cpython/Tools/c-analyzer/c_analyzer/__main__.py

539 lines
15 KiB
Python

import io
import logging
import os
import os.path
import re
import sys
from c_common import fsutil
from c_common.logging import VERBOSITY, Printer
from c_common.scriptutil import (
add_verbosity_cli,
add_traceback_cli,
add_sepval_cli,
add_progress_cli,
add_files_cli,
add_commands_cli,
process_args_by_key,
configure_logger,
get_prog,
filter_filenames,
)
from c_parser.info import KIND
from .match import filter_forward
from . import (
analyze as _analyze,
datafiles as _datafiles,
check_all as _check_all,
)
KINDS = [
KIND.TYPEDEF,
KIND.STRUCT,
KIND.UNION,
KIND.ENUM,
KIND.FUNCTION,
KIND.VARIABLE,
KIND.STATEMENT,
]
logger = logging.getLogger(__name__)
#######################################
# table helpers
TABLE_SECTIONS = {
'types': (
['kind', 'name', 'data', 'file'],
KIND.is_type_decl,
(lambda v: (v.kind.value, v.filename or '', v.name)),
),
'typedefs': 'types',
'structs': 'types',
'unions': 'types',
'enums': 'types',
'functions': (
['name', 'data', 'file'],
(lambda kind: kind is KIND.FUNCTION),
(lambda v: (v.filename or '', v.name)),
),
'variables': (
['name', 'parent', 'data', 'file'],
(lambda kind: kind is KIND.VARIABLE),
(lambda v: (v.filename or '', str(v.parent) if v.parent else '', v.name)),
),
'statements': (
['file', 'parent', 'data'],
(lambda kind: kind is KIND.STATEMENT),
(lambda v: (v.filename or '', str(v.parent) if v.parent else '', v.name)),
),
KIND.TYPEDEF: 'typedefs',
KIND.STRUCT: 'structs',
KIND.UNION: 'unions',
KIND.ENUM: 'enums',
KIND.FUNCTION: 'functions',
KIND.VARIABLE: 'variables',
KIND.STATEMENT: 'statements',
}
def _render_table(items, columns, relroot=None):
# XXX improve this
header = '\t'.join(columns)
div = '--------------------'
yield header
yield div
total = 0
for item in items:
rowdata = item.render_rowdata(columns)
row = [rowdata[c] for c in columns]
if relroot and 'file' in columns:
index = columns.index('file')
row[index] = os.path.relpath(row[index], relroot)
yield '\t'.join(row)
total += 1
yield div
yield f'total: {total}'
def build_section(name, groupitems, *, relroot=None):
info = TABLE_SECTIONS[name]
while type(info) is not tuple:
if name in KINDS:
name = info
info = TABLE_SECTIONS[info]
columns, match_kind, sortkey = info
items = (v for v in groupitems if match_kind(v.kind))
items = sorted(items, key=sortkey)
def render():
yield ''
yield f'{name}:'
yield ''
for line in _render_table(items, columns, relroot):
yield line
return items, render
#######################################
# the checks
CHECKS = {
#'globals': _check_globals,
}
def add_checks_cli(parser, checks=None, *, add_flags=None):
default = False
if not checks:
checks = list(CHECKS)
default = True
elif isinstance(checks, str):
checks = [checks]
if (add_flags is None and len(checks) > 1) or default:
add_flags = True
process_checks = add_sepval_cli(parser, '--check', 'checks', checks)
if add_flags:
for check in checks:
parser.add_argument(f'--{check}', dest='checks',
action='append_const', const=check)
return [
process_checks,
]
def _get_check_handlers(fmt, printer, verbosity=VERBOSITY):
div = None
def handle_after():
pass
if not fmt:
div = ''
def handle_failure(failure, data):
data = repr(data)
if verbosity >= 3:
logger.info(f'failure: {failure}')
logger.info(f'data: {data}')
else:
logger.warn(f'failure: {failure} (data: {data})')
elif fmt == 'raw':
def handle_failure(failure, data):
print(f'{failure!r} {data!r}')
elif fmt == 'brief':
def handle_failure(failure, data):
parent = data.parent or ''
funcname = parent if isinstance(parent, str) else parent.name
name = f'({funcname}).{data.name}' if funcname else data.name
failure = failure.split('\t')[0]
print(f'{data.filename}:{name} - {failure}')
elif fmt == 'summary':
def handle_failure(failure, data):
print(_fmt_one_summary(data, failure))
elif fmt == 'full':
div = ''
def handle_failure(failure, data):
name = data.shortkey if data.kind is KIND.VARIABLE else data.name
parent = data.parent or ''
funcname = parent if isinstance(parent, str) else parent.name
known = 'yes' if data.is_known else '*** NO ***'
print(f'{data.kind.value} {name!r} failed ({failure})')
print(f' file: {data.filename}')
print(f' func: {funcname or "-"}')
print(f' name: {data.name}')
print(f' data: ...')
print(f' type unknown: {known}')
else:
if fmt in FORMATS:
raise NotImplementedError(fmt)
raise ValueError(f'unsupported fmt {fmt!r}')
return handle_failure, handle_after, div
#######################################
# the formats
def fmt_raw(analysis):
for item in analysis:
yield from item.render('raw')
def fmt_brief(analysis):
# XXX Support sorting.
items = sorted(analysis)
for kind in KINDS:
if kind is KIND.STATEMENT:
continue
for item in items:
if item.kind is not kind:
continue
yield from item.render('brief')
yield f' total: {len(items)}'
def fmt_summary(analysis):
# XXX Support sorting and grouping.
items = list(analysis)
total = len(items)
def section(name):
_, render = build_section(name, items)
yield from render()
yield from section('types')
yield from section('functions')
yield from section('variables')
yield from section('statements')
yield ''
# yield f'grand total: {len(supported) + len(unsupported)}'
yield f'grand total: {total}'
def _fmt_one_summary(item, extra=None):
parent = item.parent or ''
funcname = parent if isinstance(parent, str) else parent.name
if extra:
return f'{item.filename:35}\t{funcname or "-":35}\t{item.name:40}\t{extra}'
else:
return f'{item.filename:35}\t{funcname or "-":35}\t{item.name}'
def fmt_full(analysis):
# XXX Support sorting.
items = sorted(analysis, key=lambda v: v.key)
yield ''
for item in items:
yield from item.render('full')
yield ''
yield f'total: {len(items)}'
FORMATS = {
'raw': fmt_raw,
'brief': fmt_brief,
'summary': fmt_summary,
'full': fmt_full,
}
def add_output_cli(parser, *, default='summary'):
parser.add_argument('--format', dest='fmt', default=default, choices=tuple(FORMATS))
def process_args(args, *, argv=None):
pass
return process_args
#######################################
# the commands
def _cli_check(parser, checks=None, **kwargs):
if isinstance(checks, str):
checks = [checks]
if checks is False:
process_checks = None
elif checks is None:
process_checks = add_checks_cli(parser)
elif len(checks) == 1 and type(checks) is not dict and re.match(r'^<.*>$', checks[0]):
check = checks[0][1:-1]
def process_checks(args, *, argv=None):
args.checks = [check]
else:
process_checks = add_checks_cli(parser, checks=checks)
process_progress = add_progress_cli(parser)
process_output = add_output_cli(parser, default=None)
process_files = add_files_cli(parser, **kwargs)
return [
process_checks,
process_progress,
process_output,
process_files,
]
def cmd_check(filenames, *,
checks=None,
ignored=None,
fmt=None,
failfast=False,
iter_filenames=None,
relroot=fsutil.USE_CWD,
track_progress=None,
verbosity=VERBOSITY,
_analyze=_analyze,
_CHECKS=CHECKS,
**kwargs
):
if not checks:
checks = _CHECKS
elif isinstance(checks, str):
checks = [checks]
checks = [_CHECKS[c] if isinstance(c, str) else c
for c in checks]
printer = Printer(verbosity)
(handle_failure, handle_after, div
) = _get_check_handlers(fmt, printer, verbosity)
filenames, relroot = fsutil.fix_filenames(filenames, relroot=relroot)
filenames = filter_filenames(filenames, iter_filenames, relroot)
if track_progress:
filenames = track_progress(filenames)
logger.info('analyzing files...')
analyzed = _analyze(filenames, **kwargs)
analyzed.fix_filenames(relroot, normalize=False)
decls = filter_forward(analyzed, markpublic=True)
logger.info('checking analysis results...')
failed = []
for data, failure in _check_all(decls, checks, failfast=failfast):
if data is None:
printer.info('stopping after one failure')
break
if div is not None and len(failed) > 0:
printer.info(div)
failed.append(data)
handle_failure(failure, data)
handle_after()
printer.info('-------------------------')
logger.info(f'total failures: {len(failed)}')
logger.info('done checking')
if fmt == 'summary':
print('Categorized by storage:')
print()
from .match import group_by_storage
grouped = group_by_storage(failed, ignore_non_match=False)
for group, decls in grouped.items():
print()
print(group)
for decl in decls:
print(' ', _fmt_one_summary(decl))
print(f'subtotal: {len(decls)}')
if len(failed) > 0:
sys.exit(len(failed))
def _cli_analyze(parser, **kwargs):
process_progress = add_progress_cli(parser)
process_output = add_output_cli(parser)
process_files = add_files_cli(parser, **kwargs)
return [
process_progress,
process_output,
process_files,
]
# XXX Support filtering by kind.
def cmd_analyze(filenames, *,
fmt=None,
iter_filenames=None,
relroot=fsutil.USE_CWD,
track_progress=None,
verbosity=None,
_analyze=_analyze,
formats=FORMATS,
**kwargs
):
verbosity = verbosity if verbosity is not None else 3
try:
do_fmt = formats[fmt]
except KeyError:
raise ValueError(f'unsupported fmt {fmt!r}')
filenames, relroot = fsutil.fix_filenames(filenames, relroot=relroot)
filenames = filter_filenames(filenames, iter_filenames, relroot)
if track_progress:
filenames = track_progress(filenames)
logger.info('analyzing files...')
analyzed = _analyze(filenames, **kwargs)
analyzed.fix_filenames(relroot, normalize=False)
decls = filter_forward(analyzed, markpublic=True)
for line in do_fmt(decls):
print(line)
def _cli_data(parser, filenames=None, known=None):
ArgumentParser = type(parser)
common = ArgumentParser(add_help=False)
# These flags will get processed by the top-level parse_args().
add_verbosity_cli(common)
add_traceback_cli(common)
subs = parser.add_subparsers(dest='datacmd')
sub = subs.add_parser('show', parents=[common])
if known is None:
sub.add_argument('--known', required=True)
if filenames is None:
sub.add_argument('filenames', metavar='FILE', nargs='+')
sub = subs.add_parser('dump', parents=[common])
if known is None:
sub.add_argument('--known')
sub.add_argument('--show', action='store_true')
process_progress = add_progress_cli(sub)
sub = subs.add_parser('check', parents=[common])
if known is None:
sub.add_argument('--known', required=True)
def process_args(args, *, argv):
if args.datacmd == 'dump':
process_progress(args, argv)
return process_args
def cmd_data(datacmd, filenames, known=None, *,
_analyze=_analyze,
formats=FORMATS,
extracolumns=None,
relroot=fsutil.USE_CWD,
track_progress=None,
**kwargs
):
kwargs.pop('verbosity', None)
usestdout = kwargs.pop('show', None)
if datacmd == 'show':
do_fmt = formats['summary']
if isinstance(known, str):
known, _ = _datafiles.get_known(known, extracolumns, relroot)
for line in do_fmt(known):
print(line)
elif datacmd == 'dump':
filenames, relroot = fsutil.fix_filenames(filenames, relroot=relroot)
if track_progress:
filenames = track_progress(filenames)
analyzed = _analyze(filenames, **kwargs)
analyzed.fix_filenames(relroot, normalize=False)
if known is None or usestdout:
outfile = io.StringIO()
_datafiles.write_known(analyzed, outfile, extracolumns,
relroot=relroot)
print(outfile.getvalue())
else:
_datafiles.write_known(analyzed, known, extracolumns,
relroot=relroot)
elif datacmd == 'check':
raise NotImplementedError(datacmd)
else:
raise ValueError(f'unsupported data command {datacmd!r}')
COMMANDS = {
'check': (
'analyze and fail if the given C source/header files have any problems',
[_cli_check],
cmd_check,
),
'analyze': (
'report on the state of the given C source/header files',
[_cli_analyze],
cmd_analyze,
),
'data': (
'check/manage local data (e.g. known types, ignored vars, caches)',
[_cli_data],
cmd_data,
),
}
#######################################
# the script
def parse_args(argv=sys.argv[1:], prog=sys.argv[0], *, subset=None):
import argparse
parser = argparse.ArgumentParser(
prog=prog or get_prog(),
)
processors = add_commands_cli(
parser,
commands={k: v[1] for k, v in COMMANDS.items()},
commonspecs=[
add_verbosity_cli,
add_traceback_cli,
],
subset=subset,
)
args = parser.parse_args(argv)
ns = vars(args)
cmd = ns.pop('cmd')
verbosity, traceback_cm = process_args_by_key(
args,
argv,
processors[cmd],
['verbosity', 'traceback_cm'],
)
# "verbosity" is sent to the commands, so we put it back.
args.verbosity = verbosity
return cmd, ns, verbosity, traceback_cm
def main(cmd, cmd_kwargs):
try:
run_cmd = COMMANDS[cmd][0]
except KeyError:
raise ValueError(f'unsupported cmd {cmd!r}')
run_cmd(**cmd_kwargs)
if __name__ == '__main__':
cmd, cmd_kwargs, verbosity, traceback_cm = parse_args()
configure_logger(verbosity)
with traceback_cm:
main(cmd, cmd_kwargs)