зеркало из https://github.com/mozilla/gecko-dev.git
195 строки
5.8 KiB
Python
195 строки
5.8 KiB
Python
# This script exists to auto-generate Http2HuffmanIncoming.h from the table
|
|
# contained in the HPACK spec. It's pretty simple to run:
|
|
# python make_incoming_tables.py < http2_huffman_table.txt > Http2HuffmanIncoming.h
|
|
# where huff_incoming.txt is copy/pasted text from the latest version of the
|
|
# HPACK spec, with all non-relevant lines removed (the most recent version
|
|
# of huff_incoming.txt also lives in this directory as an example).
|
|
import sys
|
|
|
|
def char_cmp(x, y):
|
|
rv = cmp(x['nbits'], y['nbits'])
|
|
if not rv:
|
|
rv = cmp(x['bpat'], y['bpat'])
|
|
if not rv:
|
|
rv = cmp(x['ascii'], y['ascii'])
|
|
return rv
|
|
|
|
characters = []
|
|
|
|
for line in sys.stdin:
|
|
line = line.rstrip()
|
|
obracket = line.rfind('[')
|
|
nbits = int(line[obracket + 1:-1])
|
|
|
|
oparen = line.find(' (')
|
|
ascii = int(line[oparen + 2:oparen + 5].strip())
|
|
|
|
bar = line.find('|', oparen)
|
|
space = line.find(' ', bar)
|
|
bpat = line[bar + 1:space].strip().rstrip('|')
|
|
|
|
characters.append({'ascii': ascii, 'nbits': nbits, 'bpat': bpat})
|
|
|
|
characters.sort(cmp=char_cmp)
|
|
raw_entries = []
|
|
for c in characters:
|
|
raw_entries.append((c['ascii'], c['bpat']))
|
|
|
|
class DefaultList(list):
|
|
def __init__(self, default=None):
|
|
self.__default = default
|
|
|
|
def __ensure_size(self, sz):
|
|
while sz > len(self):
|
|
self.append(self.__default)
|
|
|
|
def __getitem__(self, idx):
|
|
self.__ensure_size(idx + 1)
|
|
rv = super(DefaultList, self).__getitem__(idx)
|
|
return rv
|
|
|
|
def __setitem__(self, idx, val):
|
|
self.__ensure_size(idx + 1)
|
|
super(DefaultList, self).__setitem__(idx, val)
|
|
|
|
def expand_to_8bit(bstr):
|
|
while len(bstr) < 8:
|
|
bstr += '0'
|
|
return int(bstr, 2)
|
|
|
|
table = DefaultList()
|
|
for r in raw_entries:
|
|
ascii, bpat = r
|
|
ascii = int(ascii)
|
|
bstrs = bpat.split('|')
|
|
curr_table = table
|
|
while len(bstrs) > 1:
|
|
idx = expand_to_8bit(bstrs[0])
|
|
if curr_table[idx] is None:
|
|
curr_table[idx] = DefaultList()
|
|
curr_table = curr_table[idx]
|
|
bstrs.pop(0)
|
|
|
|
idx = expand_to_8bit(bstrs[0])
|
|
curr_table[idx] = {'prefix_len': len(bstrs[0]),
|
|
'mask': int(bstrs[0], 2),
|
|
'value': ascii}
|
|
|
|
|
|
def output_table(table, name_suffix=''):
|
|
max_prefix_len = 0
|
|
for i, t in enumerate(table):
|
|
if isinstance(t, dict):
|
|
if t['prefix_len'] > max_prefix_len:
|
|
max_prefix_len = t['prefix_len']
|
|
elif t is not None:
|
|
output_table(t, '%s_%s' % (name_suffix, i))
|
|
|
|
tablename = 'HuffmanIncoming%s' % (name_suffix if name_suffix else 'Root',)
|
|
entriestable = tablename.replace('HuffmanIncoming', 'HuffmanIncomingEntries')
|
|
nexttable = tablename.replace('HuffmanIncoming', 'HuffmanIncomingNextTables')
|
|
sys.stdout.write('static const HuffmanIncomingEntry %s[] = {\n' %
|
|
(entriestable,))
|
|
prefix_len = 0
|
|
value = 0
|
|
i = 0
|
|
while i < 256:
|
|
t = table[i]
|
|
if isinstance(t, dict):
|
|
value = t['value']
|
|
prefix_len = t['prefix_len']
|
|
elif t is not None:
|
|
break
|
|
|
|
sys.stdout.write(' { %s, %s }' %
|
|
(value, prefix_len))
|
|
sys.stdout.write(',\n')
|
|
i += 1
|
|
|
|
indexOfFirstNextTable = i
|
|
if i < 256:
|
|
sys.stdout.write('};\n')
|
|
sys.stdout.write('\n')
|
|
sys.stdout.write('static const HuffmanIncomingTable* %s[] = {\n' %
|
|
(nexttable,))
|
|
while i < 256:
|
|
subtable = '%s_%s' % (name_suffix, i)
|
|
ptr = '&HuffmanIncoming%s' % (subtable,)
|
|
sys.stdout.write(' %s' %
|
|
(ptr))
|
|
sys.stdout.write(',\n')
|
|
i += 1
|
|
else:
|
|
nexttable = 'nullptr'
|
|
|
|
sys.stdout.write('};\n')
|
|
sys.stdout.write('\n')
|
|
sys.stdout.write('static const HuffmanIncomingTable %s = {\n' % (tablename,))
|
|
sys.stdout.write(' %s,\n' % (entriestable,))
|
|
sys.stdout.write(' %s,\n' % (nexttable,))
|
|
sys.stdout.write(' %s,\n' % (indexOfFirstNextTable,))
|
|
sys.stdout.write(' %s\n' % (max_prefix_len,))
|
|
sys.stdout.write('};\n')
|
|
sys.stdout.write('\n')
|
|
|
|
sys.stdout.write('''/*
|
|
* THIS FILE IS AUTO-GENERATED. DO NOT EDIT!
|
|
*/
|
|
#ifndef mozilla__net__Http2HuffmanIncoming_h
|
|
#define mozilla__net__Http2HuffmanIncoming_h
|
|
|
|
namespace mozilla {
|
|
namespace net {
|
|
|
|
struct HuffmanIncomingTable;
|
|
|
|
struct HuffmanIncomingEntry {
|
|
uint16_t mValue:9; // 9 bits so it can hold 0..256
|
|
uint16_t mPrefixLen:7; // only holds 1..8
|
|
};
|
|
|
|
// The data members are public only so they can be statically constructed. All
|
|
// accesses should be done through the functions.
|
|
struct HuffmanIncomingTable {
|
|
// The normal entries, for indices in the range 0..(mNumEntries-1).
|
|
const HuffmanIncomingEntry* const mEntries;
|
|
|
|
// The next tables, for indices in the range mNumEntries..255. Must be
|
|
// |nullptr| if mIndexOfFirstNextTable is 256.
|
|
const HuffmanIncomingTable** const mNextTables;
|
|
|
|
// The index of the first next table (equal to the number of entries in
|
|
// mEntries). This cannot be a uint8_t because it can have the value 256,
|
|
// in which case there are no next tables and mNextTables must be |nullptr|.
|
|
const uint16_t mIndexOfFirstNextTable;
|
|
|
|
const uint8_t mPrefixLen;
|
|
|
|
bool IndexHasANextTable(uint8_t aIndex) const
|
|
{
|
|
return aIndex >= mIndexOfFirstNextTable;
|
|
}
|
|
|
|
const HuffmanIncomingEntry* Entry(uint8_t aIndex) const
|
|
{
|
|
MOZ_ASSERT(aIndex < mIndexOfFirstNextTable);
|
|
return &mEntries[aIndex];
|
|
}
|
|
|
|
const HuffmanIncomingTable* NextTable(uint8_t aIndex) const
|
|
{
|
|
MOZ_ASSERT(aIndex >= mIndexOfFirstNextTable);
|
|
return mNextTables[aIndex - mIndexOfFirstNextTable];
|
|
}
|
|
};
|
|
|
|
''')
|
|
|
|
output_table(table)
|
|
|
|
sys.stdout.write('''} // namespace net
|
|
} // namespace mozilla
|
|
|
|
#endif // mozilla__net__Http2HuffmanIncoming_h
|
|
''')
|