110 lines
3.1 KiB
Python
110 lines
3.1 KiB
Python
|
#!/usr/bin/env python
|
||
|
#
|
||
|
# This script takes directories which contain the hpack-test-case json
|
||
|
# files, and calculates the compression ratio in each file and outputs
|
||
|
# the result in table formatted in rst.
|
||
|
#
|
||
|
# The each directory contains the result of various HPACK compressor.
|
||
|
#
|
||
|
# The table is laid out so that we can see that how input header set
|
||
|
# in one json file is compressed in each compressor.
|
||
|
#
|
||
|
# For hpack-test-case, see https://github.com/Jxck/hpack-test-case
|
||
|
#
|
||
|
import sys, json, os, re
|
||
|
|
||
|
class Stat:
|
||
|
|
||
|
def __init__(self, complen, srclen):
|
||
|
self.complen = complen
|
||
|
self.srclen = srclen
|
||
|
|
||
|
def compute_stat(jsdata):
|
||
|
complen = 0
|
||
|
srclen = 0
|
||
|
for item in jsdata['cases']:
|
||
|
complen += len(item['wire']) // 2
|
||
|
srclen += \
|
||
|
sum([len(list(x.keys())[0]) + len(list(x.values())[0]) \
|
||
|
for x in item['headers']])
|
||
|
return Stat(complen, srclen)
|
||
|
|
||
|
def format_result(r):
|
||
|
return '{:.02f} ({}/{}) '.format(r.complen/r.srclen, r.complen, r.srclen)
|
||
|
|
||
|
if __name__ == '__main__':
|
||
|
entries = [(os.path.basename(re.sub(r'/+$', '', p)), p) \
|
||
|
for p in sys.argv[1:]]
|
||
|
maxnamelen = 0
|
||
|
maxstorynamelen = 0
|
||
|
res = {}
|
||
|
|
||
|
stories = set()
|
||
|
for name, ent in entries:
|
||
|
files = [p for p in os.listdir(ent) if p.endswith('.json')]
|
||
|
res[name] = {}
|
||
|
maxnamelen = max(maxnamelen, len(name))
|
||
|
for fn in files:
|
||
|
stories.add(fn)
|
||
|
maxstorynamelen = max(maxstorynamelen, len(fn))
|
||
|
with open(os.path.join(ent, fn)) as f:
|
||
|
input = f.read()
|
||
|
rv = compute_stat(json.loads(input))
|
||
|
res[name][fn] = rv
|
||
|
maxnamelen = max(maxnamelen, len(format_result(rv)))
|
||
|
stories = list(stories)
|
||
|
stories.sort()
|
||
|
|
||
|
storynameformat = '{{:{}}} '.format(maxstorynamelen)
|
||
|
nameformat = '{{:{}}} '.format(maxnamelen)
|
||
|
|
||
|
|
||
|
sys.stdout.write('''\
|
||
|
hpack-test-case compression ratio
|
||
|
=================================
|
||
|
|
||
|
The each cell has ``X (Y/Z)`` format:
|
||
|
|
||
|
X
|
||
|
Y / Z
|
||
|
Y
|
||
|
number of bytes after compression
|
||
|
Z
|
||
|
number of bytes before compression
|
||
|
|
||
|
''')
|
||
|
|
||
|
def write_border():
|
||
|
sys.stdout.write('='*maxstorynamelen)
|
||
|
sys.stdout.write(' ')
|
||
|
for _ in entries:
|
||
|
sys.stdout.write('='*maxnamelen)
|
||
|
sys.stdout.write(' ')
|
||
|
sys.stdout.write('\n')
|
||
|
|
||
|
write_border()
|
||
|
|
||
|
sys.stdout.write(storynameformat.format('story'))
|
||
|
for name, _ in entries:
|
||
|
sys.stdout.write(nameformat.format(name))
|
||
|
sys.stdout.write('\n')
|
||
|
|
||
|
write_border()
|
||
|
|
||
|
for story in stories:
|
||
|
sys.stdout.write(storynameformat.format(story))
|
||
|
srclen = -1
|
||
|
for name, _ in entries:
|
||
|
stats = res[name]
|
||
|
if story not in stats:
|
||
|
sys.stdout.write(nameformat.format('N/A'))
|
||
|
continue
|
||
|
if srclen == -1:
|
||
|
srclen = stats[story].srclen
|
||
|
elif srclen != stats[story].srclen:
|
||
|
raise Exception('Bad srclen')
|
||
|
sys.stdout.write(nameformat.format(format_result(stats[story])))
|
||
|
sys.stdout.write('\n')
|
||
|
|
||
|
write_border()
|