|
| 1 | +#!/usr/bin/env python3 |
| 2 | + |
| 3 | +import argparse |
| 4 | +import functools |
| 5 | +import os |
| 6 | +import pathlib |
| 7 | +import re |
| 8 | +import statistics |
| 9 | +import subprocess |
| 10 | +import sys |
| 11 | +import tempfile |
| 12 | + |
| 13 | +import plotly |
| 14 | +import tqdm |
| 15 | + |
| 16 | +@functools.total_ordering |
| 17 | +class Commit: |
| 18 | + """ |
| 19 | + This class represents a commit inside a given Git repository. |
| 20 | + """ |
| 21 | + |
| 22 | + def __init__(self, git_repo, sha): |
| 23 | + self._git_repo = git_repo |
| 24 | + self._sha = sha |
| 25 | + |
| 26 | + def __eq__(self, other): |
| 27 | + """ |
| 28 | + Return whether two commits refer to the same commit. |
| 29 | +
|
| 30 | + This doesn't take into account the content of the Git tree at those commits, only the |
| 31 | + 'identity' of the commits themselves. |
| 32 | + """ |
| 33 | + return self.fullrev == other.fullrev |
| 34 | + |
| 35 | + def __lt__(self, other): |
| 36 | + """ |
| 37 | + Return whether a commit is an ancestor of another commit in the Git repository. |
| 38 | + """ |
| 39 | + # Is self._sha an ancestor of other._sha? |
| 40 | + res = subprocess.run(['git', '-C', self._git_repo, 'merge-base', '--is-ancestor', self._sha, other._sha]) |
| 41 | + if res.returncode not in (0, 1): |
| 42 | + raise RuntimeError(f'Error when trying to obtain the commit order for {self._sha} and {other._sha}') |
| 43 | + return res.returncode == 0 |
| 44 | + |
| 45 | + def show(self, include_diff=False): |
| 46 | + """ |
| 47 | + Return the commit information equivalent to `git show` associated to this commit. |
| 48 | + """ |
| 49 | + cmd = ['git', '-C', self._git_repo, 'show', self._sha] |
| 50 | + if not include_diff: |
| 51 | + cmd.append('--no-patch') |
| 52 | + return subprocess.check_output(cmd, text=True) |
| 53 | + |
| 54 | + @functools.cached_property |
| 55 | + def shortrev(self): |
| 56 | + """ |
| 57 | + Return the shortened version of the given SHA. |
| 58 | + """ |
| 59 | + return subprocess.check_output(['git', '-C', self._git_repo, 'rev-parse', '--short', self._sha], text=True).strip() |
| 60 | + |
| 61 | + @functools.cached_property |
| 62 | + def fullrev(self): |
| 63 | + """ |
| 64 | + Return the full SHA associated to this commit. |
| 65 | + """ |
| 66 | + return subprocess.check_output(['git', '-C', self._git_repo, 'rev-parse', self._sha], text=True).strip() |
| 67 | + |
| 68 | + def prefetch(self): |
| 69 | + """ |
| 70 | + Prefetch cached properties associated to this commit object. |
| 71 | +
|
| 72 | + This makes it possible to control when time is spent recovering that information from Git for |
| 73 | + e.g. better reporting to the user. |
| 74 | + """ |
| 75 | + self.shortrev |
| 76 | + self.fullrev |
| 77 | + |
| 78 | + def __str__(self): |
| 79 | + return self._sha |
| 80 | + |
| 81 | +def truncate_lines(string, n, marker=None): |
| 82 | + """ |
| 83 | + Truncate the given string at a certain number of lines. |
| 84 | +
|
| 85 | + Optionally, add a marker on the last line to identify that truncation has happened. |
| 86 | + """ |
| 87 | + lines = string.splitlines() |
| 88 | + truncated = lines[:n] |
| 89 | + if marker is not None and len(lines) > len(truncated): |
| 90 | + truncated[-1] = marker |
| 91 | + assert len(truncated) <= n, "broken post-condition" |
| 92 | + return '\n'.join(truncated) |
| 93 | + |
| 94 | +def create_plot(commits, benchmarks, data): |
| 95 | + """ |
| 96 | + Create a plot object showing the evolution of each benchmark throughout the given commits. |
| 97 | + """ |
| 98 | + figure = plotly.graph_objects.Figure(layout_title_text=f'{commits[0].shortrev} to {commits[-1].shortrev}') |
| 99 | + |
| 100 | + # Create the X axis and the hover information |
| 101 | + x_axis = [commit.shortrev for commit in commits] |
| 102 | + hover_info = [truncate_lines(commit.show(), 30, marker='...').replace('\n', '<br>') for commit in commits] |
| 103 | + |
| 104 | + # For each benchmark, get the metric for that benchmark for each commit. |
| 105 | + # |
| 106 | + # Some commits may not have any data associated to a benchmark (e.g. runtime or compilation error). |
| 107 | + # Use None, which is handled properly by plotly. |
| 108 | + for benchmark in benchmarks: |
| 109 | + series = [commit_data.get(benchmark, None) for commit_data in data] |
| 110 | + scatter = plotly.graph_objects.Scatter(x=x_axis, y=series, text=hover_info, name=benchmark) |
| 111 | + figure.add_trace(scatter) |
| 112 | + |
| 113 | + return figure |
| 114 | + |
| 115 | +def directory_path(string): |
| 116 | + if os.path.isdir(string): |
| 117 | + return pathlib.Path(string) |
| 118 | + else: |
| 119 | + raise NotADirectoryError(string) |
| 120 | + |
| 121 | +def parse_lnt(lines): |
| 122 | + """ |
| 123 | + Parse lines in LNT format and return a dictionnary of the form: |
| 124 | +
|
| 125 | + { |
| 126 | + 'benchmark1': { |
| 127 | + 'metric1': [float], |
| 128 | + 'metric2': [float], |
| 129 | + ... |
| 130 | + }, |
| 131 | + 'benchmark2': { |
| 132 | + 'metric1': [float], |
| 133 | + 'metric2': [float], |
| 134 | + ... |
| 135 | + }, |
| 136 | + ... |
| 137 | + } |
| 138 | +
|
| 139 | + Each metric may have multiple values. |
| 140 | + """ |
| 141 | + results = {} |
| 142 | + for line in lines: |
| 143 | + line = line.strip() |
| 144 | + if not line: |
| 145 | + continue |
| 146 | + |
| 147 | + (identifier, value) = line.split(' ') |
| 148 | + (name, metric) = identifier.split('.') |
| 149 | + if name not in results: |
| 150 | + results[name] = {} |
| 151 | + if metric not in results[name]: |
| 152 | + results[name][metric] = [] |
| 153 | + results[name][metric].append(float(value)) |
| 154 | + return results |
| 155 | + |
| 156 | +def main(argv): |
| 157 | + parser = argparse.ArgumentParser( |
| 158 | + prog='visualize-historical', |
| 159 | + description='Visualize historical data in LNT format. This program generates a HTML file that embeds an ' |
| 160 | + 'interactive plot with the provided data. The HTML file can then be opened in a browser to ' |
| 161 | + 'visualize the data as a chart.', |
| 162 | + epilog='This script depends on the `plotly` and the `tqdm` Python modules.') |
| 163 | + parser.add_argument('directory', type=directory_path, |
| 164 | + help='Path to a valid directory containing benchmark data in LNT format, each file being named <commit>.lnt. ' |
| 165 | + 'This is also the format generated by the `benchmark-historical` utility.') |
| 166 | + parser.add_argument('--output', '-o', type=pathlib.Path, required=False, |
| 167 | + help='Optional path where to output the resulting HTML file. If it already exists, it is overwritten. ' |
| 168 | + 'Defaults to a temporary file which is opened automatically once generated, but not removed after ' |
| 169 | + 'creation.') |
| 170 | + parser.add_argument('--metric', type=str, default='execution_time', |
| 171 | + help='The metric to compare. LNT data may contain multiple metrics (e.g. code size, execution time, etc) -- ' |
| 172 | + 'this option allows selecting which metric is being visualized. The default is "execution_time".') |
| 173 | + parser.add_argument('--filter', type=str, required=False, |
| 174 | + help='An optional regular expression used to filter the benchmarks included in the chart. ' |
| 175 | + 'Only benchmarks whose names match the regular expression will be included. ' |
| 176 | + 'Since the chart is interactive, it generally makes most sense to include all the benchmarks ' |
| 177 | + 'and to then filter them in the browser, but in some cases producing a chart with a reduced ' |
| 178 | + 'number of data series is useful.') |
| 179 | + parser.add_argument('--git-repo', type=directory_path, default=pathlib.Path(os.getcwd()), |
| 180 | + help='Path to the git repository to use for ordering commits in time. ' |
| 181 | + 'By default, the current working directory is used.') |
| 182 | + parser.add_argument('--open', action='store_true', |
| 183 | + help='Whether to automatically open the generated HTML file when finished. If no output file is provided, ' |
| 184 | + 'the resulting benchmark is opened automatically by default.') |
| 185 | + args = parser.parse_args(argv) |
| 186 | + |
| 187 | + # Extract benchmark data from the directory and keep only the metric we're interested in. |
| 188 | + # |
| 189 | + # Some data points may have multiple values associated to the metric (e.g. if we performed |
| 190 | + # multiple runs to reduce noise), in which case we aggregate them using a median. |
| 191 | + historical_data = [] |
| 192 | + files = [f for f in args.directory.glob('*.lnt')] |
| 193 | + for file in tqdm.tqdm(files, desc='Parsing LNT files'): |
| 194 | + (commit, _) = os.path.splitext(os.path.basename(file)) |
| 195 | + commit = Commit(args.git_repo, commit) |
| 196 | + with open(file, 'r') as f: |
| 197 | + lnt_data = parse_lnt(f.readlines()) |
| 198 | + commit_data = {} |
| 199 | + for (bm, metrics) in lnt_data.items(): |
| 200 | + commit_data[bm] = statistics.median(metrics[args.metric]) if args.metric in metrics else None |
| 201 | + historical_data.append((commit, commit_data)) |
| 202 | + |
| 203 | + # Obtain commit information which is then cached throughout the program. Do this |
| 204 | + # eagerly so we can provide a progress bar. |
| 205 | + for (commit, _) in tqdm.tqdm(historical_data, desc='Prefetching Git information'): |
| 206 | + commit.prefetch() |
| 207 | + |
| 208 | + # Sort the data based on the ordering of commits inside the provided Git repository |
| 209 | + historical_data.sort(key=lambda x: x[0]) |
| 210 | + |
| 211 | + # Filter the benchmarks if needed |
| 212 | + benchmarks = {b for (_, commit_data) in historical_data for b in commit_data.keys()} |
| 213 | + if args.filter is not None: |
| 214 | + regex = re.compile(args.filter) |
| 215 | + benchmarks = {b for b in benchmarks if regex.search(b)} |
| 216 | + |
| 217 | + # Plot the data for all the required benchmarks |
| 218 | + figure = create_plot([commit for (commit, _) in historical_data], |
| 219 | + sorted(list(benchmarks)), |
| 220 | + [data for (_, data) in historical_data]) |
| 221 | + do_open = args.output is None or args.open |
| 222 | + output = args.output if args.output is not None else tempfile.NamedTemporaryFile(suffix='.html').name |
| 223 | + plotly.io.write_html(figure, file=output, auto_open=do_open) |
| 224 | + |
| 225 | +if __name__ == '__main__': |
| 226 | + main(sys.argv[1:]) |
0 commit comments