mirror of https://github.com/arendst/Tasmota.git
521 lines
23 KiB
Python
521 lines
23 KiB
Python
#!/usr/bin/env python
|
|
# -*- coding: utf-8 -*-
|
|
"""
|
|
Python Class for compressing short strings.
|
|
|
|
This class contains a highly modified and optimized version of Unishox
|
|
for Tasmota converted in C ported to Pyhton3.
|
|
|
|
It was basically developed to individually compress and decompress small strings
|
|
(see https://github.com/siara-cc/Unishox)
|
|
In general compression utilities such as zip, gzip do not compress short strings
|
|
well and often expand them. They also use lots of memory which makes them unusable
|
|
in constrained environments like Arduino.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License.
|
|
"""
|
|
|
|
class Unishox:
|
|
"""
|
|
This is a highly modified and optimized version of Unishox
|
|
for Tasmota, aimed at compressing `Rules` which are typically
|
|
short strings from 50 to 500 bytes.
|
|
|
|
@author Stephan Hadinger
|
|
@revised Norbert Richter
|
|
"""
|
|
|
|
# pylint: disable=bad-continuation,bad-whitespace,line-too-long
|
|
#cl_95 = [0x4000 + 3, 0x3F80 + 11, 0x3D80 + 11, 0x3C80 + 10, 0x3BE0 + 12, 0x3E80 + 10, 0x3F40 + 11, 0x3EC0 + 10, 0x3BA0 + 11, 0x3BC0 + 11, 0x3D60 + 11, 0x3B60 + 11, 0x3A80 + 10, 0x3AC0 + 10, 0x3A00 + 9, 0x3B00 + 10, 0x38C0 + 10, 0x3900 + 10, 0x3940 + 11, 0x3960 + 11, 0x3980 + 11, 0x39A0 + 11, 0x39C0 + 11, 0x39E0 + 12, 0x39F0 + 12, 0x3880 + 10, 0x3CC0 + 10, 0x3C00 + 9, 0x3D00 + 10, 0x3E00 + 9, 0x3F00 + 10, 0x3B40 + 11, 0x3BF0 + 12, 0x2B00 + 8, 0x21C0 + 11, 0x20C0 + 10, 0x2100 + 10, 0x2600 + 7, 0x2300 + 11, 0x21E0 + 12, 0x2140 + 11, 0x2D00 + 8, 0x2358 + 13, 0x2340 + 12, 0x2080 + 10, 0x21A0 + 11, 0x2E00 + 8, 0x2C00 + 8, 0x2180 + 11, 0x2350 + 13, 0x2F80 + 9, 0x2F00 + 9, 0x2A00 + 8, 0x2160 + 11, 0x2330 + 12, 0x21F0 + 12, 0x2360 + 13, 0x2320 + 12, 0x2368 + 13, 0x3DE0 + 12, 0x3FA0 + 11, 0x3DF0 + 12, 0x3D40 + 11, 0x3F60 + 11, 0x3FF0 + 12, 0xB000 + 4, 0x1C00 + 7, 0x0C00 + 6, 0x1000 + 6, 0x6000 + 3, 0x3000 + 7, 0x1E00 + 8, 0x1400 + 7, 0xD000 + 4, 0x3580 + 9, 0x3400 + 8, 0x0800 + 6, 0x1A00 + 7, 0xE000 + 4, 0xC000 + 4, 0x1800 + 7, 0x3500 + 9, 0xF800 + 5, 0xF000 + 5, 0xA000 + 4, 0x1600 + 7, 0x3300 + 8, 0x1F00 + 8, 0x3600 + 9, 0x3200 + 8, 0x3680 + 9, 0x3DA0 + 11, 0x3FC0 + 11, 0x3DC0 + 11, 0x3FE0 + 12]
|
|
cl_95 = [0x4000 + 3, 0x3F80 + 11, 0x3D80 + 11, 0x3C80 + 10, 0x3BE0 + 12, 0x3E80 + 10, 0x3F40 + 11, 0x3EC0 + 10, 0x3BA0 + 11, 0x3BC0 + 11, 0x3D60 + 11, 0x3B60 + 11, 0x3A80 + 10, 0x3AC0 + 10, 0x3A00 + 9, 0x3B00 + 10, 0x38C0 + 10, 0x3900 + 10, 0x3940 + 11, 0x3960 + 11, 0x3980 + 11, 0x39A0 + 11, 0x39C0 + 11, 0x39E0 + 12, 0x39F0 + 12, 0x3880 + 10, 0x3CC0 + 10, 0x3C00 + 9, 0x3D00 + 10, 0x3E00 + 9, 0x3F00 + 10, 0x3B40 + 11, 0x3BF0 + 12, 0x2B00 + 8, 0x21C0 + 11, 0x20C0 + 10, 0x2100 + 10, 0x2600 + 7, 0x2300 + 11, 0x21E0 + 12, 0x2140 + 11, 0x2D00 + 8, 0x46B0 + 13, 0x2340 + 12, 0x2080 + 10, 0x21A0 + 11, 0x2E00 + 8, 0x2C00 + 8, 0x2180 + 11, 0x46A0 + 13, 0x2F80 + 9, 0x2F00 + 9, 0x2A00 + 8, 0x2160 + 11, 0x2330 + 12, 0x21F0 + 12, 0x46C0 + 13, 0x2320 + 12, 0x46D0 + 13, 0x3DE0 + 12, 0x3FA0 + 11, 0x3DF0 + 12, 0x3D40 + 11, 0x3F60 + 11, 0x3FF0 + 12, 0xB000 + 4, 0x1C00 + 7, 0x0C00 + 6, 0x1000 + 6, 0x6000 + 3, 0x3000 + 7, 0x1E00 + 8, 0x1400 + 7, 0xD000 + 4, 0x3580 + 9, 0x3400 + 8, 0x0800 + 6, 0x1A00 + 7, 0xE000 + 4, 0xC000 + 4, 0x1800 + 7, 0x3500 + 9, 0xF800 + 5, 0xF000 + 5, 0xA000 + 4, 0x1600 + 7, 0x3300 + 8, 0x1F00 + 8, 0x3600 + 9, 0x3200 + 8, 0x3680 + 9, 0x3DA0 + 11, 0x3FC0 + 11, 0x3DC0 + 11, 0x3FE0 + 12]
|
|
|
|
# enum {SHX_STATE_1 = 1, SHX_STATE_2}; // removed Unicode state
|
|
SHX_STATE_1 = 1
|
|
SHX_STATE_2 = 2
|
|
|
|
SHX_SET1 = 0
|
|
SHX_SET1A = 1
|
|
SHX_SET1B = 2
|
|
SHX_SET2 = 3
|
|
|
|
sets = [['\0', ' ', 'e', '\0', 't', 'a', 'o', 'i', 'n', 's', 'r'],
|
|
['\0', 'l', 'c', 'd', 'h', 'u', 'p', 'm', 'b', 'g', 'w'],
|
|
['f', 'y', 'v', 'k', 'q', 'j', 'x', 'z', '\0', '\0', '\0'],
|
|
['\0', '9', '0', '1', '2', '3', '4', '5', '6', '7', '8'],
|
|
['.', ',', '-', '/', '?', '+', ' ', '(', ')', '$', '@'],
|
|
[';', '#', ':', '<', '^', '*', '"', '{', '}', '[', ']'],
|
|
['=', '%', '\'', '>', '&', '_', '!', '\\', '|', '~', '`']]
|
|
|
|
us_vcode = [2 + (0 << 3), 3 + (3 << 3), 3 + (1 << 3), 4 + (6 << 3), 0,
|
|
# 5, 6, 7, 8, 9, 10
|
|
4 + (4 << 3), 3 + (2 << 3), 4 + (8 << 3), 0, 0, 0,
|
|
# 11, 12, 13, 14, 15
|
|
4 + (7 << 3), 0, 4 + (5 << 3), 0, 5 + (9 << 3),
|
|
# 16, 17, 18, 19, 20, 21, 22, 23
|
|
0, 0, 0, 0, 0, 0, 0, 0,
|
|
# 24, 25, 26, 27, 28, 29, 30, 31
|
|
0, 0, 0, 0, 0, 0, 0, 5 + (10 << 3) ]
|
|
# 0, 1, 2, 3, 4, 5, 6, 7,
|
|
us_hcode = [1 + (1 << 3), 2 + (0 << 3), 0, 3 + (2 << 3), 0, 0, 0, 5 + (3 << 3),
|
|
# 8, 9, 10, 11, 12, 13, 14, 15,
|
|
0, 0, 0, 0, 0, 0, 0, 5 + (5 << 3),
|
|
# 16, 17, 18, 19, 20, 21, 22, 23
|
|
0, 0, 0, 0, 0, 0, 0, 5 + (4 << 3),
|
|
# 24, 25, 26, 27, 28, 29, 30, 31
|
|
0, 0, 0, 0, 0, 0, 0, 5 + (6 << 3) ]
|
|
# pylint: enable=bad-continuation,bad-whitespace
|
|
|
|
ESCAPE_MARKER = 0x2A
|
|
|
|
TERM_CODE = 0x37C0
|
|
# TERM_CODE_LEN = 10
|
|
DICT_CODE = 0x0000
|
|
DICT_CODE_LEN = 5
|
|
#DICT_OTHER_CODE = 0x0000
|
|
#DICT_OTHER_CODE_LEN = 6
|
|
RPT_CODE_TASMOTA = 0x3780
|
|
RPT_CODE_TASMOTA_LEN = 10
|
|
BACK2_STATE1_CODE = 0x2000
|
|
BACK2_STATE1_CODE_LEN = 4
|
|
#BACK_FROM_UNI_CODE = 0xFE00
|
|
#BACK_FROM_UNI_CODE_LEN = 8
|
|
LF_CODE = 0x3700
|
|
LF_CODE_LEN = 9
|
|
TAB_CODE = 0x2400
|
|
TAB_CODE_LEN = 7
|
|
ALL_UPPER_CODE = 0x2200
|
|
ALL_UPPER_CODE_LEN = 8
|
|
SW2_STATE2_CODE = 0x3800
|
|
SW2_STATE2_CODE_LEN = 7
|
|
ST2_SPC_CODE = 0x3B80
|
|
ST2_SPC_CODE_LEN = 11
|
|
BIN_CODE_TASMOTA = 0x8000
|
|
BIN_CODE_TASMOTA_LEN = 3
|
|
|
|
NICE_LEN = 5
|
|
|
|
mask = [0x80, 0xC0, 0xE0, 0xF0, 0xF8, 0xFC, 0xFE, 0xFF]
|
|
|
|
# pylint: disable=missing-function-docstring,invalid-name
|
|
|
|
# Input
|
|
# out = bytearray
|
|
def append_bits(self, out, ol, code, clen, state):
|
|
#print("Append bits {ol} {code} {clen} {state}".format(ol=ol, code=code, clen=clen, state=state))
|
|
if state == self.SHX_STATE_2:
|
|
# remove change state prefix
|
|
if (code >> 9) == 0x1C:
|
|
code <<= 7
|
|
clen -= 7
|
|
while clen > 0:
|
|
cur_bit = ol % 8
|
|
blen = 8 if (clen > 8) else clen
|
|
a_byte = (code >> 8) & self.mask[blen - 1]
|
|
#print("append_bits a_byte {ab} blen {blen}".format(ab=a_byte,blen=blen))
|
|
a_byte >>= cur_bit
|
|
if blen + cur_bit > 8:
|
|
blen = (8 - cur_bit)
|
|
if cur_bit == 0:
|
|
out[ol // 8] = a_byte
|
|
else:
|
|
out[ol // 8] |= a_byte
|
|
code <<= blen
|
|
ol += blen
|
|
if 0 == ol % 8: # pylint: disable=misplaced-comparison-constant
|
|
# we completed a full byte
|
|
last_c = out[(ol // 8) - 1]
|
|
if last_c in (0, self.ESCAPE_MARKER):
|
|
out[ol // 8] = 1 + last_c # increment to 0x01 or 0x2B
|
|
out[(ol // 8) -1] = self.ESCAPE_MARKER # replace old value with marker
|
|
ol += 8 # add one full byte
|
|
clen -= blen
|
|
return ol
|
|
|
|
codes = [0x82, 0xC3, 0xE5, 0xED, 0xF5] # pylint: disable=bad-whitespace
|
|
bit_len = [ 5, 7, 9, 12, 16] # pylint: disable=bad-whitespace
|
|
|
|
def encodeCount(self, out, ol, count):
|
|
#print("encodeCount ol = {ol}, count = {count}".format(ol=ol, count=count))
|
|
till = 0
|
|
base = 0
|
|
for i in range(len(self.bit_len)):
|
|
bit_len_i = self.bit_len[i]
|
|
till += (1 << bit_len_i)
|
|
if count < till:
|
|
codes_i = self.codes[i]
|
|
ol = self.append_bits(out, ol, (codes_i & 0xF8) << 8, codes_i & 0x07, 1)
|
|
#print("encodeCount append_bits ol = {ol}, code = {code}, len = {len}".format(ol=ol,code=(codes_i & 0xF8) << 8,len=codes_i & 0x07))
|
|
ol = self.append_bits(out, ol, (count - base) << (16 - bit_len_i), bit_len_i, 1)
|
|
#print("encodeCount append_bits ol = {ol}, code = {code}, len = {len}".format(ol=ol,code=(count - base) << (16 - bit_len_i),len=bit_len_i))
|
|
return ol
|
|
base = till
|
|
return ol
|
|
|
|
# Returns (int, ol, state, is_all_upper)
|
|
def matchOccurance(self, inn, len_, l_, out, ol, state, is_all_upper):
|
|
# int j, k;
|
|
longest_dist = 0
|
|
longest_len = 0
|
|
#for (j = l_ - self.NICE_LEN; j >= 0; j--) {
|
|
j = l_ - self.NICE_LEN
|
|
while j >= 0:
|
|
k = l_
|
|
#for (k = l_; k < len && j + k - l_ < l_; k++) {
|
|
while k < len_ and j + k - l_ < l_:
|
|
if inn[k] != inn[j + k - l_]:
|
|
break
|
|
k += 1
|
|
if k - l_ > self.NICE_LEN - 1:
|
|
match_len = k - l_ - self.NICE_LEN
|
|
match_dist = l_ - j - self.NICE_LEN + 1
|
|
if match_len > longest_len:
|
|
longest_len = match_len
|
|
longest_dist = match_dist
|
|
j -= 1
|
|
|
|
if longest_len:
|
|
#print("longest_len {ll}".format(ll=longest_len))
|
|
#ol_save = ol
|
|
if state == self.SHX_STATE_2 or is_all_upper:
|
|
is_all_upper = 0
|
|
state = self.SHX_STATE_1
|
|
ol = self.append_bits(out, ol, self.BACK2_STATE1_CODE, self.BACK2_STATE1_CODE_LEN, state)
|
|
|
|
ol = self.append_bits(out, ol, self.DICT_CODE, self.DICT_CODE_LEN, 1)
|
|
ol = self.encodeCount(out, ol, longest_len)
|
|
ol = self.encodeCount(out, ol, longest_dist)
|
|
#print("longest_len {ll} longest_dist {ld} ol {ols}-{ol}".format(ll=longest_len, ld=longest_dist, ol=ol, ols=ol_save))
|
|
l_ += longest_len + self.NICE_LEN
|
|
l_ -= 1
|
|
|
|
return l_, ol, state, is_all_upper
|
|
return -l_, ol, state, is_all_upper
|
|
|
|
|
|
def compress(self, inn, len_, out, len_out):
|
|
ol = 0
|
|
state = self.SHX_STATE_1
|
|
is_all_upper = 0
|
|
l = 0
|
|
while l < len_:
|
|
# for (l=0; l<len_; l++) {
|
|
|
|
c_in = inn[l]
|
|
|
|
if l and l < len_ - 4:
|
|
if c_in == inn[l - 1] and c_in == inn[l + 1] and c_in == inn[l + 2] and c_in == inn[l + 3]:
|
|
rpt_count = l + 4
|
|
while rpt_count < len_ and inn[rpt_count] == c_in:
|
|
rpt_count += 1
|
|
rpt_count -= l
|
|
|
|
if state == self.SHX_STATE_2 or is_all_upper:
|
|
is_all_upper = 0
|
|
state = self.SHX_STATE_1
|
|
ol = self.append_bits(out, ol, self.BACK2_STATE1_CODE, self.BACK2_STATE1_CODE_LEN, state) # back to lower case and Set1
|
|
|
|
ol = self.append_bits(out, ol, self.RPT_CODE_TASMOTA, self.RPT_CODE_TASMOTA_LEN, 1) # reusing CRLF for RPT
|
|
ol = self.encodeCount(out, ol, rpt_count - 4)
|
|
l += rpt_count
|
|
#l -= 1
|
|
continue
|
|
|
|
if l < (len_ - self.NICE_LEN + 1):
|
|
#l_old = l
|
|
(l, ol, state, is_all_upper) = self.matchOccurance(inn, len_, l, out, ol, state, is_all_upper)
|
|
if l > 0:
|
|
#print("matchOccurance l = {l} l_old = {lo}".format(l=l,lo=l_old))
|
|
l += 1 # for loop
|
|
continue
|
|
|
|
l = -l
|
|
|
|
if state == self.SHX_STATE_2: # if Set2
|
|
if ord(' ') <= c_in <= ord('@') or ord('[') <= c_in <= ord('`') or ord('{') <= c_in <= ord('~'):
|
|
pass
|
|
else:
|
|
state = self.SHX_STATE_1 # back to Set1 and lower case
|
|
ol = self.append_bits(out, ol, self.BACK2_STATE1_CODE, self.BACK2_STATE1_CODE_LEN, state)
|
|
|
|
is_upper = 0
|
|
if ord('A') <= c_in <= ord('Z'):
|
|
is_upper = 1
|
|
else:
|
|
if is_all_upper:
|
|
is_all_upper = 0
|
|
ol = self.append_bits(out, ol, self.BACK2_STATE1_CODE, self.BACK2_STATE1_CODE_LEN, state)
|
|
|
|
if 32 <= c_in <= 126:
|
|
if is_upper and not is_all_upper:
|
|
ll = l+5
|
|
# for (ll=l+5; ll>=l && ll<len_; ll--) {
|
|
while l <= ll < len_:
|
|
if inn[ll] < ord('A') or inn[ll] > ord('Z'):
|
|
break
|
|
|
|
ll -= 1
|
|
|
|
if ll == l-1:
|
|
ol = self.append_bits(out, ol, self.ALL_UPPER_CODE, self.ALL_UPPER_CODE_LEN, state) # CapsLock
|
|
is_all_upper = 1
|
|
|
|
if state == self.SHX_STATE_1 and ord('0') <= c_in <= ord('9'):
|
|
ol = self.append_bits(out, ol, self.SW2_STATE2_CODE, self.SW2_STATE2_CODE_LEN, state) # Switch to sticky Set2
|
|
state = self.SHX_STATE_2
|
|
|
|
c_in -= 32
|
|
if is_all_upper and is_upper:
|
|
c_in += 32
|
|
if c_in == 0 and state == self.SHX_STATE_2:
|
|
ol = self.append_bits(out, ol, self.ST2_SPC_CODE, self.ST2_SPC_CODE_LEN, state) # space from Set2 ionstead of Set1
|
|
else:
|
|
# ol = self.append_bits(out, ol, pgm_read_word(&c_95[c_in]), pgm_read_byte(&l_95[c_in]), state); // original version with c/l in split arrays
|
|
cl = self.cl_95[c_in]
|
|
cl_code = cl & 0xFFF0
|
|
cl_len = cl & 0x000F
|
|
if cl_len == 13:
|
|
cl_code = cl_code >> 1
|
|
ol = self.append_bits(out, ol, cl_code, cl_len, state)
|
|
|
|
elif c_in == 10:
|
|
ol = self.append_bits(out, ol, self.LF_CODE, self.LF_CODE_LEN, state) # LF
|
|
elif c_in == '\t':
|
|
ol = self.append_bits(out, ol, self.TAB_CODE, self.TAB_CODE_LEN, state) # TAB
|
|
else:
|
|
ol = self.append_bits(out, ol, self.BIN_CODE_TASMOTA, self.BIN_CODE_TASMOTA_LEN, state) # Binary, we reuse the Unicode marker which 3 bits instead of 9
|
|
ol = self.encodeCount(out, ol, (255 - c_in) & 0xFF)
|
|
|
|
|
|
# check that we have some headroom in the output buffer
|
|
if ol // 8 >= len_out - 4:
|
|
return -1 # we risk overflow and crash
|
|
|
|
l += 1
|
|
|
|
bits = ol % 8
|
|
if bits:
|
|
ol = self.append_bits(out, ol, self.TERM_CODE, 8 - bits, 1) # 0011 0111 1100 0000 TERM = 0011 0111 11
|
|
return (ol + 7) // 8
|
|
# return ol // 8 + 1 if (ol%8) else 0
|
|
|
|
|
|
def getBitVal(self, inn, bit_no, count):
|
|
c_in = inn[bit_no >> 3]
|
|
if bit_no >> 3 and self.ESCAPE_MARKER == inn[(bit_no >> 3) - 1]:
|
|
c_in -= 1
|
|
r = 1 << count if (c_in & (0x80 >> (bit_no % 8))) else 0
|
|
#print("getBitVal r={r}".format(r=r))
|
|
return r
|
|
|
|
# Returns:
|
|
# 0..11
|
|
# or -1 if end of stream
|
|
def getCodeIdx(self, code_type, inn, len_, bit_no_p):
|
|
code = 0
|
|
count = 0
|
|
while count < 5:
|
|
if bit_no_p >= len_:
|
|
return -1, bit_no_p
|
|
# detect marker
|
|
if self.ESCAPE_MARKER == inn[bit_no_p >> 3]:
|
|
bit_no_p += 8 # skip marker
|
|
|
|
if bit_no_p >= len_:
|
|
return -1, bit_no_p
|
|
|
|
code += self.getBitVal(inn, bit_no_p, count)
|
|
bit_no_p += 1
|
|
count += 1
|
|
code_type_code = code_type[code]
|
|
if code_type_code and (code_type_code & 0x07) == count:
|
|
#print("getCodeIdx = {r}".format(r=code_type_code >> 3))
|
|
return code_type_code >> 3, bit_no_p
|
|
|
|
#print("getCodeIdx not found = {r}".format(r=1))
|
|
return 1, bit_no_p
|
|
|
|
def getNumFromBits(self, inn, bit_no, count):
|
|
ret = 0
|
|
while count:
|
|
count -= 1
|
|
if self.ESCAPE_MARKER == inn[bit_no >> 3]:
|
|
bit_no += 8 # skip marker
|
|
ret += self.getBitVal(inn, bit_no, count)
|
|
bit_no += 1
|
|
return ret
|
|
|
|
def readCount(self, inn, bit_no_p, len_):
|
|
(idx, bit_no_p) = self.getCodeIdx(self.us_hcode, inn, len_, bit_no_p)
|
|
if idx >= 1:
|
|
idx -= 1 # we skip v = 1 (code '0') since we no more accept 2 bits encoding
|
|
if idx >= 5 or idx < 0:
|
|
return 0, bit_no_p # unsupported or end of stream
|
|
till = 0
|
|
bit_len_idx = 0
|
|
base = 0
|
|
#for (uint32_t i = 0; i <= idx; i++) {
|
|
i = 0
|
|
while i <= idx:
|
|
# for i in range(idx):
|
|
base = till
|
|
bit_len_idx = self.bit_len[i]
|
|
till += (1 << bit_len_idx)
|
|
i += 1
|
|
|
|
count = self.getNumFromBits(inn, bit_no_p, bit_len_idx) + base
|
|
#print("readCount getNumFromBits = {count} ({bl})".format(count=count,bl=bit_len_idx))
|
|
|
|
bit_no_p += bit_len_idx
|
|
return count, bit_no_p
|
|
|
|
def decodeRepeat(self, inn, len_, out, ol, bit_no):
|
|
#print("decodeRepeat Enter")
|
|
(dict_len, bit_no) = self.readCount(inn, bit_no, len_)
|
|
dict_len += self.NICE_LEN
|
|
(dist, bit_no) = self.readCount(inn, bit_no, len_)
|
|
dist += self.NICE_LEN - 1
|
|
#memcpy(out + ol, out + ol - dist, dict_len);
|
|
i = 0
|
|
while i < dict_len:
|
|
#for i in range(dict_len):
|
|
out[ol + i] = out[ol - dist + i]
|
|
i += 1
|
|
ol += dict_len
|
|
|
|
return ol, bit_no
|
|
|
|
def decompress(self, inn, len_, out, len_out):
|
|
ol = 0
|
|
bit_no = 0
|
|
dstate = self.SHX_SET1
|
|
is_all_upper = 0
|
|
|
|
len_ <<= 3 # *8, len_ in bits
|
|
out[ol] = 0
|
|
while bit_no < len_:
|
|
c = 0
|
|
is_upper = is_all_upper
|
|
(v, bit_no) = self.getCodeIdx(self.us_vcode, inn, len_, bit_no) # read vCode
|
|
#print("bit_no {b}. v = {v}".format(b=bit_no,v=v))
|
|
if v < 0:
|
|
break # end of stream
|
|
h = dstate # Set1 or Set2
|
|
if v == 0: # Switch which is common to Set1 and Set2, first entry
|
|
(h, bit_no) = self.getCodeIdx(self.us_hcode, inn, len_, bit_no) # read hCode
|
|
#print("bit_no {b}. h = {h}".format(b=bit_no,h=h))
|
|
if h < 0:
|
|
break # end of stream
|
|
if h == self.SHX_SET1: # target is Set1
|
|
if dstate == self.SHX_SET1: # Switch from Set1 to Set1 us UpperCase
|
|
if is_all_upper: # if CapsLock, then back to LowerCase
|
|
is_upper = 0
|
|
is_all_upper = 0
|
|
continue
|
|
|
|
(v, bit_no) = self.getCodeIdx(self.us_vcode, inn, len_, bit_no) # read again vCode
|
|
if v < 0:
|
|
break # end of stream
|
|
if v == 0:
|
|
(h, bit_no) = self.getCodeIdx(self.us_hcode, inn, len_, bit_no) # read second hCode
|
|
if h < 0:
|
|
break # end of stream
|
|
if h == self.SHX_SET1: # If double Switch Set1, the CapsLock
|
|
is_all_upper = 1
|
|
continue
|
|
|
|
is_upper = 1 # anyways, still uppercase
|
|
else:
|
|
dstate = self.SHX_SET1 # if Set was not Set1, switch to Set1
|
|
continue
|
|
|
|
elif h == self.SHX_SET2: # If Set2, switch dstate to Set2
|
|
if dstate == self.SHX_SET1:
|
|
dstate = self.SHX_SET2
|
|
continue
|
|
|
|
if h != self.SHX_SET1: # all other Sets (why not else)
|
|
(v, bit_no) = self.getCodeIdx(self.us_vcode, inn, len_, bit_no) # we changed set, now read vCode for char
|
|
if v < 0:
|
|
break # end of stream
|
|
|
|
if v == 0 and h == self.SHX_SET1A:
|
|
#print("v = 0, h = self.SHX_SET1A")
|
|
if is_upper:
|
|
(temp, bit_no) = self.readCount(inn, bit_no, len_)
|
|
out[ol] = 255 - temp # binary
|
|
ol += 1
|
|
else:
|
|
(ol, bit_no) = self.decodeRepeat(inn, len_, out, ol, bit_no) # dist
|
|
continue
|
|
|
|
if h == self.SHX_SET1 and v == 3:
|
|
# was Unicode, will do Binary instead
|
|
(temp, bit_no) = self.readCount(inn, bit_no, len_)
|
|
out[ol] = 255 - temp # binary
|
|
ol += 1
|
|
continue
|
|
|
|
if h < 7 and v < 11:
|
|
#print("h {h} v {v}".format(h=h,v=v))
|
|
c = ord(self.sets[h][v])
|
|
if ord('a') <= c <= ord('z'):
|
|
if is_upper:
|
|
c -= 32 # go to UpperCase for letters
|
|
else: # handle all other cases
|
|
if is_upper and dstate == self.SHX_SET1 and v == 1:
|
|
c = ord('\t') # If UpperCase Space, change to TAB
|
|
if h == self.SHX_SET1B:
|
|
if 8 == v: # was LF or RPT, now only LF # pylint: disable=misplaced-comparison-constant
|
|
out[ol] = ord('\n')
|
|
ol += 1
|
|
continue
|
|
|
|
if 9 == v: # was CRLF, now RPT # pylint: disable=misplaced-comparison-constant
|
|
(count, bit_no) = self.readCount(inn, bit_no, len_)
|
|
count += 4
|
|
if ol + count >= len_out:
|
|
return -1 # overflow
|
|
|
|
rpt_c = out[ol - 1]
|
|
while count:
|
|
count -= 1
|
|
out[ol] = rpt_c
|
|
ol += 1
|
|
continue
|
|
|
|
if 10 == v: # pylint: disable=misplaced-comparison-constant
|
|
break # TERM, stop decoding
|
|
|
|
out[ol] = c
|
|
ol += 1
|
|
|
|
if ol >= len_out:
|
|
return -1 # overflow
|
|
|
|
return ol
|
|
|
|
# pylint: enable=missing-function-docstring
|
|
|
|
|
|
if __name__ == "__main__":
|
|
# pylint: disable=line-too-long
|
|
UNISHOX = Unishox()
|
|
BYTES_ = bytearray(2048)
|
|
INN = bytearray(b'ON Switch1#State==1 DO Add1 1 ENDON ON Var1#State==0 DO ShutterStop1 ENDON ON Var1#State==1 DO ShutterClose1 ENDON ON Var1#State>=2 DO Var1 0 ENDON ON Shutter1#Close DO Var1 0 ENDON ON Switch2#State==1 DO Add2 1 ENDON ON Var2#State==0 DO ShutterStop1 ENDON ON Var2#State==1 DO ShutterOpen1 ENDON ON Var2#State>=2 DO Var2 0 ENDON ON Shutter1#Open DO Var2 0 ENDON')
|
|
LEN_ = UNISHOX.compress(INN, len(INN), BYTES_, len(BYTES_))
|
|
print("Compressed from {fromm} to {to} ({p}%)".format(fromm=len(INN), to=LEN_, p=(100-LEN_/len(INN)*100)))
|
|
|
|
OUT = bytearray(2048)
|
|
LEN_ = UNISHOX.decompress(BYTES_, LEN_, OUT, len(OUT))
|
|
print(str(OUT, 'utf-8').split('\x00')[0])
|