# Copyright (c) 2013 - present Facebook, Inc. # All rights reserved. # # This source code is licensed under the BSD style license found in the # LICENSE file in the root directory of this source tree. An additional grant # of patent rights can be found in the PATENTS file in the same directory. from __future__ import absolute_import from __future__ import division from __future__ import print_function from __future__ import unicode_literals import argparse import csv import logging import multiprocessing import os import shutil import subprocess import sys import time from . import config, issues, utils # Increase the limit of the CSV parser to sys.maxlimit csv.field_size_limit(sys.maxsize) INFER_ANALYZE_BINARY = 'InferAnalyze' def get_infer_version(): try: return subprocess.check_output([ utils.get_cmd_in_bin_dir(INFER_ANALYZE_BINARY), '-version']) except subprocess.CalledProcessError: utils.stdout('Failed to run {0} binary, exiting' .format(INFER_ANALYZE_BINARY)) sys.exit(os.EX_UNAVAILABLE) # https://github.com/python/cpython/blob/aa8ea3a6be22c92e774df90c6a6ee697915ca8ec/Lib/argparse.py class VersionAction(argparse._VersionAction): def __call__(self, parser, namespace, values, option_string=None): # set self.version so that argparse version action knows it self.version = get_infer_version() super(VersionAction, self).__call__(parser, namespace, values, option_string) def get_pwd(): pwd = os.getenv('PWD') if pwd is not None: try: # Compare whether 'PWD' and '.' point to same place # Approach is borrowed from llvm implementation of # llvm::sys::fs::current_path (implemented in Path.inc file) pwd_stat = os.stat(pwd) dot_stat = os.stat('.') if pwd_stat.st_dev == dot_stat.st_dev and \ pwd_stat.st_ino == dot_stat.st_ino: return pwd except OSError: # fallthrough to default case pass return os.getcwd() base_parser = argparse.ArgumentParser(add_help=False) base_group = base_parser.add_argument_group('global arguments') base_group.add_argument('-o', '--out', metavar='', default=utils.encode(config.DEFAULT_INFER_OUT), dest='infer_out', type=utils.decode, action=utils.AbsolutePathAction, help='Set the Infer results directory') base_group.add_argument('-r', '--reactive', action='store_true', help='''Analyze in reactive propagation mode starting from changed files.''') base_group.add_argument('-c', '--continue', action='store_true', dest='continue_capture', help='''Continue the capture for the reactive analysis, increasing the changed files/procedures.''') base_group.add_argument('--debug-exceptions', action='store_true', help='''Generate lightweight debugging information: just print the internal exceptions during analysis''') base_group.add_argument('-g', '--debug', action='store_true', help='Generate all debugging information') base_group.add_argument('-a', '--analyzer', help='Select the analyzer within: {0}'.format( ', '.join(config.ANALYZERS)), default=config.ANALYZER_INFER) base_group.add_argument('-nf', '--no-filtering', action='store_true', help='''Also show the results from the experimental checks. Warning: some checks may contain many false alarms''') base_group.add_argument('--android-harness', action='store_true', help='''[experimental] Create harness to detect bugs involving the Android lifecycle''') base_parser.add_argument('-v', '--version', help='''Print the version of Infer and exit''', action=VersionAction) base_group.add_argument('--pmd-xml', action='store_true', help='''Output issues in (PMD) XML format.''') infer_parser = argparse.ArgumentParser(parents=[base_parser]) infer_group = infer_parser.add_argument_group('backend arguments') infer_group.add_argument('-j', '--multicore', metavar='n', type=int, default=multiprocessing.cpu_count(), dest='multicore', help='Set the number of cores to ' 'be used for the analysis (default uses all cores)') infer_group.add_argument('-l', '--load-average', metavar='', type=float, help='Specifies that no new jobs (commands) should ' 'be started if there are others jobs running and the ' 'load average is at least .') infer_group.add_argument('--buck', action='store_true', dest='buck', help='To use when run with buck') infer_group.add_argument('-pr', '--project_root', dest='project_root', default=get_pwd(), help='Location of the project root ' '(default is current directory)') infer_group.add_argument('--absolute-paths', action='store_true', default=False, help='Report errors with absolute paths') infer_group.add_argument('--java-jar-compiler', metavar='') def remove_infer_out(infer_out): # it is safe to ignore errors here because recreating the infer_out # directory will fail later shutil.rmtree(infer_out, True) def create_results_dir(results_dir): utils.mkdir_if_not_exists(results_dir) utils.mkdir_if_not_exists(os.path.join(results_dir, 'specs')) utils.mkdir_if_not_exists(os.path.join(results_dir, 'captured')) utils.mkdir_if_not_exists(os.path.join(results_dir, 'sources')) def reset_start_file(results_dir, touch_if_present=False): start_path = os.path.join(results_dir, '.start') if (not os.path.exists(start_path)) or touch_if_present: # create new empty file - this will update modified timestamp open(start_path, 'w').close() def clean(infer_out): directories = [ 'multicore', 'classnames', 'sources', config.JAVAC_FILELISTS_FILENAME, ] extensions = ['.cfg', '.cg'] for root, dirs, files in os.walk(infer_out): for d in dirs: if d in directories: path = os.path.join(root, d) shutil.rmtree(path) for f in files: for ext in extensions: if f.endswith(ext): path = os.path.join(root, f) os.remove(path) def help_exit(message): utils.stdout(message) infer_parser.print_usage() exit(1) def run_command(cmd, debug_mode, javac_arguments, step, analyzer): if debug_mode: utils.stdout('\n{0}\n'.format(' '.join(cmd))) try: return subprocess.check_call(cmd) except subprocess.CalledProcessError as e: error_msg = 'Failure during {0}, original command was\n\n{1}\n\n' infer_cmd = ['infer', '-g', '-a', analyzer] failing_cmd = infer_cmd + ['--', 'javac'] + javac_arguments logging.error(error_msg.format( step, failing_cmd )) raise e class AnalyzerWrapper(object): javac = None def __init__(self, args): self.args = args if self.args.analyzer not in config.ANALYZERS: help_exit('Unknown analysis mode \"{0}\"' .format(self.args.analyzer)) if self.args.infer_out is None: help_exit('Expect Infer results directory') try: os.mkdir(self.args.infer_out) except OSError as e: if not os.path.isdir(self.args.infer_out): raise e self.stats = {'int': {}} self.timing = {} def clean_exit(self): if os.path.isdir(self.args.infer_out): utils.stdout('removing {}'.format(self.args.infer_out)) shutil.rmtree(self.args.infer_out) exit(os.EX_OK) def analyze(self): logging.info('Starting analysis') infer_analyze = [ utils.get_cmd_in_bin_dir(INFER_ANALYZE_BINARY), '-results_dir', self.args.infer_out ] infer_options = [] # remove specs if possible so that old issues are less likely # to be reported infer_options += ['-allow_specs_cleanup'] infer_options += ['-inferconfig_home', utils.decode(os.getcwd())] if self.args.analyzer == config.ANALYZER_ERADICATE: infer_options += ['-eradicate'] elif self.args.analyzer == config.ANALYZER_CRASHCONTEXT: infer_options += ['-crashcontext'] elif self.args.analyzer == config.ANALYZER_CHECKERS: infer_options += ['-checkers'] elif self.args.analyzer == config.ANALYZER_QUANDARY: infer_options += ['-quandary'] else: if self.args.analyzer == config.ANALYZER_TRACING: infer_options.append('-tracing') if os.path.isfile(config.MODELS_JAR): infer_options += ['-models', config.MODELS_JAR] if self.args.debug: infer_options += [ '-developer_mode', '-html', '-dotty', '-print_types', '-trace_error', '-print_buckets', '-notest' ] self.args.no_filtering = True elif self.args.debug_exceptions: infer_options.append('-developer_mode') infer_options.append('-print_buckets') self.args.no_filtering = True exit_status = os.EX_OK if self.javac is not None and self.args.buck: infer_options += ['-project_root', utils.decode(os.getcwd()), '-java'] if self.javac.args.classpath is not None: for path in self.javac.args.classpath.split(os.pathsep): if os.path.isfile(path): infer_options += ['-ziplib', os.path.abspath(path)] elif self.args.project_root: infer_options += ['-project_root', self.args.project_root] infer_options = map(utils.decode_or_not, infer_options) infer_options_str = ' '.join(infer_options) os.environ['INFER_OPTIONS'] = utils.encode(infer_options_str) javac_original_arguments = \ self.javac.original_arguments if self.javac is not None else [] if self.args.analyzer == config.ANALYZER_TRACING: os.environ['INFER_LAZY_DYNAMIC_DISPATCH'] = 'Y' if self.args.multicore == 1: analysis_start_time = time.time() analyze_cmd = infer_analyze + infer_options exit_status = run_command( analyze_cmd, self.args.debug, javac_original_arguments, 'analysis', self.args.analyzer ) elapsed = utils.elapsed_time(analysis_start_time) self.timing['analysis'] = elapsed self.timing['makefile_generation'] = 0 else: multicore_dir = os.path.join(self.args.infer_out, 'multicore') pwd = os.getcwd() if os.path.isdir(multicore_dir): shutil.rmtree(multicore_dir) os.mkdir(multicore_dir) os.chdir(multicore_dir) analyze_cmd = infer_analyze + ['-makefile', 'Makefile'] analyze_cmd += infer_options makefile_generation_start_time = time.time() makefile_status = run_command( analyze_cmd, self.args.debug, javac_original_arguments, 'create_makefile', self.args.analyzer ) elapsed = utils.elapsed_time(makefile_generation_start_time) self.timing['makefile_generation'] = elapsed exit_status += makefile_status if makefile_status == os.EX_OK: make_cmd = ['make', '-k'] make_cmd += ['-j', str(self.args.multicore)] if self.args.load_average is not None: make_cmd += ['-l', str(self.args.load_average)] if not self.args.debug: make_cmd += ['-s'] analysis_start_time = time.time() make_status = run_command( make_cmd, self.args.debug, javac_original_arguments, 'run_makefile', self.args.analyzer ) elapsed = utils.elapsed_time(analysis_start_time) self.timing['analysis'] = elapsed os.chdir(pwd) exit_status += make_status if self.args.buck and exit_status == os.EX_OK: clean(self.args.infer_out) return exit_status def update_stats_with_warnings(self, csv_report): with open(csv_report, 'r') as file_in: reader = utils.locale_csv_reader(file_in) rows = [row for row in reader][1:] for row in rows: key = row[issues.CSV_INDEX_TYPE] previous_value = self.stats['int'].get(key, 0) self.stats['int'][key] = previous_value + 1 def create_report(self): """Report statistics about the computation and create a CSV file containing the list or errors found during the analysis""" out_dir = self.args.infer_out csv_report = os.path.join(out_dir, config.CSV_REPORT_FILENAME) json_report = os.path.join(out_dir, config.JSON_REPORT_FILENAME) procs_report = os.path.join(self.args.infer_out, 'procs.csv') infer_print_cmd = [utils.get_cmd_in_bin_dir('InferPrint')] infer_print_options = [ '-q', '-results_dir', self.args.infer_out, '-bugs', csv_report, '-bugs_json', json_report, '-procs', procs_report, '-analyzer', self.args.analyzer ] if self.args.debug or self.args.debug_exceptions: infer_print_options.append('-with_infer_src_loc') exit_status = subprocess.check_call( infer_print_cmd + infer_print_options ) if exit_status != os.EX_OK: logging.error( 'Error with InferPrint with the command: {}'.format( infer_print_cmd)) else: issues.clean_csv(self.args, csv_report) issues.clean_json(self.args, json_report) self.update_stats_with_warnings(csv_report) return exit_status def read_proc_stats(self): proc_stats_path = os.path.join( self.args.infer_out, config.PROC_STATS_FILENAME) # capture and compile mode do not create proc_stats.json if os.path.isfile(proc_stats_path): proc_stats = utils.load_json_from_path(proc_stats_path) self.stats['int'].update(proc_stats) def save_stats(self): """Print timing information to infer_out/stats.json""" self.stats['float'] = { 'capture_time': self.timing.get('capture', 0.0), 'makefile_generation_time': self.timing.get( 'makefile_generation', 0.0), 'analysis_time': self.timing.get('analysis', 0.0), 'reporting_time': self.timing.get('reporting', 0.0), } self.stats['normal'] = { 'analyzer': self.args.analyzer, 'infer_version': utils.infer_version() } stats_path = os.path.join(self.args.infer_out, config.STATS_FILENAME) utils.dump_json_to_path(self.stats, stats_path) def report_proc_stats(self): self.read_proc_stats() self.print_analysis_stats() def report(self): reporting_start_time = time.time() report_status = self.create_report() elapsed = utils.elapsed_time(reporting_start_time) self.timing['reporting'] = elapsed if report_status == os.EX_OK and not self.args.buck: json_report = os.path.join(self.args.infer_out, config.JSON_REPORT_FILENAME) bugs_out = os.path.join(self.args.infer_out, config.BUGS_FILENAME) xml_out = None if self.args.pmd_xml: xml_out = os.path.join(self.args.infer_out, config.PMD_XML_FILENAME) issues.print_and_save_errors(json_report, bugs_out, xml_out) def analyze_and_report(self): if self.args.analyzer not in [config.ANALYZER_COMPILE, config.ANALYZER_CAPTURE]: if self.args.analyzer == config.ANALYZER_LINTERS: self.report() elif self.analyze() == os.EX_OK: self.report_proc_stats() self.report() def print_analysis_stats(self): files_total = self.stats['int']['files'] files_str = utils.get_plural('file', files_total) print('Analyzed {}'.format(files_str))