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

126 lines
3.4 KiB
Python
Raw Normal View History

import os.path
from c_common import fsutil
import c_common.tables as _tables
import c_parser.info as _info
import c_parser.match as _match
import c_parser.datafiles as _parser
from . import analyze as _analyze
#############################
# "known" decls
EXTRA_COLUMNS = [
#'typedecl',
]
def get_known(known, extracolumns=None, *,
analyze_resolved=None,
handle_unresolved=True,
relroot=fsutil.USE_CWD,
):
if isinstance(known, str):
known = read_known(known, extracolumns, relroot)
return analyze_known(
known,
handle_unresolved=handle_unresolved,
analyze_resolved=analyze_resolved,
)
def read_known(infile, extracolumns=None, relroot=fsutil.USE_CWD):
extracolumns = EXTRA_COLUMNS + (
list(extracolumns) if extracolumns else []
)
known = {}
for decl, extra in _parser.iter_decls_tsv(infile, extracolumns, relroot):
known[decl] = extra
return known
def analyze_known(known, *,
analyze_resolved=None,
handle_unresolved=True,
):
knowntypes = knowntypespecs = {}
collated = _match.group_by_kinds(known)
types = {decl: None for decl in collated['type']}
typespecs = _analyze.get_typespecs(types)
def analyze_decl(decl):
return _analyze.analyze_decl(
decl,
typespecs,
knowntypespecs,
types,
knowntypes,
analyze_resolved=analyze_resolved,
)
_analyze.analyze_type_decls(types, analyze_decl, handle_unresolved)
return types, typespecs
def write_known(rows, outfile, extracolumns=None, *,
relroot=fsutil.USE_CWD,
backup=True,
):
extracolumns = EXTRA_COLUMNS + (
list(extracolumns) if extracolumns else []
)
_parser.write_decls_tsv(
rows,
outfile,
extracolumns,
relroot=relroot,
backup=backup,
)
#############################
# ignored vars
IGNORED_COLUMNS = [
'filename',
'funcname',
'name',
'reason',
]
IGNORED_HEADER = '\t'.join(IGNORED_COLUMNS)
def read_ignored(infile, relroot=fsutil.USE_CWD):
return dict(_iter_ignored(infile, relroot))
def _iter_ignored(infile, relroot):
if relroot and relroot is not fsutil.USE_CWD:
relroot = os.path.abspath(relroot)
bogus = {_tables.EMPTY, _tables.UNKNOWN}
for row in _tables.read_table(infile, IGNORED_HEADER, sep='\t'):
*varidinfo, reason = row
if _tables.EMPTY in varidinfo or _tables.UNKNOWN in varidinfo:
varidinfo = tuple(None if v in bogus else v
for v in varidinfo)
if reason in bogus:
reason = None
varid = _info.DeclID.from_row(varidinfo)
varid = varid.fix_filename(relroot, formatted=False, fixroot=False)
yield varid, reason
def write_ignored(variables, outfile, relroot=fsutil.USE_CWD):
raise NotImplementedError
if relroot and relroot is not fsutil.USE_CWD:
relroot = os.path.abspath(relroot)
reason = '???'
#if not isinstance(varid, DeclID):
# varid = getattr(varid, 'parsed', varid).id
decls = (d.fix_filename(relroot, fixroot=False) for d in decls)
_tables.write_table(
outfile,
IGNORED_HEADER,
sep='\t',
rows=(r.render_rowdata() + (reason,) for r in decls),
)