#!/usr/bin/env python3

from argparse import ArgumentParser, FileType

import numpy

import pandas

def main():

# process command-line options

parser = ArgumentParser()

parser.add_argument(

"--csv",

help="CSV file with raw results from time trials (default: %(default)s)",

type=FileType(),

default="build/time-trials.csv",

)

raw_results_csv = parser.parse_args().csv

# load results, then identify successful and skipped tests

frame = pandas.read_csv(raw_results_csv)

results = frame["resultType"]

successes = results == "SUCCESS"

skips = results == "SKIPPED"

# report failed tests, if any

failures = frame[~(successes | skips)]

if not failures.empty:

print("failed tests:")

print(failures)

return

# aggregate multiple trials of each individual test method

frame["elapsedTime"] = frame["endTime"] - frame["startTime"]

grouped = frame[successes].groupby(["className", "name"])["elapsedTime"]

# print very wide tables in full; assume user can scroll horizontally

pandas.set_option("display.width", None)

pandas.set_option("display.max_colwidth", None)

# summarize distribution of elapsed times, including fine-grained

# percentiles at the upper (slowest) end

times = grouped.mean()

times.sort_values(inplace=True)

print("Overall distribution and percentiles of elapsed times:\n")

print(

times.describe(

percentiles=numpy.arange(0.8, 1, 0.01),

)

)

print("\n")

# print slowest individual tests, showing only those in the 95% percentile

# or higher (slower)

ranked_times = pandas.DataFrame(times)

ranked_times["percentRank"] = times.rank(pct=True)

elapsed_times = ranked_times["elapsedTime"]

ranked_times["percentOfTotal"] = elapsed_times / elapsed_times.sum()

is_slow = ranked_times["percentRank"] >= 0.95

slowest_tests = ranked_times[is_slow][::-1]

print("Slowest individual tests:")

format_as_percent = "{:.1%}".format

print(

slowest_tests.to_string(

float_format=format_as_percent,

formatters={"elapsedTime": "{:,.0f}".format},

)

)

if __name__ == "__main__":

main()