
if self.error_annotation_lines:
fields["error_annotation"] = " ".join(
[x.text for x in self.error_annotation_lines]
)
fields["error_annotation_lineno"] = str(
self.error_annotation_lines[0].number
)
if len(self.error_annotation_lines) > 1:
fields["error_annotation_lineno"] += ":" + str(
self.error_annotation_lines[-1].number
)
fmtstrs += [
" additional output on stderr:{error_annotation_lineno}:",
" {BOLD}{error_annotation}{RESET}",
]
if self.diff:
fmtstrs += [" Context:"]
lasthi = 0
lastcheckline = None
for d in self.diff.get_grouped_opcodes():
for op, alo, ahi, blo, bhi in d:
color = "{BOLD}"
if op == "replace" or op == "delete":
color = "{RED}"
# We got a new chunk, so we print a marker.
if alo > lasthi:
fmtstrs += [
" [...] from line "
+ str(self.checks[blo].line.number)
+ " ("
+ self.lines[alo].file
+ ":"
+ str(self.lines[alo].number)
+ "):"
]
lasthi = ahi
# We print one "no more checks" after the last check and then skip any markers
lastcheck = False
for a, b in zip_longest(self.lines[alo:ahi], self.checks[blo:bhi]):
# Clean up strings for use in a format string - double up the curlies.
astr = (
color + a.escaped_text(for_formatting=True) + "{RESET}"
if a
else ""
)
if b:
bstr = (
"'{BLUE}"
+ b.line.escaped_text(for_formatting=True)
+ "{RESET}'"
+ " on line "
+ str(b.line.number)
)
lastcheckline = b.line.number
if op == "equal":
fmtstrs += [" " + astr]
elif b and a:
fmtstrs += [
" "
+ astr
+ " <= does not match "
+ b.type
+ " "
+ bstr
]
elif b:
fmtstrs += [
" "
+ astr
+ " <= nothing to match "
+ b.type
+ " "
+ bstr
]
elif not b:
string = " " + astr
if bhi == len(self.checks):
if not lastcheck:
string += " <= no more checks"
lastcheck = True
elif lastcheckline is not None:
string += (
" <= no check matches this, previous check on line "
+ str(lastcheckline)
)
else:
string += " <= no check matches"
fmtstrs.append(string)
fmtstrs.append("")
fmtstrs += [" when running command:", " {subbed_command}"]
return "\n".join(fmtstrs).format(**fields)
def print_message(self):
""" Print our message to stdout. """
print(self.message())
def perform_substitution(input_str, subs):
"""Perform the substitutions described by subs to str
Return the substituted string.
"""
# Sort our substitutions into a list of tuples (key, value), descending by length.
# It needs to be descending because we need to try longer substitutions first.
subs_ordered = sorted(subs.items(), key=lambda s: len(s[0]), reverse=True)
def subber(m):
# We get the entire sequence of characters.
# Replace just the prefix and return it.
text = m.group(1)
for key, replacement in subs_ordered:
if text.startswith(key):
return replacement + text[len(key) :]
# No substitution found, so we default to running it as-is,
# which will end up running it via $PATH.
return text
return re.sub(r"%(%|[a-zA-Z0-9_-]+)", subber, input_str)
def runproc(cmd):
""" Wrapper around subprocess.Popen to save typing """
PIPE = subprocess.PIPE
proc = subprocess.Popen(
cmd,
stdin=PIPE,
stdout=PIPE,
stderr=PIPE,
shell=True,
close_fds=True, # For Python 2.6 as shipped on RHEL 6
)
return proc
class TestRun(object):
def __init__(self, name, runcmd, checker, subs, config):
self.name = name
self.runcmd = runcmd
self.subbed_command = perform_substitution(runcmd.args, subs)
self.checker = checker
self.subs = subs
self.config = config
def check(self, lines, checks):
# Reverse our lines and checks so we can pop off the end.
lineq = lines[::-1]
checkq = checks[::-1]
usedlines = []
usedchecks = []
mismatches = []
while lineq and checkq:
line = lineq[-1]
check = checkq[-1]
if check == line:
# This line matched this checker, continue on.
usedlines.append(line)
usedchecks.append(check)
lineq.pop()
checkq.pop()
elif line.is_empty_space():
# Skip all whitespace input lines.
lineq.pop()
else:
usedlines.append(line)
usedchecks.append(check)
mismatches.append((line, check))
# Failed to match.
lineq.pop()
checkq.pop()
# Drain empties
while lineq and lineq[-1].is_empty_space():
lineq.pop()
# Store the remaining lines for the diff
for i in lineq[::-1]:
if not i.is_empty_space():
usedlines.append(i)
# Store remaining checks for the diff
for i in checkq[::-1]:
usedchecks.append(i)
# If we have no more output, there's no reason to give
# SCREENFULS of text.
# So we truncate the check list.
if len(usedchecks) > len(usedlines):
usedchecks = usedchecks[:len(usedlines) + 5]
# Do a SequenceMatch! This gives us a diff-like thing.
diff = SequenceMatcher(a=usedlines, b=usedchecks, autojunk=False)
# If there's a mismatch or still lines or checkers, we have a failure.
# Otherwise it's success.
if mismatches:
return TestFailure(
mismatches[0][0],
mismatches[0][1],
self,
diff=diff,
lines=usedlines,
checks=usedchecks,
)
elif lineq:
return TestFailure(
lineq[-1], None, self, diff=diff, lines=usedlines, checks=usedchecks
)
elif checkq:
return TestFailure(
None, checkq[-1], self, diff=diff, lines=usedlines, checks=usedchecks
)
else:
# Success!
return None
def run(self):
""" Run the command. Return a TestFailure, or None. """
def split_by_newlines(s):
"""Decode a string and split it by newlines only,
retaining the newlines.
"""
return [s + "\n" for s in s.decode("utf-8").split("\n")]
if self.config.verbose:
print(self.subbed_command)
proc = runproc(self.subbed_command)
stdout, stderr = proc.communicate()
# HACK: This is quite cheesy: POSIX specifies that sh should return 127 for a missing command.
# It's also possible that it'll be returned in other situations,
# most likely when the last command in a shell script doesn't exist.
# So we check if the command *we execute* exists, and complain then.
status = proc.returncode
cmd = shlex.split(self.subbed_command)[0]
if status == 127 and not find_command(cmd):
raise CheckerError("Command could not be found: " + cmd)
if status == 126 and not find_command(cmd):
raise CheckerError("Command is not executable: " + cmd)
outlines = [
Line(text, idx + 1, "stdout")
for idx, text in enumerate(split_by_newlines(stdout))
]
errlines = [
Line(text, idx + 1, "stderr")
for idx, text in enumerate(split_by_newlines(stderr))
]
outfail = self.check(outlines, self.checker.outchecks)
errfail = self.check(errlines, self.checker.errchecks)
# It's possible that something going wrong on stdout resulted in new
# text being printed on stderr. If we have an outfailure, and either
# non-matching or unmatched stderr text, then annotate the outfail
# with it.
if outfail and errfail and errfail.line:
outfail.error_annotation_lines = errlines[errfail.line.number - 1 :]
# Trim a trailing newline
if outfail.error_annotation_lines[-1].text == "\n":
del outfail.error_annotation_lines[-1]
failure = outfail if outfail else errfail
if failure and status < 0:
# Process was killed by a signal and failed,
# add a message.
import signal
# Unfortunately strsignal only exists in python 3.8+,
# and signal.signals is 3.5+.
if hasattr(signal, "Signals"):
try:
sig = signal.Signals(-status)
failure.signal = sig.name + " (" + signal.strsignal(sig.value) + ")"
except ValueError:
failure.signal = str(-status)
else:
# No easy way to get the full list,
# make up a dict.
signals = {
signal.SIGABRT: "SIGABRT",
signal.SIGBUS: "SIGBUS",
signal.SIGFPE: "SIGFPE",
signal.SIGILL: "SIGILL",
signal.SIGSEGV: "SIGSEGV",
signal.SIGTERM: "SIGTERM",
}
failure.signal = signals.get(-status, str(-status))
return failure
class CheckCmd(object):
def __init__(self, line, checktype, regex):
self.line = line
self.type = checktype
self.regex = regex
def __hash__(self):
# HACK: We pass this to the Sequencematcher, which puts the Checks into a dict.
# To force it to match the regexes, we return a hash collision intentionally,
# so it falls back on __eq__().
#
# Line has the same thing.
return 0
def __eq__(self, other):
# "Magical" comparison with lines and strings.
# Typically I wouldn't use this, but it allows us to check if a line matches any check in a dict or list via
# the `in` operator.
if other is None:
return False
if isinstance(other, CheckCmd):
return self.regex == other.regex
if isinstance(other, Line):
return self.regex.match(other.text)
if isinstance(other, str):
return self.regex.match(other)
raise NotImplementedError
@staticmethod
def parse(line, checktype):
# type: (Line) -> CheckCmd
# Everything inside {{}} is a regular expression.
# Everything outside of it is a literal string.
# Split around {{...}}. Then every odd index will be a regex, and
# evens will be literals.
# Note that if {{...}} appears first we will get an empty string in
# the split array, so the {{...}} matches are always at odd indexes.
bracket_re = re.compile(
r"""
\{\{ # Two open brackets
(.*?) # Nongreedy capture
\}\} # Two close brackets
""",
re.VERBOSE,
)
pieces = bracket_re.split(line.text)
even = True
re_strings = []
for piece in pieces:
if even:
# piece is a literal string.
re_strings.append(re.escape(piece))
else:
# piece is a regex (found inside {{...}}).
# Verify the regex can be compiled.
try:
re.compile(piece)
except re.error:
raise CheckerError("Invalid regular expression: '%s'" % piece, line)
re_strings.append(piece)
even = not even
# Enclose each piece in a non-capturing group.
# This ensures that lower-precedence operators don't trip up catenation.
# For example: {{b|c}}d would result in /b|cd/ which is different.
# Backreferences are assumed to match across the entire string.
re_strings = ["(?:%s)" % s for s in re_strings]
# Anchor at beginning and end (allowing arbitrary whitespace), and maybe
# a terminating newline.
# We need the anchors because Python's match() matches an arbitrary prefix,
# not the entire string.
re_strings = [r"^\s*"] + re_strings + [r"\s*\n?$"]
full_re = re.compile("".join(re_strings))
return CheckCmd(line, checktype, full_re)
class Checker(object):
def __init__(self, name, lines):
self.name = name
# Helper to yield subline containing group1 from all matching lines.
def group1s(regex):
for line in lines:
m = regex.match(line.text)
if m:
yield line.subline(m.group(1))
# Find run commands.
self.runcmds = [RunCmd.parse(sl) for sl in group1s(RUN_RE)]
self.shebang_cmd = None
if not self.runcmds:
# If no RUN command has been given, fall back to the shebang.
if lines[0].text.startswith("#!"):
# Remove the "#!" at the beginning, and the newline at the end.
cmd = lines[0].text[2:-1]
self.shebang_cmd = cmd
self.runcmds = [RunCmd(cmd + " %s", lines[0])]
else:
raise CheckerError("No runlines ('# RUN') found")
self.requirecmds = [RunCmd.parse(sl) for sl in group1s(REQUIRES_RE)]
# Find check cmds.
self.outchecks = [
CheckCmd.parse(sl, "CHECK") for sl in group1s(CHECK_STDOUT_RE)
]
self.errchecks = [
CheckCmd.parse(sl, "CHECKERR") for sl in group1s(CHECK_STDERR_RE)
]
def check_file(input_file, name, subs, config, failure_handler):
""" Check a single file. Return a True on success, False on error. """
success = True
lines = Line.readfile(input_file, name)
checker = Checker(name, lines)
# Run all the REQUIRES lines first,
# if any of them fail it's a SKIP
for reqcmd in checker.requirecmds:
proc = runproc(
perform_substitution(reqcmd.args, subs)
)
proc.communicate()
if proc.returncode > 0:
return SKIP
if checker.shebang_cmd is not None and not find_command(checker.shebang_cmd):
raise CheckerError("Command could not be found: " + checker.shebang_cmd)
# Only then run the RUN lines.
for runcmd in checker.runcmds:
failure = TestRun(name, runcmd, checker, subs, config).run()
if failure:
failure_handler(failure)
success = False
return success
def check_path(path, subs, config, failure_handler):
with io.open(path, encoding="utf-8") as fd:
return check_file(fd, path, subs, config, failure_handler)
def parse_subs(subs):
"""Given a list of input substitutions like 'foo=bar',
return a dictionary like {foo:bar}, or exit if invalid.
"""
result = {}
for sub in subs:
try:
key, val = sub.split("=", 1)
if not key:
print("Invalid substitution %s: empty key" % sub)
sys.exit(1)
if not val:
print("Invalid substitution %s: empty value" % sub)
sys.exit(1)
result[key] = val
except ValueError:
print("Invalid substitution %s: equal sign not found" % sub)
sys.exit(1)
return result
def get_argparse():
""" Return a littlecheck argument parser. """
parser = argparse.ArgumentParser(
description="littlecheck: command line tool tester."
)
parser.add_argument(
"-s",
"--substitute",
type=str,
help="Add a new substitution for RUN lines. Example: bash=/bin/bash",
action="append",
default=[],
)
parser.add_argument(
"-p",
"--progress",
action="store_true",
dest="progress",
help="Show the files to be checked",
default=False,
)
parser.add_argument(
"--force-color",
action="store_true",
dest="force_color",
help="Force usage of color even if not connected to a terminal",
default=False,
)
parser.add_argument("file", nargs="+", help="File to check")
return parser
def main():
args = get_argparse().parse_args()
# Default substitution is %% -> %
def_subs = {"%": "%"}
def_subs.update(parse_subs(args.substitute))
tests_count = 0
failed = False
skip_count = 0
config = Config()
config.colorize = args.force_color or sys.stdout.isatty()
config.progress = args.progress
fields = config.colors()
for path in args.file:
tests_count += 1
fields["path"] = path
if config.progress:
print("Testing file {path} ... ".format(**fields), end="")
sys.stdout.flush()
subs = def_subs.copy()
subs["s"] = path
starttime = datetime.datetime.now()
ret = check_path(path, subs, config, TestFailure.print_message)
if ret is SKIP:
skip_count += 1
if not ret:
failed = True
elif config.progress:
endtime = datetime.datetime.now()
duration_ms = round((endtime - starttime).total_seconds() * 1000)
reason = "ok"
color = "{GREEN}"
if ret is SKIP:
reason = "SKIPPED"
color = "{BLUE}"
print(
(color + "{reason}{RESET} ({duration} ms)").format(
duration=duration_ms, reason=reason, **fields
)
)
# To facilitate integration with testing frameworks, use exit code 125 to indicate that all
# tests have been skipped (primarily for use when tests are run one at a time). Exit code 125 is
# used to indicate to automated `git bisect` runs that a revision has been skipped; we use it
# for the same reasons git does.
if skip_count > 0 and skip_count == tests_count:
sys.exit(125)
sys.exit(1 if failed else 0)
if __name__ == "__main__":
main()