Diff two runs of nm on similar executables to find large symbol size increases

diff_big_sizes.py

#! /usr/bin/env python3

"""
Build to compare two outputs of this line on fairly identical executables
  nm --print-size --radix=d <executable> | grep ' .* .* ' | cut -d' ' -f 2-10

To just see the top symbols, can run e.g
  nm --print-size --size-sort --radix=d <executable>
"""

from collections import Counter
import re
import subprocess
import sys
import textwrap
from typing import NamedTuple


class Diff(NamedTuple):
    old_max: int
    new_max: int
    section: str

    @property
    def size(self):
        return self.new_max - self.old_max


def get_diffs():
    diff = subprocess.run(
        ["diff", "-u0"] + sys.argv[1:], capture_output=True
    ).stdout.decode("utf-8")

    # strip filenames
    diff = diff[diff.index("@") :]

    diff_sections = re.split(r"^@@.*@@$", diff, flags=re.MULTILINE)

    def sizes_histogram(section, prefix):
        return Counter(
            int(line.split()[0].strip("-+"))
            for line in section.splitlines()
            if line.startswith(prefix)
        )

    for section in diff_sections:
        section = section.strip()
        pluses = sizes_histogram(section, prefix="+")
        minuses = sizes_histogram(section, prefix="-")

        if pluses == minuses:
            # no change
            continue

        if not pluses or not minuses:
            # not a diff
            continue

        # some sizes changed
        yield Diff(section=section, old_max=max(minuses), new_max=max(pluses))


def main():
    MB = 1048576
    big_diffs = sorted(
        (x for x in get_diffs() if x.size > MB), key=lambda x: x.size, reverse=True
    )

    for diff in big_diffs[:20]:
        print(
            f"""
=====
{diff.size/1048576:.1f} MB increase | {100*diff.new_max/diff.old_max - 100:.1f}% | {diff.old_max} => {diff.new_max}
-----
{diff.section}
""".strip()
        )


if __name__ == "__main__":
    main()