mprof.py

import glob
import os
import os.path as osp
import sys
import re
import copy
import time
import math
import logging
import itertools
from ast import literal_eval

from collections import defaultdict
from argparse import ArgumentParser, ArgumentError, REMAINDER, RawTextHelpFormatter

import importlib
import memory_profiler as mp

ALL_ACTIONS = ("run", "rm", "clean", "list", "plot", "attach", "peak")
help_msg = """
Available commands:

    run      run a given command or python file
    attach   alias for 'run --attach': attach to an existing process by pid or name
    rm       remove a given file generated by mprof
    clean    clean the current directory from files created by mprof
    list     display existing profiles, with indices
    plot     plot memory consumption generated by mprof run
    peak     print the maximum memory used by an mprof run

Type mprof <command> --help for usage help on a specific command.
For example, mprof plot --help will list all plotting options.
"""

logger = logging.getLogger(__name__)
logging.basicConfig()


def print_usage():
    print("Usage: %s <command> <options> <arguments>"
          % osp.basename(sys.argv[0]))
    print(help_msg)


def get_action():
    """Pop first argument, check it is a valid action."""
    if len(sys.argv) <= 1:
        print_usage()
        sys.exit(1)
    if not sys.argv[1] in ALL_ACTIONS:
        print_usage()
        sys.exit(1)

    return sys.argv.pop(1)


def get_profile_filenames(args):
    """Return list of profile filenames.

    Parameters
    ==========
    args (list)
        list of filename or integer. An integer is the index of the
        profile in the list of existing profiles. 0 is the oldest,
        -1 in the more recent.
        Non-existing files cause a ValueError exception to be thrown.

    Returns
    =======
    filenames (list)
        list of existing memory profile filenames. It is guaranteed
        that an given file name will not appear twice in this list.
    """
    profiles = glob.glob("mprofile_??????????????.dat")
    profiles.sort()

    if args == "all":
        filenames = copy.copy(profiles)
    else:
        filenames = []
        for arg in args:
            if arg == "--":  # workaround
                continue
            try:
                index = int(arg)
            except ValueError:
                index = None
            if index is not None:
                try:
                    filename = profiles[index]
                except IndexError:
                    raise ValueError("Invalid index (non-existing file): %s" % arg)

                if filename not in filenames:
                    filenames.append(filename)
            else:
                if osp.isfile(arg):
                    if arg not in filenames:
                        filenames.append(arg)
                elif osp.isdir(arg):
                    raise ValueError("Path %s is a directory" % arg)
                else:
                    raise ValueError("File %s not found" % arg)

    # Add timestamp files, if any
    for filename in reversed(filenames):
        parts = osp.splitext(filename)
        timestamp_file = parts[0] + "_ts" + parts[1]
        if osp.isfile(timestamp_file) and timestamp_file not in filenames:
            filenames.append(timestamp_file)

    return filenames


def list_action():
    """Display existing profiles, with indices."""
    parser = ArgumentParser(
            usage='mprof list\nThis command takes no argument.')
    parser.add_argument('--version', action='version', version=mp.__version__)
    args = parser.parse_args()

    filenames = get_profile_filenames("all")
    for n, filename in enumerate(filenames):
        ts = osp.splitext(filename)[0].split('_')[-1]
        print("{index} {filename} {hour}:{min}:{sec} {day}/{month}/{year}"
              .format(index=n, filename=filename,
                      year=ts[:4], month=ts[4:6], day=ts[6:8],
                      hour=ts[8:10], min=ts[10:12], sec=ts[12:14]))


def rm_action():
    """TODO: merge with clean_action (@pgervais)"""
    parser = ArgumentParser(usage='mprof rm [options] numbers_or_filenames')
    parser.add_argument('--version', action='version', version=mp.__version__)
    parser.add_argument("--dry-run", dest="dry_run", default=False,
                        action="store_true",
                        help="""Show what will be done, without actually doing it.""")
    parser.add_argument("numbers_or_filenames", nargs='*',
                        help="""numbers or filenames removed""")
    args = parser.parse_args()

    if len(args.numbers_or_filenames) == 0:
        print("A profile to remove must be provided (number or filename)")
        sys.exit(1)

    filenames = get_profile_filenames(args.numbers_or_filenames)
    if args.dry_run:
        print("Files to be removed: ")
        for filename in filenames:
            print(filename)
    else:
        for filename in filenames:
            os.remove(filename)


def clean_action():
    """Remove every profile file in current directory."""
    parser = ArgumentParser(
            usage='mprof clean\nThis command takes no argument.')
    parser.add_argument('--version', action='version', version=mp.__version__)
    parser.add_argument("--dry-run", dest="dry_run", default=False,
                        action="store_true",
                        help="""Show what will be done, without actually doing it.""")
    args = parser.parse_args()

    filenames = get_profile_filenames("all")
    if args.dry_run:
        print("Files to be removed: ")
        for filename in filenames:
            print(filename)
    else:
        for filename in filenames:
            os.remove(filename)


def get_cmd_line(args):
    """Given a set or arguments, compute command-line."""
    blanks = set(' \t')
    args = [s if blanks.isdisjoint(s) else "'" + s + "'" for s in args]
    return ' '.join(args)

def find_first_process(name):
    for i in mp.psutil.process_iter():
        if name in i.name():
            return i
    return None

def attach_action():
    argv = sys.argv
    sys.argv = argv[:1] + ['--attach'] + argv[1:]
    run_action()

def run_action():
    import time, subprocess
    parser = ArgumentParser(usage="mprof run [options] program", formatter_class=RawTextHelpFormatter)
    parser.add_argument('--version', action='version', version=mp.__version__)
    parser.add_argument("--python", dest="python", action="store_true",
                        help="""Activates extra features when the profiling executable is a Python program (currently: function timestamping.)""")
    parser.add_argument("--nopython", dest="nopython", action="store_true",
                        help="""Disables extra features when the profiled executable is a Python program (currently: function timestamping.)""")
    parser.add_argument("--interval", "-T", dest="interval", default="0.1", type=float, action="store",
                        help="Sampling period (in seconds), defaults to 0.1")
    parser.add_argument("--include-children", "-C", dest="include_children", action="store_true",
                        help="""Monitors forked processes as well (sum up all process memory)""")
    parser.add_argument("--multiprocess", "-M", dest="multiprocess", action="store_true",
                        help="""Monitors forked processes creating individual plots for each child (disables --python features)""")
    parser.add_argument("--exit-code", "-E", dest="exit_code", action="store_true", help="""Propagate the exit code""")
    attach_arg = parser.add_argument("--attach", "-a", dest="attach_existing", action="store_true",
                        help="Attach to an existing process, by process name or by pid")
    parser.add_argument("--timeout", "-t", dest="timeout", action="store", type=int,
                        help="timeout in seconds for the profiling, default new process has no timeout, attach existing is 1 hour")
    parser.add_argument("--output", "-o", dest="filename",
                        default="mprofile_%s.dat" % time.strftime("%Y%m%d%H%M%S", time.localtime()),
                        help="""File to store results in, defaults to 'mprofile_<YYYYMMDDhhmmss>.dat' in the current directory,
(where <YYYYMMDDhhmmss> is the date-time of the program start).
This file contains the process memory consumption, in Mb (one value per line).""")
    parser.add_argument("--backend", dest="backend", choices=["psutil", "psutil_pss", "psutil_uss", "posix", "tracemalloc"],
                        default="psutil",
                        help="Current supported backends: 'psutil', 'psutil_pss', 'psutil_uss', 'posix', 'tracemalloc'. Defaults to 'psutil'.")
    parser.add_argument("program", nargs=REMAINDER,
                        help='Option 1: "<EXECUTABLE> <ARG1> <ARG2>..." - profile executable\n'
                             'Option 2: "<PYTHON_SCRIPT> <ARG1> <ARG2>..." - profile python script\n'
                             'Option 3: (--python flag present) "<PYTHON_EXECUTABLE> <PYTHON_SCRIPT> <ARG1> <ARG2>..." - profile python script with specified interpreter\n'
                             'Option 4: (--python flag present) "<PYTHON_MODULE> <ARG1> <ARG2>..." - profile python module\n'
                        )
    args = parser.parse_args()

    if len(args.program) == 0:
        print("A program to run must be provided. Use -h for help")
        sys.exit(1)

    print("{1}: Sampling memory every {0}s".format(
        args.interval, osp.basename(sys.argv[0])))

    mprofile_output = args.filename

    program = args.program
    if args.attach_existing:
        print('attaching to existing process, using hint: {}'.format(program[0]))
        if program[0].isdigit():
            p = literal_eval(program[0])
            cmd_line = get_cmd_line(program)
        else:
            proc = find_first_process(program[0])
            if proc is None:
                raise ArgumentError(attach_arg, '\nWhen attaching, program should be process name or pid.\nFailed to find a process using hint: {}'.format(program[0]))
            
            p = proc.pid
            try:
                cmd_line = proc.cmdline()
            except:
                cmd_line = get_cmd_line(program)
        if args.timeout is None:
            args.timeout = 3600
    else:
        print('running new process')
        # .. TODO: more than one script as argument ? ..
        if program[0].endswith('.py') and not args.nopython:
            if args.multiprocess:
                # in multiprocessing mode you want to spawn a separate
                # python process
                if not program[0].startswith("python"):
                    program.insert(0, sys.executable)
                args.python = False
            else:
                args.python = True
        if args.python:
            print("running as a Python program...")
            if not program[0].startswith("python"):
                program.insert(0, sys.executable)
            cmd_line = get_cmd_line(program)
            extra_args = ["-m", "memory_profiler", "--timestamp", "-o", mprofile_output]
            if args.include_children:
                extra_args.append("--include-children")
            program[1:1] = extra_args
            p = subprocess.Popen(program)
        else:
            cmd_line = get_cmd_line(program)
            p = subprocess.Popen(program)

    with open(mprofile_output, "a") as f:
        f.write("CMDLINE {0}\n".format(cmd_line))
        mp.memory_usage(proc=p, interval=args.interval, timeout=args.timeout, timestamps=True,
                        include_children=args.include_children,
                        multiprocess=args.multiprocess, stream=f, backend=args.backend)

    if args.exit_code:
        if p.returncode != 0:
            logger.error('Program resulted with a non-zero exit code: %s', p.returncode)
        sys.exit(p.returncode)


def add_brackets(xloc, yloc, xshift=0, color="r", label=None, options=None):
    """Add two brackets on the memory line plot.

    This function uses the current figure.

    Parameters
    ==========
    xloc: tuple with 2 values
        brackets location (on horizontal axis).
    yloc: tuple with 2 values
        brackets location (on vertical axis)
    xshift: float
        value to subtract to xloc.
    """
    try:
        import pylab as pl
    except ImportError as e:
        print("matplotlib is needed for plotting.")
        print(e)
        sys.exit(1)
    height_ratio = 20.
    vsize = (pl.ylim()[1] - pl.ylim()[0]) / height_ratio
    hsize = (pl.xlim()[1] - pl.xlim()[0]) / (3. * height_ratio)

    bracket_x = pl.asarray([hsize, 0, 0, hsize])
    bracket_y = pl.asarray([vsize, vsize, -vsize, -vsize])

    # Matplotlib workaround: labels starting with _ aren't displayed
    if label[0] == '_':
        label = ' ' + label
    if options.xlim is None or options.xlim[0] <= (xloc[0] - xshift) <= options.xlim[1]:
        pl.plot(bracket_x + xloc[0] - xshift, bracket_y + yloc[0],
                "-" + color, linewidth=2, label=label)
    if options.xlim is None or options.xlim[0] <= (xloc[1] - xshift) <= options.xlim[1]:
        pl.plot(-bracket_x + xloc[1] - xshift, bracket_y + yloc[1],
                "-" + color, linewidth=2)

        # TODO: use matplotlib.patches.Polygon to draw a colored background for
        # each function.

        # with maplotlib 1.2, use matplotlib.path.Path to create proper markers
        # see http://matplotlib.org/examples/pylab_examples/marker_path.html
        # This works with matplotlib 0.99.1
        ## pl.plot(xloc[0], yloc[0], "<"+color, markersize=7, label=label)
        ## pl.plot(xloc[1], yloc[1], ">"+color, markersize=7)


def read_mprofile_file(filename):
    """Read an mprofile file and return its content.

    Returns
    =======
    content: dict
        Keys:

        - "mem_usage": (list) memory usage values, in MiB
        - "timestamp": (list) time instant for each memory usage value, in
            second
        - "func_timestamp": (dict) for each function, timestamps and memory
            usage upon entering and exiting.
        - 'cmd_line': (str) command-line ran for this profile.
    """
    func_ts = {}
    mem_usage = []
    timestamp = []
    children  = defaultdict(list)
    cmd_line = None
    f = open(filename, "r")
    for l in f:
        if l == '\n':
            raise ValueError('Sampling time was too short')
        field, value = l.split(' ', 1)
        if field == "MEM":
            # mem, timestamp
            values = value.split(' ')
            mem_usage.append(float(values[0]))
            timestamp.append(float(values[1]))

        elif field == "FUNC":
            values = value.split(' ')
            f_name, mem_start, start, mem_end, end = values[:5]
            ts = func_ts.get(f_name, [])
            to_append = [float(start), float(end), float(mem_start), float(mem_end)]
            if len(values) >= 6:
                # There is a stack level field
                stack_level = values[5]
                to_append.append(int(stack_level))
            ts.append(to_append)
            func_ts[f_name] = ts

        elif field == "CHLD":
            values = value.split(' ')
            chldnum = values[0]
            children[chldnum].append(
                (float(values[1]), float(values[2]))
            )

        elif field == "CMDLINE":
            cmd_line = value
        else:
            pass
    f.close()

    return {"mem_usage": mem_usage, "timestamp": timestamp,
            "func_timestamp": func_ts, 'filename': filename,
            'cmd_line': cmd_line, 'children': children}


def plot_file(filename, index=0, timestamps=True, children=True, options=None):
    try:
        import pylab as pl
    except ImportError as e:
        print("matplotlib is needed for plotting.")
        print(e)
        sys.exit(1)
    import numpy as np  # pylab requires numpy anyway
    mprofile = read_mprofile_file(filename)

    if len(mprofile['timestamp']) == 0:
        print('** No memory usage values have been found in the profile '
              'file.**\nFile path: {0}\n'
              'File may be empty or invalid.\n'
              'It can be deleted with "mprof rm {0}"'.format(
            mprofile['filename']))
        sys.exit(0)

    # Merge function timestamps and memory usage together
    ts = mprofile['func_timestamp']
    t = mprofile['timestamp']
    mem = mprofile['mem_usage']
    chld = mprofile['children']

    if len(ts) > 0:
        for values in ts.values():
            for v in values:
                t.extend(v[:2])
                mem.extend(v[2:4])

    mem = np.asarray(mem)
    t = np.asarray(t)
    ind = t.argsort()
    mem = mem[ind]
    t = t[ind]

    # Plot curves
    global_start = float(t[0])
    t = t - global_start

    max_mem = mem.max()
    max_mem_ind = mem.argmax()

    all_colors = ("c", "y", "g", "r", "b")
    mem_line_colors = ("k", "b", "r", "g", "c", "y", "m")

    show_trend_slope = options is not None and hasattr(options, 'slope') and options.slope is True

    mem_line_label = time.strftime("%d / %m / %Y - start at %H:%M:%S",
                                   time.localtime(global_start)) \
                     + ".{0:03d}".format(int(round(math.modf(global_start)[0] * 1000)))

    mem_trend = None
    if show_trend_slope:
        # Compute trend line
        mem_trend = np.polyfit(t, mem, 1)

        # Append slope to label
        mem_line_label = mem_line_label + " slope {0:.5f}".format(mem_trend[0])

    pl.plot(t, mem, "+-" + mem_line_colors[index % len(mem_line_colors)],
            label=mem_line_label)

    if show_trend_slope:
        # Plot the trend line
        pl.plot(t, t*mem_trend[0] + mem_trend[1], "--", linewidth=0.5, color="#00e3d8")

    bottom, top = pl.ylim()
    bottom += 0.001
    top -= 0.001

    # plot children, if any
    if len(chld) > 0 and children:
        cmpoint = (0,0) # maximal child memory

        for idx, (proc, data) in enumerate(chld.items()):
            # Create the numpy arrays from the series data
            cts  = np.asarray([item[1] for item in data]) - global_start
            cmem = np.asarray([item[0] for item in data])

            cmem_trend = None
            child_mem_trend_label = ""
            if show_trend_slope:
                # Compute trend line
                cmem_trend = np.polyfit(cts, cmem, 1)

                child_mem_trend_label = " slope {0:.5f}".format(cmem_trend[0])

            # Plot the line to the figure
            pl.plot(cts, cmem, "+-" + mem_line_colors[(idx + 1) % len(mem_line_colors)],
                    label="child {}{}".format(proc, child_mem_trend_label))

            if show_trend_slope:
                # Plot the trend line
                pl.plot(cts, cts*cmem_trend[0] + cmem_trend[1], "--", linewidth=0.5, color="black")

            # Detect the maximal child memory point
            cmax_mem = cmem.max()
            if cmax_mem > cmpoint[1]:
                cmpoint = (cts[cmem.argmax()], cmax_mem)

        # Add the marker lines for the maximal child memory usage
        pl.vlines(cmpoint[0], pl.ylim()[0]+0.001, pl.ylim()[1] - 0.001, 'r', '--')
        pl.hlines(cmpoint[1], pl.xlim()[0]+0.001, pl.xlim()[1] - 0.001, 'r', '--')

    # plot timestamps, if any
    if len(ts) > 0 and timestamps:
        func_num = 0
        f_labels = function_labels(ts.keys())
        for f, exec_ts in ts.items():
            for execution in exec_ts:
                add_brackets(execution[:2], execution[2:], xshift=global_start,
                             color=all_colors[func_num % len(all_colors)],
                             label=f_labels[f]
                                   + " %.3fs" % (execution[1] - execution[0]), options=options)
            func_num += 1

    if timestamps:
        pl.hlines(max_mem,
                  pl.xlim()[0] + 0.001, pl.xlim()[1] - 0.001,
                  colors="r", linestyles="--")
        pl.vlines(t[max_mem_ind], bottom, top,
                  colors="r", linestyles="--")
    return mprofile



FLAME_PLOTTER_VARS = {
    'hovered_rect': None,
    'hovered_text': None,
    'alpha': None
}

def flame_plotter(filename, index=0, timestamps=True, children=True, options=None):
    try:
        import pylab as pl
    except ImportError as e:
        print("matplotlib is needed for plotting.")
        print(e)
        sys.exit(1)
    import numpy as np  # pylab requires numpy anyway
    mprofile = read_mprofile_file(filename)

    if len(mprofile['timestamp']) == 0:
        print('** No memory usage values have been found in the profile '
              'file.**\nFile path: {0}\n'
              'File may be empty or invalid.\n'
              'It can be deleted with "mprof rm {0}"'.format(
            mprofile['filename']))
        sys.exit(0)

    # Merge function timestamps and memory usage together
    ts = mprofile['func_timestamp']
    t = mprofile['timestamp']
    mem = mprofile['mem_usage']
    chld = mprofile['children']

    if len(ts) > 0:
        for values in ts.values():
            for v in values:
                t.extend(v[:2])
                mem.extend(v[2:4])

    mem = np.asarray(mem)
    t = np.asarray(t)
    ind = t.argsort()
    mem = mem[ind]
    t = t[ind]

    if ts:
        stack_size = 1 + max(ex[4] for executions in ts.values() for ex in executions)
    else:
        stack_size = 0
    def level_to_saturation(level):
        return 1 - 0.75 * level / stack_size

    colors = [
        itertools.cycle([
            pl.matplotlib.colors.hsv_to_rgb((0, level_to_saturation(level), 1)),
            pl.matplotlib.colors.hsv_to_rgb((0.1, level_to_saturation(level), 1)),
        ]) for level in range(stack_size)
    ]

    # Plot curves
    global_start = float(t[0])
    t = t - global_start

    max_mem = mem.max()
    max_mem_ind = mem.argmax()

    # cmap = pl.cm.get_cmap('gist_rainbow')
    mem_line_colors = ("k", "b", "r", "g", "c", "y", "m")
    mem_line_label = time.strftime("%d / %m / %Y - start at %H:%M:%S",
                                   time.localtime(global_start)) \
                     + ".{0:03d}".format(int(round(math.modf(global_start)[0] * 1000)))

    pl.plot(t, mem, "-" + mem_line_colors[index % len(mem_line_colors)],
            label=mem_line_label)

    bottom, top = pl.ylim()
    bottom += 0.001
    top -= 0.001

    ax = pl.gca()
    ax.grid(True)
    timestamp_ax = ax.twinx()
    timestamp_ax.set_yticks([])
    timestamp_ax.set_ylim((0, stack_size + 1))
    timestamp_ax.grid(False)

    # plot children, if any
    if len(chld) > 0 and children:
        cmpoint = (0,0) # maximal child memory

        for idx, (proc, data) in enumerate(chld.items()):
            # Create the numpy arrays from the series data
            cts  = np.asarray([item[1] for item in data]) - global_start
            cmem = np.asarray([item[0] for item in data])

            # Plot the line to the figure
            pl.plot(cts, cmem, "+-"  + mem_line_colors[(idx+1) % len(mem_line_colors)],
                     label="child {}".format(proc))

            # Detect the maximal child memory point
            cmax_mem = cmem.max()
            if cmax_mem > cmpoint[1]:
                cmpoint = (cts[cmem.argmax()], cmax_mem)

        # Add the marker lines for the maximal child memory usage
        pl.vlines(cmpoint[0], pl.ylim()[0]+0.001, pl.ylim()[1] - 0.001, 'r', '--')
        pl.hlines(cmpoint[1], pl.xlim()[0]+0.001, pl.xlim()[1] - 0.001, 'r', '--')

    def mouse_motion_handler(event):
        x, y = event.xdata, event.ydata
        if x is not None and y is not None:
            for coord, (name, text, rect) in rectangles.items():
                x0, y0, x1, y1 = coord
                if x0 < x < x1 and y0 < y < y1:
                    if FLAME_PLOTTER_VARS['hovered_rect'] == rect:
                        return

                    if FLAME_PLOTTER_VARS['hovered_rect'] is not None:
                        FLAME_PLOTTER_VARS['hovered_rect'].set_alpha(FLAME_PLOTTER_VARS['alpha'])
                        FLAME_PLOTTER_VARS['hovered_text'].set_color((0, 0, 0, 0))
                        FLAME_PLOTTER_VARS['hovered_rect'].set_linewidth(1)

                    FLAME_PLOTTER_VARS['hovered_text'] = text
                    FLAME_PLOTTER_VARS['hovered_rect'] = rect
                    FLAME_PLOTTER_VARS['alpha'] = rect.get_alpha()
                    FLAME_PLOTTER_VARS['hovered_rect'].set_alpha(0.8)
                    FLAME_PLOTTER_VARS['hovered_rect'].set_linewidth(3)
                    FLAME_PLOTTER_VARS['hovered_text'].set_color((0, 0, 0, 1))
                    pl.draw()
                    return

        if FLAME_PLOTTER_VARS['hovered_rect'] is not None:
            FLAME_PLOTTER_VARS['hovered_text'].set_color((0, 0, 0, 0))
            FLAME_PLOTTER_VARS['hovered_rect'].set_alpha(FLAME_PLOTTER_VARS['alpha'])
            FLAME_PLOTTER_VARS['hovered_rect'].set_linewidth(1)
            pl.draw()
            FLAME_PLOTTER_VARS['hovered_rect'] = None
            FLAME_PLOTTER_VARS['hovered_text'] = None

    def mouse_click_handler(event):
        x, y = event.xdata, event.ydata
        if x is None or y is None:
            return

        for coord, _ in rectangles.items():
            x0, y0, x1, y1 = coord
            if x0 < x < x1 and y0 < y < y1:
                toolbar = pl.gcf().canvas.toolbar
                toolbar.push_current()
                timestamp_ax.set_xlim(x0, x1)
                timestamp_ax.set_ylim(y0, stack_size + 1)
                toolbar.push_current()
                pl.draw()
                return

    # plot timestamps, if any
    if len(ts) > 0 and timestamps:
        func_num = 0
        f_labels = function_labels(ts.keys())
        rectangles = {}
        for f, exec_ts in ts.items():
            for execution in exec_ts:
                x0, x1 = execution[:2]
                y0 = execution[4]
                y1 = y0 + 1
                x0 -= global_start
                x1 -= global_start
                color = next(colors[y0])
                rect, text = add_timestamp_rectangle(
                    timestamp_ax,
                    x0, x1, y0, y1, f,
                    color=color
                )
                rectangles[(x0, y0, x1, y1)] = (f, text, rect)
            func_num += 1

        # Disable hovering if there are too many rectangle to prevent slow down
        if len(rectangles) < 100:
            pl.gcf().canvas.mpl_connect('motion_notify_event', mouse_motion_handler)
        pl.gcf().canvas.mpl_connect('button_release_event', mouse_click_handler)

    if timestamps:
        pl.hlines(max_mem,
                  pl.xlim()[0] + 0.001, pl.xlim()[1] - 0.001,
                  colors="r", linestyles="--")
        pl.vlines(t[max_mem_ind], bottom, top,
                  colors="r", linestyles="--")

    pl.sca(ax)

    return mprofile


def add_timestamp_rectangle(ax, x0, x1, y0, y1, func_name, color='none'):
    rect = ax.fill_betweenx((y0, y1), x0, x1, color=color, alpha=0.5, linewidth=1)
    text = ax.text(x0, y1, func_name,
        horizontalalignment='left',
        verticalalignment='top',
        color=(0, 0, 0, 0)
    )
    return rect, text


def function_labels(dotted_function_names):
    state = {}

    def set_state_for(function_names, level):
        for fn in function_names:
            label = ".".join(fn.split(".")[-level:])
            label_state = state.setdefault(label, {"functions": [],
                                                   "level": level})
            label_state["functions"].append(fn)

    set_state_for(dotted_function_names, 1)

    while True:
        ambiguous_labels = [label for label in state if len(state[label]["functions"]) > 1]
        for ambiguous_label in ambiguous_labels:
            function_names = state[ambiguous_label]["functions"]
            new_level = state[ambiguous_label]["level"] + 1
            del state[ambiguous_label]
            set_state_for(function_names, new_level)
        if len(ambiguous_labels) == 0:
            break

    fn_to_label = dict((label_state["functions"][0] , label) for label, label_state in state.items())

    return fn_to_label


def plot_action():
    def xlim_type(value):
        try:
            newvalue = [float(x) for x in value.split(',')]
        except:
            raise ArgumentError("'%s' option must contain two numbers separated with a comma" % value)
        if len(newvalue) != 2:
            raise ArgumentError("'%s' option must contain two numbers separated with a comma" % value)
        return newvalue

    desc = """Plots using matplotlib the data file `file.dat` generated
using `mprof run`. If no .dat file is given, it will take the most recent
such file in the current directory."""
    parser = ArgumentParser(usage="mprof plot [options] [file.dat]", description=desc)
    parser.add_argument('--version', action='version', version=mp.__version__)
    parser.add_argument("--title", "-t", dest="title", default=None,
                        type=str, action="store",
                        help="String shown as plot title")
    parser.add_argument("--no-function-ts", "-n", dest="no_timestamps", action="store_true",
                        help="Do not display function timestamps on plot.")
    parser.add_argument("--output", "-o",
                        help="Save plot to file instead of displaying it.")
    parser.add_argument("--window", "-w", dest="xlim", type=xlim_type,
                        help="Plot a time-subset of the data. E.g. to plot between 0 and 20.5 seconds: --window 0,20.5")
    parser.add_argument("--flame", "-f", dest="flame_mode", action="store_true",
                        help="Plot the timestamps as a flame-graph instead of the default brackets")
    parser.add_argument("--slope", "-s", dest="slope", action="store_true",
                        help="Plot a trend line and its numerical slope")
    parser.add_argument("--backend",
                      help="Specify the Matplotlib backend to use")
    parser.add_argument("profiles", nargs="*",
                        help="profiles made by mprof run")
    args = parser.parse_args()

    try:
        if args.backend is not None:
            import matplotlib
            matplotlib.use(args.backend)

        import pylab as pl
    except ImportError as e:
        print("matplotlib is needed for plotting.")
        print(e)
        sys.exit(1)
    pl.ioff()

    filenames = get_profiles(args)

    fig = pl.figure(figsize=(14, 6), dpi=90)
    if not args.flame_mode:
        ax = fig.add_axes([0.1, 0.1, 0.6, 0.75])
    else:
        ax = fig.add_axes([0.1, 0.1, 0.8, 0.8])
    if args.xlim is not None:
        pl.xlim(args.xlim[0], args.xlim[1])

    if len(filenames) > 1 or args.no_timestamps:
        timestamps = False
    else:
        timestamps = True
    plotter = plot_file
    if args.flame_mode:
        plotter = flame_plotter
    for n, filename in enumerate(filenames):
        mprofile = plotter(filename, index=n, timestamps=timestamps, options=args)
    pl.xlabel("time (in seconds)")
    pl.ylabel("memory used (in MiB)")

    if args.title is None and len(filenames) == 1:
        pl.title(mprofile['cmd_line'])
    else:
        if args.title is not None:
            pl.title(args.title)

    # place legend within the plot, make partially transparent in
    # case it obscures part of the lineplot
    if not args.flame_mode:
        leg = ax.legend(loc='center left', bbox_to_anchor=(1, 0.5))
        leg.get_frame().set_alpha(0.5)
        pl.grid()

    if args.output:
        pl.savefig(args.output)
    else:
        pl.show()

def filter_mprofile_mem_usage_by_function(prof, func):
    if func is None:
        return prof["mem_usage"]

    if func not in prof["func_timestamp"]:
        raise ValueError(str(func) + " was not found.")

    time_ranges = prof["func_timestamp"][func]
    filtered_memory = []
    
    # The check here could be improved, but it's done in this
    # inefficient way to make sure we don't miss overlapping
    # ranges.
    for mib, ts in zip(prof["mem_usage"], prof["timestamp"]):
        for rng in time_ranges:
            if rng[0] <= ts <= rng[1]:
                filtered_memory.append(mib)

    return filtered_memory

def peak_action():
    desc = """Prints the peak memory used in data file `file.dat` generated
using `mprof run`. If no .dat file is given, it will take the most recent
such file in the current directory."""
    parser = ArgumentParser(usage="mprof peak [options] [file.dat]", description=desc)
    parser.add_argument("profiles", nargs="*",
                    help="profiles made by mprof run")
    parser.add_argument("--func", dest="func", default=None,
                        help="""Show the peak for this function. Does not support child processes.""")
    args = parser.parse_args()
    filenames = get_profiles(args)

    for filename in filenames:
        prof = read_mprofile_file(filename)
        try:
            mem_usage = filter_mprofile_mem_usage_by_function(prof, args.func)
        except ValueError:
            print("{}\tNaN MiB".format(prof["filename"]))
            continue

        print("{}\t{:.3f} MiB".format(prof["filename"], max(mem_usage)))
        for child, values in prof["children"].items():
            child_peak = max([ mem_ts[0] for mem_ts in values ])
            print("  Child {}\t\t\t{:.3f} MiB".format(child, child_peak))
        

def get_profiles(args):
    profiles = glob.glob("mprofile_??????????????.dat")
    profiles.sort()

    if len(args.profiles) == 0:
        if len(profiles) == 0:
            print("No input file found. \nThis program looks for "
                  "mprofile_*.dat files, generated by the "
                  "'mprof run' command.")
            sys.exit(-1)
        print("Using last profile data.")
        filenames = [profiles[-1]]
    else:
        filenames = []
        for prof in args.profiles:
            if osp.exists(prof):
                if not prof in filenames:
                    filenames.append(prof)
            else:
                try:
                    n = int(prof)
                    if not profiles[n] in filenames:
                        filenames.append(profiles[n])
                except ValueError:
                    print("Input file not found: " + prof)
    if not len(filenames):
        print("No files found from given input.")
        sys.exit(-1)

    return filenames

def main():
    # Workaround for optparse limitation: insert -- before first negative
    # number found.
    negint = re.compile("-[0-9]+")
    for n, arg in enumerate(sys.argv):
        if negint.match(arg):
            sys.argv.insert(n, "--")
            break
    actions = {"rm": rm_action,
               "clean": clean_action,
               "list": list_action,
               "run": run_action,
               "attach": attach_action,
               "plot": plot_action,
               "peak": peak_action}
    actions[get_action()]()

if __name__ == "__main__":
    main()