_clean_data benchmarking

bench.py

import os
import re
import StringIO
import cStringIO
import contextlib

# caching the compilation of the regex used
# to check for lookup calls within data
LOOKUP_REGEX=re.compile(r'lookup\s*\(')


def _clean_data(orig_data, from_remote=False, from_inventory=False):
    ''' remove jinja2 template tags from a string '''

    if not isinstance(orig_data, basestring):
        return orig_data

    data = StringIO.StringIO("")

    # when the data is marked as having come from a remote, we always
    # replace any print blocks (ie. {{var}}), however when marked as coming
    # from inventory we only replace print blocks that contain a call to
    # a lookup plugin (ie. {{lookup('foo','bar'))}})
    replace_prints = from_remote or (from_inventory and '{{' in orig_data and LOOKUP_REGEX.search(orig_data) is not None)

    # these variables keep track of opening block locations, as we only
    # want to replace matched pairs of print/block tags
    print_openings = []
    block_openings = []

    for idx,c in enumerate(orig_data):
        # if the current character is an opening brace, check to
        # see if this is a jinja2 token. Otherwise, if the current
        # character is a closing brace, we backup one character to
        # see if we have a closing.
        if c == '{' and idx < len(orig_data) - 1:
            token = orig_data[idx:idx+2]
            # if so, and we want to replace this block, push
            # this token's location onto the appropriate array
            if token == '{{' and replace_prints:
                print_openings.append(idx)
            elif token == '{%':
                block_openings.append(idx)
            # finally we write the data to the buffer and write
            data.seek(0, os.SEEK_END)
            data.write(c)
        elif c == '}' and idx > 0:
            token = orig_data[idx-1:idx+1]
            prev_idx = -1
            if token == '%}' and len(block_openings) > 0:
                prev_idx = block_openings.pop()
            elif token == '}}' and len(print_openings) > 0:
                prev_idx = print_openings.pop()
            # if we have a closing token, and we have previously found
            # the opening to the same kind of block represented by this
            # token, replace both occurrences, otherwise we just write
            # the current character to the buffer
            if prev_idx != -1:
                # replace the opening
                data.seek(prev_idx, os.SEEK_SET)
                data.write('{#')
                # replace the closing
                data.seek(-1, os.SEEK_END)
                data.write('#}')
            else:
                data.seek(0, os.SEEK_END)
                data.write(c)
        else:
            # not a jinja2 token, so we just write the current char
            # to the output buffer
            data.seek(0, os.SEEK_END)
            data.write(c)
    return_data = data.getvalue()
    data.close()
    return return_data



def _clean_data_cstringio(orig_data, from_remote=False, from_inventory=False):
    ''' remove jinja2 template tags from a string '''

    if not isinstance(orig_data, basestring):
        return orig_data

    data = cStringIO.StringIO()

    # when the data is marked as having come from a remote, we always
    # replace any print blocks (ie. {{var}}), however when marked as coming
    # from inventory we only replace print blocks that contain a call to
    # a lookup plugin (ie. {{lookup('foo','bar'))}})
    replace_prints = from_remote or (from_inventory and '{{' in orig_data and LOOKUP_REGEX.search(orig_data) is not None)

    # these variables keep track of opening block locations, as we only
    # want to replace matched pairs of print/block tags
    print_openings = []
    block_openings = []

    for idx,c in enumerate(orig_data):
        # if the current character is an opening brace, check to
        # see if this is a jinja2 token. Otherwise, if the current
        # character is a closing brace, we backup one character to
        # see if we have a closing.
        if c == '{' and idx < len(orig_data) - 1:
            token = orig_data[idx:idx+2]
            # if so, and we want to replace this block, push
            # this token's location onto the appropriate array
            if token == '{{' and replace_prints:
                print_openings.append(idx)
            elif token == '{%':
                block_openings.append(idx)
            # finally we write the data to the buffer and write
            data.seek(0, os.SEEK_END)
            data.write(c)
        elif c == '}' and idx > 0:
            token = orig_data[idx-1:idx+1]
            prev_idx = -1
            if token == '%}' and len(block_openings) > 0:
                prev_idx = block_openings.pop()
            elif token == '}}' and len(print_openings) > 0:
                prev_idx = print_openings.pop()
            # if we have a closing token, and we have previously found
            # the opening to the same kind of block represented by this
            # token, replace both occurrences, otherwise we just write
            # the current character to the buffer
            if prev_idx != -1:
                # replace the opening
                data.seek(prev_idx, os.SEEK_SET)
                data.write('{#')
                # replace the closing
                data.seek(-1, os.SEEK_END)
                data.write('#}')
            else:
                data.seek(0, os.SEEK_END)
                data.write(c)
        else:
            # not a jinja2 token, so we just write the current char
            # to the output buffer
            data.seek(0, os.SEEK_END)
            data.write(c)
    return_data = data.getvalue()
    data.close()
    return return_data


PRINT_CODE_REGEX = re.compile(r'(?:{[{%]|[%}]})')
ONLY_CODE_REGEX = re.compile(r'(?:{%|%})')

def _regex(orig_data, from_remote=False, from_inventory=False):
    ''' remove jinja2 template tags from a string '''

    if not isinstance(orig_data, basestring):
        return orig_data

    # when the data is marked as having come from a remote, we always
    # replace any print blocks (ie. {{var}}), however when marked as coming
    # from inventory we only replace print blocks that contain a call to
    # a lookup plugin (ie. {{lookup('foo','bar'))}})
    replace_prints = from_remote or (from_inventory and '{{' in orig_data and LOOKUP_REGEX.search(orig_data) is not None)

    regex = PRINT_CODE_REGEX if replace_prints else ONLY_CODE_REGEX

    with contextlib.closing(cStringIO.StringIO()) as data:
        # these variables keep track of opening block locations, as we only
        # want to replace matched pairs of print/block tags
        last_pos = 0
        print_openings = []
        block_openings = []
        for mo in regex.finditer(orig_data):
            token = mo.group(0)
            token_start = mo.start(0)
            token_end = mo.end(0)

            if token[0] == '{':
                if token == '{%':
                    block_openings.append(token_start)
                elif token == '{{':
                    print_openings.append(token_start)
                data.write(orig_data[last_pos:token_end])
            elif token[1] == '}':
                prev_idx = None
                if token == '%}' and block_openings:
                    prev_idx = block_openings.pop()
                elif token == '}}' and print_openings:
                    prev_idx = print_openings.pop()

                data.write(orig_data[last_pos:token_start])
                if prev_idx is not None:
                    # replace the opening
                    data.seek(prev_idx, os.SEEK_SET)
                    data.write('{#')
                    # replace the closing
                    data.seek(0, os.SEEK_END)
                    data.write('#}')
                else:
                    data.write(token)
            else:
                assert False, 'Unhandled regex match'
            last_pos = token_end
        
        data.write(orig_data[last_pos:])

        return data.getvalue()

def generate_test_strings():
    template = '{{ }}'
    whitespace = ' ' * len(template)
    templates = (
        template * 4,
        (template * 3) + whitespace,
        (template + whitespace) * 2,
        template + (whitespace * 3),
        whitespace * 4,
    )

    for tens in xrange(0,4):
        length = 10 ** tens

        for i, template in enumerate(templates):
            yield i / 4.0, template * length

import timeit

to_bench = {
	'orig': _clean_data,
	'orig cs': _clean_data_cstringio,
    'regex': _regex,
}

tests = {
	'none     ': dict(from_remote=False, from_inventory=False),
	'remote   ': dict(from_remote=False, from_inventory=False),
	'inventory': dict(from_remote=False, from_inventory=False),
	'both     ': dict(from_remote=False, from_inventory=False)
}

name_padding = max(len(s) for s in to_bench.keys())
for t in generate_test_strings():
	test_string = intern(t[1])
	print 'length:', len(t[1]), 'density:', t[0]
	for test, kwargs in tests.iteritems():
		for name, fn in to_bench.iteritems():
			fn(t, **kwargs)
			print name.ljust(name_padding), test, '{:.8f}'.format(min(timeit.repeat(
				stmt=lambda: fn(test_string, **kwargs),
				number=1000,
				repeat=20)))
	print

output

length: 20 density: 0.0
regex   none      0.00233603
orig cs none      0.01116395
orig    none      0.03076696
regex   inventory 0.00235701
orig cs inventory 0.01127720
orig    inventory 0.03086901
regex   both      0.00232005
orig cs both      0.01120687
orig    both      0.03084397
regex   remote    0.00232410
orig cs remote    0.01129794
orig    remote    0.03086996

length: 20 density: 0.25
regex   none      0.00231910
orig cs none      0.01046991
orig    none      0.03019309
regex   inventory 0.00233412
orig cs inventory 0.01054406
orig    inventory 0.03022504
regex   both      0.00233984
orig cs both      0.01054120
orig    both      0.03017092
regex   remote    0.00230885
orig cs remote    0.01054597
orig    remote    0.03020096

length: 20 density: 0.5
regex   none      0.00226808
orig cs none      0.00974989
orig    none      0.02947187
regex   inventory 0.00223994
orig cs inventory 0.00973392
orig    inventory 0.02951097
regex   both      0.00225091
orig cs both      0.00973988
orig    both      0.02943206
regex   remote    0.00227809
orig cs remote    0.00979209
orig    remote    0.02976108

length: 20 density: 0.75
regex   none      0.00228810
orig cs none      0.00916791
orig    none      0.02876782
regex   inventory 0.00223589
orig cs inventory 0.00901389
orig    inventory 0.02870798
regex   both      0.00226593
orig cs both      0.00899220
orig    both      0.02879286
regex   remote    0.00224280
orig cs remote    0.00903201
orig    remote    0.02879405

length: 20 density: 1.0
regex   none      0.00212598
orig cs none      0.00824499
orig    none      0.02791786
regex   inventory 0.00211096
orig cs inventory 0.00822306
orig    inventory 0.02799916
regex   both      0.00214314
orig cs both      0.00822902
orig    both      0.02803493
regex   remote    0.00210404
orig cs remote    0.00870681
orig    remote    0.02804685

length: 200 density: 0.0
regex   none      0.00437117
orig cs none      0.10380411
orig    none      0.29348207
regex   inventory 0.00436783
orig cs inventory 0.10435104
orig    inventory 0.29324698
regex   both      0.00440097
orig cs both      0.10405898
orig    both      0.29357386
regex   remote    0.00437903
orig cs remote    0.10408211
orig    remote    0.29358387

length: 200 density: 0.25
regex   none      0.00419116
orig cs none      0.09632802
orig    none      0.28626513
regex   inventory 0.00419402
orig cs inventory 0.09633589
orig    inventory 0.28691602
regex   both      0.00421095
orig cs both      0.09631491
orig    both      0.28645802
regex   remote    0.00419688
orig cs remote    0.09635997
orig    remote    0.28712702

length: 200 density: 0.5
regex   none      0.00368094
orig cs none      0.08877921
orig    none      0.27943301
regex   inventory 0.00369215
orig cs inventory 0.08880091
orig    inventory 0.27882290
regex   both      0.00374818
orig cs both      0.08862686
orig    both      0.30531192
regex   remote    0.00372386
orig cs remote    0.10351205
orig    remote    0.32544017

length: 200 density: 0.75
regex   none      0.00362611
orig cs none      0.09516382
orig    none      0.32432103
regex   inventory 0.00345397
orig cs inventory 0.09414315
orig    inventory 0.27665305
regex   both      0.00341392
orig cs both      0.08206987
orig    both      0.27330804
regex   remote    0.00338101
orig cs remote    0.08216286
orig    remote    0.27378011

length: 200 density: 1.0
regex   none      0.00282884
orig cs none      0.07388496
orig    none      0.26605392
regex   inventory 0.00289392
orig cs inventory 0.07474899
orig    inventory 0.26591420
regex   both      0.00286889
orig cs both      0.07398915
orig    both      0.26416993
regex   remote    0.00287604
orig cs remote    0.07397103
orig    remote    0.26444697

length: 2000 density: 0.0
regex   none      0.02651596
orig cs none      1.04446483
orig    none      3.16794777
regex   inventory 0.02648711
orig cs inventory 1.04291296
orig    inventory 3.17617607
regex   both      0.02652907
orig cs both      1.04543400
orig    both      3.17354608
regex   remote    0.02650285
orig cs remote    1.04383111
orig    remote    3.17507601

length: 2000 density: 0.25
regex   none      0.02319193
orig cs none      0.97022009
orig    none      3.08056402
regex   inventory 0.02326107
orig cs inventory 0.97225213
orig    inventory 3.09071088
regex   both      0.02322698
orig cs both      0.97343111
orig    both      3.08821392
regex   remote    0.02319694
orig cs remote    0.97419095
orig    remote    3.09205699

length: 2000 density: 0.5
regex   none      0.01878405
orig cs none      0.89367700
orig    none      3.02064180
regex   inventory 0.01838994
orig cs inventory 0.89543295
orig    inventory 3.01873088
regex   both      0.01818085
orig cs both      0.89289713
orig    both      3.01739812
regex   remote    0.01843214
orig cs remote    0.89358997
orig    remote    3.01982999

length: 2000 density: 0.75
regex   none      0.01398683
orig cs none      0.81412983
orig    none      2.92859387
regex   inventory 0.01400113
orig cs inventory 0.81458783
orig    inventory 2.93060899
regex   both      0.01399398
orig cs both      0.81417394
orig    both      2.92725301
regex   remote    0.01393986
orig cs remote    0.81492209
orig    remote    2.92646503

length: 2000 density: 1.0
regex   none      0.00927281
orig cs none      0.73526096
orig    none      2.84585190
regex   inventory 0.00925303
orig cs inventory 0.73610306
orig    inventory 2.85367703
regex   both      0.00924015
orig cs both      0.73535395
orig    both      2.84879208
regex   remote    0.00921488
orig cs remote    0.73591018
orig    remote    2.84457397