#!/usr/bin/env python3 import argparse import os import re import sys arg_parser = argparse.ArgumentParser( description=""" Reports time consumed for one or more task in a format similar to the standard Bash 'time' builtin. Optionally sorts tasks by real (wall-clock), user (user space CPU), or sys (kernel CPU) time. """) arg_parser.add_argument( "paths", metavar="path", nargs="+", help=""" A path containing task buildstats. If the path is a directory, e.g. build/tmp/buildstats, then all task found (recursively) in it will be processed. If the path is a single task buildstat, e.g. build/tmp/buildstats/20161018083535/foo-1.0-r0/do_compile, then just that buildstat will be processed. Multiple paths can be specified to process all of them. Files whose names do not start with "do_" are ignored. """) arg_parser.add_argument( "--sort", choices=("none", "real", "user", "sys"), default="none", help=""" The measurement to sort the output by. Defaults to 'none', which means to sort by the order paths were given on the command line. For other options, tasks are sorted in descending order from the highest value. """) args = arg_parser.parse_args() # Field names and regexes for parsing out their values from buildstat files field_regexes = (("elapsed", ".*Elapsed time: ([0-9.]+)"), ("user", "rusage ru_utime: ([0-9.]+)"), ("sys", "rusage ru_stime: ([0-9.]+)"), ("child user", "Child rusage ru_utime: ([0-9.]+)"), ("child sys", "Child rusage ru_stime: ([0-9.]+)")) # A list of (<path>, <dict>) tuples, where <path> is the path of a do_* task # buildstat file and <dict> maps fields from the file to their values task_infos = [] def save_times_for_task(path): """Saves information for the buildstat file 'path' in 'task_infos'.""" if not os.path.basename(path).startswith("do_"): return with open(path) as f: fields = {} for line in f: for name, regex in field_regexes: match = re.match(regex, line) if match: fields[name] = float(match.group(1)) break # Check that all expected fields were present for name, regex in field_regexes: if name not in fields: print("Warning: Skipping '{}' because no field matching '{}' could be found" .format(path, regex), file=sys.stderr) return task_infos.append((path, fields)) def save_times_for_dir(path): """Runs save_times_for_task() for each file in path and its subdirs, recursively.""" # Raise an exception for os.walk() errors instead of ignoring them def walk_onerror(e): raise e for root, _, files in os.walk(path, onerror=walk_onerror): for fname in files: save_times_for_task(os.path.join(root, fname)) for path in args.paths: if os.path.isfile(path): save_times_for_task(path) else: save_times_for_dir(path) def elapsed_time(task_info): return task_info[1]["elapsed"] def tot_user_time(task_info): return task_info[1]["user"] + task_info[1]["child user"] def tot_sys_time(task_info): return task_info[1]["sys"] + task_info[1]["child sys"] if args.sort != "none": sort_fn = {"real": elapsed_time, "user": tot_user_time, "sys": tot_sys_time} task_infos.sort(key=sort_fn[args.sort], reverse=True) first_entry = True # Catching BrokenPipeError avoids annoying errors when the output is piped into # e.g. 'less' or 'head' and not completely read try: for task_info in task_infos: real = elapsed_time(task_info) user = tot_user_time(task_info) sys = tot_sys_time(task_info) if not first_entry: print() first_entry = False # Mimic Bash's 'time' builtin print("{}:\n" "real\t{}m{:.3f}s\n" "user\t{}m{:.3f}s\n" "sys\t{}m{:.3f}s" .format(task_info[0], int(real//60), real%60, int(user//60), user%60, int(sys//60), sys%60)) except BrokenPipeError: pass