zephyr/scripts/filter-known-issues.py

152 lines
5.5 KiB
Python
Executable File

#! /usr/bin/python
import argparse
import logging
import mmap
import os
import re
import sys
exclude_regexs = []
noncomment_regex = re.compile(
"(^[ \t][^#]+.*\n)+"
, re.MULTILINE)
def config_import_file(filename):
"""
Imports regular expresions from any file *.conf in the given path
Each file follows the format::
#
# Comments for multiline regex 1...
#
multilineregex
multilineregex
multilineregex
#
# Comments for multiline regex 2...
#
multilineregex
multilineregex
multilineregex
etc.
"""
try:
with open(filename, "rb") as f:
mm = mmap.mmap(f.fileno(), 0, access = mmap.ACCESS_READ)
# That regex basically selects any block of
# lines that is not a comment block. The
# finditer() finds all the blocks and selects
# the bits of mmapped-file that comprises
# each--we compile it into a regex and append.
for m in re.finditer("(^\s*[^#].*\n)+", mm, re.MULTILINE):
origin = "%s:%s-%s" % (filename, m.start(), m.end())
try:
r = re.compile(mm[m.start():m.end()], re.MULTILINE)
except Exception as e:
logging.error("%s: bytes %d-%d: bad regex: %s",
filename, m.start(), m.end(), e)
raise
logging.debug("%s: found regex at bytes %d-%d: %s",
filename, m.start(), m.end(),
mm[m.start():m.end()])
exclude_regexs.append((r, origin))
logging.debug("%s: loaded", filename)
except Exception as e:
raise Exception("E: %s: can't load config file: %s" % (filename, e))
def config_import_path(path):
"""
Imports regular expresions from any file *.conf in the given path
"""
file_regex = re.compile(".*\.conf$")
try:
for dirpath, dirnames, filenames in os.walk(path):
for _filename in sorted(filenames):
filename = os.path.join(dirpath, _filename)
if not file_regex.search(_filename):
logging.debug("%s: ignored", filename)
continue
config_import_file(filename)
except Exception as e:
raise Exception("E: %s: can't load config files: %s" % (path, e))
def config_import(paths):
"""
Imports regular expresions from any file *.conf in the list of paths.
If a path is "" or None, the list of paths until then is flushed
and only the new ones are considered.
"""
_paths = []
# Go over the list, flush it if the user gave an empty path ("")
for path in paths:
if path == "" or path == None:
logging.debug("flushing current config list: %s", _paths)
_paths = []
else:
_paths.append(path)
logging.debug("config list: %s", _paths)
for path in _paths:
config_import_path(path)
arg_parser = argparse.ArgumentParser()
arg_parser.add_argument("-v", "--verbosity", action = "count", default = 0,
help = "increase verbosity")
arg_parser.add_argument("-q", "--quiet", action = "count", default = 0,
help = "decrease verbosity")
arg_parser.add_argument("-c", "--config-dir", action = "append", nargs = "?",
default = [ ".known-issues/" ],
help = "configuration directory (multiple can be "
"given; if none given, clears the current list) "
"%(default)s")
arg_parser.add_argument("FILENAMEs", nargs = "+",
help = "files to filter")
args = arg_parser.parse_args()
logging.basicConfig(level = 40 - 10 * (args.verbosity - args.quiet),
format = "%(levelname)s: %(message)s")
path = ".known-issues/"
logging.debug("Reading configuration from directory `%s`", path)
config_import(args.config_dir)
exclude_ranges = []
for filename in args.FILENAMEs:
try:
with open(filename, "r+b") as f:
logging.info("%s: filtering", filename)
# Yeah, this should be more protected in case of exception
# and such, but this is a short running program...
mm = mmap.mmap(f.fileno(), 0)
for ex, origin in exclude_regexs:
logging.info("%s: searching from %s: %s",
filename, origin, ex.pattern)
for m in re.finditer(ex.pattern, mm, re.MULTILINE):
logging.debug("%s: %s-%s: match from from %s",
filename, m.start(), m.end(), origin)
exclude_ranges.append((m.start(), m.end()))
exclude_ranges = sorted(exclude_ranges, key=lambda r: r[0])
logging.warning("%s: ranges excluded: %s", filename, exclude_ranges)
# Printd what has not been filtered
offset = 0
for b, e in exclude_ranges:
mm.seek(offset)
d = b - offset
logging.debug("%s: exclude range (%d, %d), from %d %dB",
filename, b, e, offset, d)
if b > offset:
print(mm.read(d - 1))
offset = e
mm.seek(offset)
if len(mm) != offset:
print mm.read(len(mm) - offset - 1)
del mm
except Exception as e:
logging.error("%s: cannot load: %s", filename, e)