littlefs/scripts/code.py

#!/usr/bin/env python3
#
# Script to find code size at the function level. Basically just a big wrapper
# around nm with some extra conveniences for comparing builds. Heavily inspired
# by Linux's Bloat-O-Meter.
#
# Example:
# ./scripts/code.py lfs.o lfs_util.o -Ssize
#
# Copyright (c) 2022, The littlefs authors.
# Copyright (c) 2020, Arm Limited. All rights reserved.
# SPDX-License-Identifier: BSD-3-Clause
#

# prevent local imports
__import__('sys').path.pop(0)

import collections as co
import csv
import difflib
import itertools as it
import math as mt
import os
import re
import shlex
import subprocess as sp


NM_PATH = ['nm']
NM_TYPES = 'tTrRdD'
OBJDUMP_PATH = ['objdump']


# integer fields
class RInt(co.namedtuple('RInt', 'x')):
    __slots__ = ()
    def __new__(cls, x=0):
        if isinstance(x, RInt):
            return x
        if isinstance(x, str):
            try:
                x = int(x, 0)
            except ValueError:
                # also accept +-∞ and +-inf
                if re.match('^\s*\+?\s*(?:∞|inf)\s*$', x):
                    x = mt.inf
                elif re.match('^\s*-\s*(?:∞|inf)\s*$', x):
                    x = -mt.inf
                else:
                    raise
        assert isinstance(x, int) or mt.isinf(x), x
        return super().__new__(cls, x)

    def __str__(self):
        if self.x == mt.inf:
            return '∞'
        elif self.x == -mt.inf:
            return '-∞'
        else:
            return str(self.x)

    def __int__(self):
        assert not mt.isinf(self.x)
        return self.x

    def __float__(self):
        return float(self.x)

    none = '%7s' % '-'
    def table(self):
        return '%7s' % (self,)

    def diff(self, other):
        new = self.x if self else 0
        old = other.x if other else 0
        diff = new - old
        if diff == +mt.inf:
            return '%7s' % '+∞'
        elif diff == -mt.inf:
            return '%7s' % '-∞'
        else:
            return '%+7d' % diff

    def ratio(self, other):
        new = self.x if self else 0
        old = other.x if other else 0
        if mt.isinf(new) and mt.isinf(old):
            return 0.0
        elif mt.isinf(new):
            return +mt.inf
        elif mt.isinf(old):
            return -mt.inf
        elif not old and not new:
            return 0.0
        elif not old:
            return +mt.inf
        else:
            return (new-old) / old

    def __add__(self, other):
        return self.__class__(self.x + other.x)

    def __sub__(self, other):
        return self.__class__(self.x - other.x)

    def __mul__(self, other):
        return self.__class__(self.x * other.x)

# code size results
class CodeResult(co.namedtuple('CodeResult', [
        'file', 'function',
        'size'])):
    _by = ['file', 'function']
    _fields = ['size']
    _sort = ['size']
    _types = {'size': RInt}

    __slots__ = ()
    def __new__(cls, file='', function='', size=0):
        return super().__new__(cls, file, function,
                RInt(size))

    def __add__(self, other):
        return CodeResult(self.file, self.function,
                self.size + other.size)


def openio(path, mode='r', buffering=-1):
    # allow '-' for stdin/stdout
    if path == '-':
        if 'r' in mode:
            return os.fdopen(os.dup(sys.stdin.fileno()), mode, buffering)
        else:
            return os.fdopen(os.dup(sys.stdout.fileno()), mode, buffering)
    else:
        return open(path, mode, buffering)

def collect(obj_paths, *,
        nm_path=NM_PATH,
        nm_types=NM_TYPES,
        objdump_path=OBJDUMP_PATH,
        sources=None,
        everything=False,
        **args):
    size_pattern = re.compile(
            '^(?P<size>[0-9a-fA-F]+)'
                + ' (?P<type>[%s])' % re.escape(nm_types)
                + ' (?P<func>.+?)$')
    line_pattern = re.compile(
            '^\s+(?P<no>[0-9]+)'
                '(?:\s+(?P<dir>[0-9]+))?'
                '\s+.*'
                '\s+(?P<path>[^\s]+)$')
    info_pattern = re.compile(
            '^(?:.*(?P<tag>DW_TAG_[a-z_]+).*'
                '|.*DW_AT_name.*:\s*(?P<name>[^:\s]+)\s*'
                '|.*DW_AT_decl_file.*:\s*(?P<file>[0-9]+)\s*)$')

    results = []
    for path in obj_paths:
        # guess the source, if we have debug-info we'll replace this later
        file = re.sub('(\.o)?$', '.c', path, 1)

        # find symbol sizes
        results_ = []
        # note nm-path may contain extra args
        cmd = nm_path + ['--size-sort', path]
        if args.get('verbose'):
            print(' '.join(shlex.quote(c) for c in cmd))
        proc = sp.Popen(cmd,
                stdout=sp.PIPE,
                stderr=None if args.get('verbose') else sp.DEVNULL,
                universal_newlines=True,
                errors='replace',
                close_fds=False)
        for line in proc.stdout:
            m = size_pattern.match(line)
            if m:
                func = m.group('func')
                # discard internal functions
                if not everything and func.startswith('__'):
                    continue
                results_.append(CodeResult(
                        file, func,
                        int(m.group('size'), 16)))
        proc.wait()
        if proc.returncode != 0:
            if not args.get('verbose'):
                for line in proc.stderr:
                    sys.stderr.write(line)
            sys.exit(-1)


        # try to figure out the source file if we have debug-info
        dirs = {}
        files = {}
        # note objdump-path may contain extra args
        cmd = objdump_path + ['--dwarf=rawline', path]
        if args.get('verbose'):
            print(' '.join(shlex.quote(c) for c in cmd))
        proc = sp.Popen(cmd,
                stdout=sp.PIPE,
                stderr=None if args.get('verbose') else sp.DEVNULL,
                universal_newlines=True,
                errors='replace',
                close_fds=False)
        for line in proc.stdout:
            # note that files contain references to dirs, which we
            # dereference as soon as we see them as each file table follows a
            # dir table
            m = line_pattern.match(line)
            if m:
                if not m.group('dir'):
                    # found a directory entry
                    dirs[int(m.group('no'))] = m.group('path')
                else:
                    # found a file entry
                    dir = int(m.group('dir'))
                    if dir in dirs:
                        files[int(m.group('no'))] = os.path.join(
                                dirs[dir],
                                m.group('path'))
                    else:
                        files[int(m.group('no'))] = m.group('path')
        proc.wait()
        if proc.returncode != 0:
            if not args.get('verbose'):
                for line in proc.stderr:
                    sys.stderr.write(line)
            # do nothing on error, we don't need objdump to work, source files
            # may just be inaccurate
            pass

        defs = {}
        is_func = False
        f_name = None
        f_file = None
        def append():
            # ignore non-functions and unnamed files
            if is_func and f_name:
                defs[f_name] = files.get(f_file, '?')
        # note objdump-path may contain extra args
        cmd = objdump_path + ['--dwarf=info', path]
        if args.get('verbose'):
            print(' '.join(shlex.quote(c) for c in cmd))
        proc = sp.Popen(cmd,
                stdout=sp.PIPE,
                stderr=None if args.get('verbose') else sp.DEVNULL,
                universal_newlines=True,
                errors='replace',
                close_fds=False)
        for line in proc.stdout:
            # state machine here to find definitions
            m = info_pattern.match(line)
            if m:
                if m.group('tag'):
                    append()
                    is_func = (m.group('tag') == 'DW_TAG_subprogram')
                    f_name = None
                    f_file = None
                elif m.group('name'):
                    f_name = m.group('name')
                elif m.group('file'):
                    f_file = int(m.group('file'))
        # don't forget the last function
        append()
        proc.wait()
        if proc.returncode != 0:
            if not args.get('verbose'):
                for line in proc.stderr:
                    sys.stderr.write(line)
            # do nothing on error, we don't need objdump to work, source files
            # may just be inaccurate
            pass

        for r in results_:
            # find best matching debug symbol, this may be slightly different
            # due to optimizations
            if defs:
                # exact match? avoid difflib if we can for speed
                if r.function in defs:
                    file = defs[r.function]
                else:
                    _, file = max(
                            defs.items(),
                            key=lambda d: difflib.SequenceMatcher(None,
                                d[0],
                                r.function, False).ratio())
            else:
                file = r.file

            # ignore filtered sources
            if sources is not None:
                if not any(os.path.abspath(file) == os.path.abspath(s)
                        for s in sources):
                    continue
            else:
                # default to only cwd
                if not everything and not os.path.commonpath([
                        os.getcwd(),
                        os.path.abspath(file)]) == os.getcwd():
                    continue

            # simplify path
            if os.path.commonpath([
                    os.getcwd(),
                    os.path.abspath(file)]) == os.getcwd():
                file = os.path.relpath(file)
            else:
                file = os.path.abspath(file)

            results.append(r._replace(file=file))

    return results


def fold(Result, results, by=None, defines=[]):
    if by is None:
        by = Result._by

    for k in it.chain(by or [], (k for k, _ in defines)):
        if k not in Result._by and k not in Result._fields:
            print("error: could not find field %r?" % k,
                    file=sys.stderr)
            sys.exit(-1)

    # filter by matching defines
    if defines:
        results_ = []
        for r in results:
            if all(getattr(r, k) in vs for k, vs in defines):
                results_.append(r)
        results = results_

    # organize results into conflicts
    folding = co.OrderedDict()
    for r in results:
        name = tuple(getattr(r, k) for k in by)
        if name not in folding:
            folding[name] = []
        folding[name].append(r)

    # merge conflicts
    folded = []
    for name, rs in folding.items():
        folded.append(sum(rs[1:], start=rs[0]))

    return folded

def table(Result, results, diff_results=None, *,
        by=None,
        fields=None,
        sort=None,
        summary=False,
        all=False,
        percent=False,
        **_):
    all_, all = all, __builtins__.all

    if by is None:
        by = Result._by
    if fields is None:
        fields = Result._fields
    types = Result._types

    # fold again
    results = fold(Result, results, by=by)
    if diff_results is not None:
        diff_results = fold(Result, diff_results, by=by)

    # organize by name
    table = {
            ','.join(str(getattr(r, k) or '') for k in by): r
                for r in results}
    diff_table = {
            ','.join(str(getattr(r, k) or '') for k in by): r
                for r in diff_results or []}
    names = [name
            for name in table.keys() | diff_table.keys()
            if diff_results is None
                or all_
                or any(
                    types[k].ratio(
                            getattr(table.get(name), k, None),
                            getattr(diff_table.get(name), k, None))
                        for k in fields)]

    # sort again, now with diff info, note that python's sort is stable
    names.sort()
    if diff_results is not None:
        names.sort(
                key=lambda n: tuple(
                    types[k].ratio(
                            getattr(table.get(n), k, None),
                            getattr(diff_table.get(n), k, None))
                        for k in fields),
                reverse=True)
    if sort:
        for k, reverse in reversed(sort):
            names.sort(
                    key=lambda n: tuple(
                        (getattr(table[n], k),)
                                if getattr(table.get(n), k, None) is not None
                                else ()
                            for k in (
                                [k] if k else [
                                    k for k in Result._sort
                                        if k in fields])),
                    reverse=reverse ^ (not k or k in Result._fields))


    # build up our lines
    lines = []

    # header
    header = ['%s%s' % (
                ','.join(by),
                ' (%d added, %d removed)' % (
                        sum(1 for n in table if n not in diff_table),
                        sum(1 for n in diff_table if n not in table))
                    if diff_results is not None and not percent else '')
            if not summary else '']
    if diff_results is None:
        for k in fields:
            header.append(k)
    elif percent:
        for k in fields:
            header.append(k)
    else:
        for k in fields:
            header.append('o'+k)
        for k in fields:
            header.append('n'+k)
        for k in fields:
            header.append('d'+k)
    lines.append(header)

    # entry helper
    def table_entry(name, r, diff_r=None):
        entry = [name]
        if diff_results is None:
            for k in fields:
                entry.append(
                        (getattr(r, k).table(),
                                getattr(getattr(r, k), 'notes', lambda: [])())
                            if getattr(r, k, None) is not None
                            else types[k].none)
        elif percent:
            for k in fields:
                entry.append(
                        (getattr(r, k).table()
                                if getattr(r, k, None) is not None
                                else types[k].none,
                            (lambda t: ['+∞%'] if t == +mt.inf
                                    else ['-∞%'] if t == -mt.inf
                                    else ['%+.1f%%' % (100*t)])(
                                types[k].ratio(
                                    getattr(r, k, None),
                                    getattr(diff_r, k, None)))))
        else:
            for k in fields:
                entry.append(getattr(diff_r, k).table()
                        if getattr(diff_r, k, None) is not None
                        else types[k].none)
            for k in fields:
                entry.append(getattr(r, k).table()
                        if getattr(r, k, None) is not None
                        else types[k].none)
            for k in fields:
                entry.append(
                        (types[k].diff(
                                getattr(r, k, None),
                                getattr(diff_r, k, None)),
                            (lambda t: ['+∞%'] if t == +mt.inf
                                    else ['-∞%'] if t == -mt.inf
                                    else ['%+.1f%%' % (100*t)] if t
                                    else [])(
                                types[k].ratio(
                                    getattr(r, k, None),
                                    getattr(diff_r, k, None)))))
        return entry

    # entries
    if not summary:
        for name in names:
            r = table.get(name)
            if diff_results is None:
                diff_r = None
            else:
                diff_r = diff_table.get(name)
            lines.append(table_entry(name, r, diff_r))

    # total
    r = next(iter(fold(Result, results, by=[])), None)
    if diff_results is None:
        diff_r = None
    else:
        diff_r = next(iter(fold(Result, diff_results, by=[])), None)
    lines.append(table_entry('TOTAL', r, diff_r))

    # homogenize
    lines = [
            [x if isinstance(x, tuple) else (x, []) for x in line]
                for line in lines]

    # find the best widths, note that column 0 contains the names and is
    # handled a bit differently
    widths = co.defaultdict(lambda: 7, {0: 23})
    notes = co.defaultdict(lambda: 0)
    for line in lines:
        for i, x in enumerate(line):
            widths[i] = max(widths[i], ((len(x[0])+1+4-1)//4)*4-1)
            notes[i] = max(notes[i], 1+2*len(x[1])+sum(len(n) for n in x[1]))

    # print our table
    for line in lines:
        print('%-*s  %s' % (
                widths[0], line[0][0],
                ' '.join('%*s%-*s' % (
                        widths[i], x[0],
                        notes[i], ' (%s)' % ', '.join(x[1]) if x[1] else '')
                    for i, x in enumerate(line[1:], 1))))


def main(obj_paths, *,
        by=None,
        fields=None,
        defines=[],
        sort=None,
        **args):
    # find sizes
    if not args.get('use', None):
        results = collect(obj_paths, **args)
    else:
        results = []
        with openio(args['use']) as f:
            reader = csv.DictReader(f, restval='')
            for r in reader:
                # filter by matching defines
                if not all(k in r and r[k] in vs for k, vs in defines):
                    continue

                if not any(k in r and r[k].strip()
                        for k in CodeResult._fields):
                    continue
                try:
                    results.append(CodeResult(
                            **{k: r[k] for k in CodeResult._by
                                if k in r and r[k].strip()},
                            **{k: r[k] for k in CodeResult._fields
                                if k in r and r[k].strip()}))
                except TypeError:
                    pass

    # fold
    results = fold(CodeResult, results, by=by, defines=defines)

    # sort, note that python's sort is stable
    results.sort()
    if sort:
        for k, reverse in reversed(sort):
            results.sort(
                    key=lambda r: tuple(
                        (getattr(r, k),) if getattr(r, k) is not None else ()
                            for k in ([k] if k else CodeResult._sort)),
                    reverse=reverse ^ (not k or k in CodeResult._fields))

    # write results to CSV
    if args.get('output'):
        with openio(args['output'], 'w') as f:
            writer = csv.DictWriter(f,
                    (by if by is not None else CodeResult._by)
                        + [k for k in (
                            fields if fields is not None
                                else CodeResult._fields)])
            writer.writeheader()
            for r in results:
                writer.writerow(
                        {k: getattr(r, k) for k in (
                                by if by is not None else CodeResult._by)}
                            | {k: getattr(r, k) for k in (
                                fields if fields is not None
                                    else CodeResult._fields)})

    # find previous results?
    if args.get('diff'):
        diff_results = []
        try:
            with openio(args['diff']) as f:
                reader = csv.DictReader(f, restval='')
                for r in reader:
                    # filter by matching defines
                    if not all(k in r and r[k] in vs for k, vs in defines):
                        continue

                    if not any(k in r and r[k].strip()
                            for k in CodeResult._fields):
                        continue
                    try:
                        diff_results.append(CodeResult(
                                **{k: r[k] for k in CodeResult._by
                                    if k in r and r[k].strip()},
                                **{k: r[k] for k in CodeResult._fields
                                    if k in r and r[k].strip()}))
                    except TypeError:
                        pass
        except FileNotFoundError:
            pass

        # fold
        diff_results = fold(CodeResult, diff_results, by=by, defines=defines)

    # print table
    if not args.get('quiet'):
        table(CodeResult, results,
                diff_results if args.get('diff') else None,
                by=by if by is not None else ['function'],
                fields=fields,
                sort=sort,
                **args)


if __name__ == "__main__":
    import argparse
    import sys
    parser = argparse.ArgumentParser(
            description="Find code size at the function level.",
            allow_abbrev=False)
    parser.add_argument(
            'obj_paths',
            nargs='*',
            help="Input *.o files.")
    parser.add_argument(
            '-v', '--verbose',
            action='store_true',
            help="Output commands that run behind the scenes.")
    parser.add_argument(
            '-q', '--quiet',
            action='store_true',
            help="Don't show anything, useful with -o.")
    parser.add_argument(
            '-o', '--output',
            help="Specify CSV file to store results.")
    parser.add_argument(
            '-u', '--use',
            help="Don't parse anything, use this CSV file.")
    parser.add_argument(
            '-d', '--diff',
            help="Specify CSV file to diff against.")
    parser.add_argument(
            '-a', '--all',
            action='store_true',
            help="Show all, not just the ones that changed.")
    parser.add_argument(
            '-p', '--percent',
            action='store_true',
            help="Only show percentage change, not a full diff.")
    parser.add_argument(
            '-b', '--by',
            action='append',
            choices=CodeResult._by,
            help="Group by this field.")
    parser.add_argument(
            '-f', '--field',
            dest='fields',
            action='append',
            choices=CodeResult._fields,
            help="Show this field.")
    parser.add_argument(
            '-D', '--define',
            dest='defines',
            action='append',
            type=lambda x: (
                lambda k, vs: (
                    k.strip(),
                    {v.strip() for v in vs.split(',')})
                )(*x.split('=', 1)),
            help="Only include results where this field is this value.")
    class AppendSort(argparse.Action):
        def __call__(self, parser, namespace, value, option):
            if namespace.sort is None:
                namespace.sort = []
            namespace.sort.append((value, True if option == '-S' else False))
    parser.add_argument(
            '-s', '--sort',
            nargs='?',
            action=AppendSort,
            help="Sort by this field.")
    parser.add_argument(
            '-S', '--reverse-sort',
            nargs='?',
            action=AppendSort,
            help="Sort by this field, but backwards.")
    parser.add_argument(
            '-Y', '--summary',
            action='store_true',
            help="Only show the total.")
    parser.add_argument(
            '-F', '--source',
            dest='sources',
            action='append',
            help="Only consider definitions in this file. Defaults to "
                "anything in the current directory.")
    parser.add_argument(
            '--everything',
            action='store_true',
            help="Include builtin and libc specific symbols.")
    parser.add_argument(
            '--nm-types',
            default=NM_TYPES,
            help="Type of symbols to report, this uses the same "
                "single-character type-names emitted by nm. Defaults to "
                "%r." % NM_TYPES)
    parser.add_argument(
            '--nm-path',
            type=lambda x: x.split(),
            default=NM_PATH,
            help="Path to the nm executable, may include flags. "
                "Defaults to %r." % NM_PATH)
    parser.add_argument(
            '--objdump-path',
            type=lambda x: x.split(),
            default=OBJDUMP_PATH,
            help="Path to the objdump executable, may include flags. "
                "Defaults to %r." % OBJDUMP_PATH)
    sys.exit(main(**{k: v
            for k, v in vars(parser.parse_intermixed_args()).items()
            if v is not None}))