Compare commits

6 Commits
master ... dev

Author SHA1 Message Date
Daniel Dybing
a15ba67b1a feat: Optimize .t42 loading and improve decoder fidelity
All checks were successful
Build Linux / Build Linux (push) Successful in 1m29s
Build Windows / Build Windows (push) Successful in 4m46s
2026-02-21 20:44:26 +01:00
Daniel Dybing
18fef7b049 fix: Align CRC-16 calculation with ETSI EN 300 706 and improve retrieval
All checks were successful
Build Linux / Build Linux (push) Successful in 1m28s
Build Windows / Build Windows (push) Successful in 4m52s
2026-02-08 19:51:28 +01:00
9b846970b8 fix: Align CRC calculation with ETSI EN 300 706 standard
All checks were successful
Build Linux / Build Linux (push) Successful in 1m29s
Build Windows / Build Windows (push) Successful in 4m45s
2026-02-07 14:27:37 +01:00
de296b4711 fix(ci): Add robust apt flags for Windows build container
All checks were successful
Build Linux / Build Linux (push) Successful in 1m30s
Build Windows / Build Windows (push) Successful in 5m20s
2026-02-07 10:57:44 +01:00
84d1094d16 fix: Update CRC calc to use 0 init and full 16-bit stored value
Some checks failed
Build Linux / Build Linux (push) Successful in 1m27s
Build Windows / Build Windows (push) Failing after 17s
2026-02-07 10:47:29 +01:00
6a6df63980 feat: Add CRC checksum calculation and display
All checks were successful
Build Linux / Build Linux (push) Successful in 1m34s
Build Windows / Build Windows (push) Successful in 4m42s
2026-02-07 10:12:25 +01:00
5 changed files with 300 additions and 147 deletions

View File

@@ -10,8 +10,9 @@ jobs:
steps:
- name: Install Node.js
run: |
apt-get clean
apt-get update
apt-get install -y nodejs npm
apt-get install -y --fix-missing nodejs npm
- name: Checkout
uses: actions/checkout@v3

View File

@@ -5,69 +5,82 @@ from .models import Packet, Page, TeletextService
def load_t42(file_path: str, progress_callback: Optional[Callable[[int, int], None]] = None) -> TeletextService:
service = TeletextService()
if not os.path.exists(file_path):
return service
total_bytes = os.path.getsize(file_path)
# Each packet is 42 bytes
total_packets = total_bytes // 42
processed_packets = 0
# Magazine buffers: magazine -> {row_num: Packet}
magazine_buffers = {m: {} for m in range(1, 9)}
# Active page lookup: magazine -> Page object (for O(1) access)
active_pages = {m: None for m in range(1, 9)}
with open(file_path, 'rb') as f:
while True:
chunk = f.read(42)
if not chunk:
break
if len(chunk) < 42:
# Should not happen in a valid T42 stream, or we just ignore incomplete tail
break
if not chunk: break
if len(chunk) < 42: break
processed_packets += 1
if progress_callback and processed_packets % 100 == 0:
if progress_callback and processed_packets % 500 == 0:
progress_callback(processed_packets, total_packets)
packet = Packet(chunk)
service.all_packets.append(packet)
# Logic to group into pages.
# This is non-trivial because packets for a page might be interleaved or sequential.
# Standard implementation: Packets arrive in order. Row 0 starts a new page/subpage.
mag = packet.magazine
buffer = magazine_buffers[mag]
if packet.row == 0:
# Start of a new page header.
# Byte 2-9 of header contain Page Number, Subcode, Control bits etc.
# We need to parse the header to identify the page.
p_num, sub_code, control_bits, language = parse_header(packet.data)
# Header format (after Mag/Row):
# Bytes: P1 P2 S1 S2 S3 S4 C1 C2 ...
# All Hamming 8/4 encoded.
# Check Erase Page bit (C4 is bit 0 of control_bits)
erase_page = bool(control_bits & 1)
# For now, let's just create a new page entry for every Header we see,
# or find the existing one if we want to support updates (but T42 usually is a stream capture).
# If it's an editor file, it's likely sequential.
if erase_page:
magazine_buffers[mag] = {0: packet}
buffer = magazine_buffers[mag]
else:
buffer[0] = packet
p_num, sub_code, language = parse_header(packet.data)
# Create snapshot
new_page = Page(
magazine=mag,
page_number=p_num,
sub_code=sub_code,
control_bits=control_bits,
language=language
)
# Efficient cloning: use the existing Packet objects where possible,
# but we MUST clone the data bytearray if we plan to edit it later.
for r_num, pkt in sorted(buffer.items()):
# Create a new packet shell sharing the original_data but with its own data bytearray
cloned_pkt = Packet(pkt.original_data)
cloned_pkt.data = bytearray(pkt.data)
new_page.packets.append(cloned_pkt)
# Create new page
new_page = Page(magazine=packet.magazine, page_number=p_num, sub_code=sub_code, language=language)
new_page.packets.append(packet)
service.pages.append(new_page)
else:
# Add to the "current" page of this magazine.
# We need to track the current active page for each magazine.
# A simplistic approach: add to the last page added that matches the magazine ??
# Robust approach: Maintain a dict of current_pages_by_magazine.
active_pages[mag] = new_page # Update active page lookup
# Let's find the last page in service that matches the packet's magazine
# This is O(N) but N (pages) is small.
target_page = None
for p in reversed(service.pages):
if p.magazine == packet.magazine:
target_page = p
break
elif 1 <= packet.row <= 31:
# Update the running buffer
buffer[packet.row] = packet
# Update the active snapshot immediately
target_page = active_pages[mag]
if target_page:
# Update row in the current active page
found_row = False
for i, p in enumerate(target_page.packets):
if p.row == packet.row:
target_page.packets[i] = packet
found_row = True
break
if not found_row:
target_page.packets.append(packet)
else:
# Packet without a header? Orphaned. Just keep in all_packets
pass
return service
@@ -182,52 +195,51 @@ def decode_hamming_8_4(byte_val):
(((byte_val >> 7) & 1) << 3)
def parse_header(data: bytearray):
# Data is 40 bytes.
# Bytes 0-7 are Page Num (2), Subcode (4), Control (2) - ALL Hamming encoded.
# 0: Page Units (PU)
# 1: Page Tens (PT)
# Data is 40 bytes (after MRAG).
# Byte 0: Page Units (PU)
# Byte 1: Page Tens (PT)
# Byte 2: Subcode S1 (bits 0-3)
# Byte 3: Subcode S2 (bits 4-6), C4 (bit 7)
# Byte 4: Subcode S3 (bits 8-11)
# Byte 5: Subcode S4 (bits 12-13), C5 (bit 14), C6 (bit 15)
# Byte 6: C7-C10
# Byte 7: C11-C14 (C12-C14 are Language)
pu = decode_hamming_8_4(data[0])
pt = decode_hamming_8_4(data[1])
# Use BCD/Hex-like storage: High nibble is Tens, Low nibble is Units.
# This preserves Hex pages (A-F) without colliding with decimal pages.
# E.g. Page 1FF -> Tens=F(15), Units=F(15) -> 0xFF (255)
# Page 12E -> Tens=2, Units=E(14) -> 0x2E (46)
# Page 134 -> Tens=3, Units=4 -> 0x34 (52)
# 0x2E != 0x34. No collision.
# Page number: pt (tens), pu (units). 0x00 to 0xFF.
page_num = ((pt & 0xF) << 4) | (pu & 0xF)
# Subcode: S1, S2, S3, S4
# S1 (low), S2, S3, S4 (high)
# Subcode (13 bits)
s1 = decode_hamming_8_4(data[2])
s2 = decode_hamming_8_4(data[3])
s3 = decode_hamming_8_4(data[4])
s4 = decode_hamming_8_4(data[5])
# Subcode logic is a bit complex with specific bit mapping for "Time" vs "Subcode"
# But usually just combining them gives the raw subcode value.
# S1: bits 0-3
# S2: bits 4-6 (bit 4 is C4) -> actually S2 has 3 bits of subcode + 1 control bit usually?
# Let's simplify and just concat them for a unique identifier.
sub_code = (s1 & 0xF) | \
((s2 & 0x7) << 4) | \
((s3 & 0xF) << 7) | \
((s4 & 0x3) << 11)
sub_code = s1 | (s2 << 4) | (s3 << 8) | (s4 << 12)
# Control bits C4-C14
c4 = (s2 >> 3) & 1
c5 = (s4 >> 2) & 1
c6 = (s4 >> 3) & 1
# Control bits C12, C13, C14 are in Byte 8 (index 8)
# They determine the National Option (Language)
c_bits_2 = decode_hamming_8_4(data[8])
c_7_10 = decode_hamming_8_4(data[6])
c_11_14 = decode_hamming_8_4(data[7])
# Fix for Language Detection:
# It seems C12 and C13 are swapped in the Hamming decoding or file format relative to expected values.
# C12 is bit 0, C13 is bit 1.
# We swap them so D1 maps to C13 (Swedish bit) and D2 maps to C12 (German bit).
# Original: language = c_bits_2 & 0b111
# bitmask starting at index 0 for C4
control_bits = c4 | (c5 << 1) | (c6 << 2) | \
((c_7_10 & 0xF) << 3) | \
((c_11_14 & 0xF) << 7)
language = ((c_bits_2 & 1) << 1) | ((c_bits_2 & 2) >> 1) | (c_bits_2 & 4)
# Language (C12, C13, C14)
# c_11_14: bit 0:C11, bit 1:C12, bit 2:C13, bit 3:C14
language = (c_11_14 >> 1) & 0x7
return page_num, sub_code, language
return page_num, sub_code, control_bits, language
def save_tti(file_path: str, page: Page):
"""

View File

@@ -1,6 +1,13 @@
from dataclasses import dataclass, field
from typing import List, Optional
def decode_hamming_8_4(byte_val):
# Extract data bits: bits 1, 3, 5, 7
return ((byte_val >> 1) & 1) | \
(((byte_val >> 3) & 1) << 1) | \
(((byte_val >> 5) & 1) << 2) | \
(((byte_val >> 7) & 1) << 3)
@dataclass
class Packet:
"""
@@ -27,22 +34,6 @@ class Packet:
b2 = self.original_data[1]
# De-interleave Hamming bits to get M (3 bits) and R (5 bits)
# This is the "basic" interpretation.
# For a robust editor we assume the input T42 is valid or we just store bytes.
# But we need Mag/Row to organize pages.
# Decode Hamming 8/4 logic is complex to implementation from scratch correctly
# without a reference, but usually D1, D2, D3, D4 are at bit positions 1, 3, 5, 7
# (0-indexed, where 0 is LSB).
# Let's perform a simple extraction assuming no bit errors for now.
def decode_hamming_8_4(byte_val):
# Extract data bits: bits 1, 3, 5, 7
return ((byte_val >> 1) & 1) | \
(((byte_val >> 3) & 1) << 1) | \
(((byte_val >> 5) & 1) << 2) | \
(((byte_val >> 7) & 1) << 3)
d1 = decode_hamming_8_4(b1)
d2 = decode_hamming_8_4(b2)
@@ -74,9 +65,13 @@ class Page:
Can have multiple subpages.
"""
magazine: int
page_number: int # 00-99
sub_code: int = 0 # Subpage code (0000 to 3F7F hex usually, simplest is 0-99 equivalent)
language: int = 0 # National Option (0-7)
page_number: int # 00-99 (Hex storage: 0x00-0xFF)
sub_code: int = 0 # 13-bit subcode (0000 to 3F7F hex)
# Control bits C4-C14
control_bits: int = 0
language: int = 0 # National Option (0-7, from C12-C14)
packets: List[Packet] = field(default_factory=list)
@property
@@ -84,6 +79,112 @@ class Page:
# Format as Hex to support A-F pages
return f"{self.magazine}{self.page_number:02X}"
def get_control_bit(self, n: int) -> bool:
""" Returns value of control bit Cn (4-14) """
if 4 <= n <= 14:
return bool((self.control_bits >> (n - 4)) & 1)
return False
def set_control_bit(self, n: int, value: bool):
""" Sets value of control bit Cn (4-14) """
if 4 <= n <= 14:
if value:
self.control_bits |= (1 << (n - 4))
else:
self.control_bits &= ~(1 << (n - 4))
def calculate_crc(self) -> int:
"""
Calculates the CRC-16 checksum for the page.
According to ETSI EN 300 706 (Section 9.6.1 & Figure 13):
- G(x) = x^16 + x^12 + x^9 + x^7 + 1 (Poly 0x1281)
- Initial value: 0.
- Processed bits b8 to b1 (MSB first for stored bytes).
- Total 1024 bytes (32 packets * 32 bytes).
- Packet X/0: Bytes 14 to 37 (24 bytes) + 8 spaces.
- Packets X/1 to X/25: Bytes 14 to 45 (32 bytes).
- Packets X/26 to X/31: 32 spaces each.
"""
crc = 0
poly = 0x1281
# Helper to update CRC with a byte (MSB first)
def update_crc(c, val):
v = (val << 8) & 0xFFFF
for _ in range(8):
if (c ^ v) & 0x8000:
c = (c << 1) ^ poly
else:
c = c << 1
v <<= 1
c &= 0xFFFF
return c
# Organize packets by row
rows = {p.row: p for p in self.packets}
for r in range(32): # Process 32 slots (0-31)
start, end = 0, 0
padding = 0
if r == 0:
# Row 0: Bytes 14-37 (24 bytes)
start, end = 8, 32 # data[8..31]
padding = 8
elif 1 <= r <= 25:
# Rows 1-25: Bytes 14-45 (32 bytes)
start, end = 8, 40 # data[8..39]
padding = 0
else:
# Rows 26-31: 32 spaces each
padding = 32
# Process packet data if available
if r in rows and start < end:
p_data = rows[r].data
for i in range(start, end):
byte_val = (p_data[i] & 0x7F) if i < len(p_data) else 0x20
crc = update_crc(crc, byte_val)
elif start < end:
# Missing packet but slot exists
for _ in range(start, end):
crc = update_crc(crc, 0x20)
# Add padding for this slot
for _ in range(padding):
crc = update_crc(crc, 0x20)
return crc
def get_stored_crc(self) -> Optional[int]:
"""
Attempts to retrieve the stored CRC from Packet 27/0 if present.
Returns None if not found.
"""
# Look for Packet 27
for p in self.packets:
if p.row == 27:
# Check Designation Code (Byte 0)
try:
if len(p.data) >= 40:
b0 = p.data[0]
# Decode Hamming 8/4
designation = decode_hamming_8_4(b0)
# Packets X/27/0 to X/27/3 exist, but only X/27/0 has the CRC.
# We also check if b0 is raw 0 as a fallback for some captures.
if designation == 0 or b0 == 0:
# Packet 27/0
# Checksum is in bytes 38 and 39 (TBytes 44 and 45).
hi = p.data[38]
lo = p.data[39]
crc = (hi << 8) | lo
return crc
except:
pass
return None
@dataclass
class TeletextService:
"""

View File

@@ -214,10 +214,18 @@ class TeletextCanvas(QWidget):
painter.end()
return
# Draw each packet
# Initialize a grid of empty chars
grid = [None] * 26 # 0-25
# Check Control Bits for "Inhibit Display" (C10)
# In our bitmask (from parse_header):
# C4:0, C5:1, C6:2, C7:3, C8:4, C9:5, C10:6, C11:7, C12:8, C13:9, C14:10
inhibit_display = bool((self.page.control_bits >> 6) & 1)
if inhibit_display:
painter.setPen(Qt.GlobalColor.gray)
painter.drawText(10, 20, f"Page {self.page.full_page_number} - INHIBIT DISPLAY (C10 set)")
painter.end()
return
# Organize each packet by row
grid = [None] * 26 # 0-25
for p in self.page.packets:
if 0 <= p.row <= 25:
grid[p.row] = p
@@ -243,6 +251,10 @@ class TeletextCanvas(QWidget):
# Output mask for the next row
next_occlusion_mask = [False] * 40
# Check for Suppress Header (C7)
# C7:3, so bit 3 of control_bits
suppress_header = bool((self.page.control_bits >> 3) & 1)
# Default State at start of row
fg = COLORS[7] # White
bg = COLORS[0] # Black
@@ -272,29 +284,18 @@ class TeletextCanvas(QWidget):
for c in range(40):
x = c * self.cell_w
# If this cell is occluded by the row above, skip drawing and attribute processing?
# Spec says "The characters in the row below are ignored."
# Ideally we shouldn't even process attributes, but for simple renderer we just skip draw.
# However, if we skip attribute processing, state (fg/bg) won't update.
# Teletext attributes are serial.
# BUT, if the row above covers it, the viewer sees the row above.
# Does the hidden content affect the *rest* of the row?
# Likely yes, attributes usually propagate.
# But the spec says "ignored". Let's assume we skip *everything* for this cell visually,
# but maybe we should technically maintain state?
# For "Double Height" visual correctness, skipping drawing is the key.
# We will Process attributes (to keep state consistent) but Skip Drawing if occluded.
# Wait, if we process attributes, we might set double_height=True for the NEXT row?
# If this cell is occluded, it shouldn't trigger DH for the next row.
is_occluded = occlusion_mask[c]
# Decide byte value
if row == 0 and c < 8:
# Use generated header prefix
if row == 0:
if c < 8:
# Column 0-7: Header prefix
byte_val = ord(header_prefix[c])
elif suppress_header and c < 32:
# Column 8-31: Hide header if C7 set
byte_val = 0x20
else:
byte_val = data[c] if c < len(data) else 0x20
else:
byte_val = data[c] if c < len(data) else 0x20

View File

@@ -320,6 +320,16 @@ class MainWindow(QMainWindow):
lang_layout.addWidget(btn_set_lang)
right_layout.addLayout(lang_layout)
right_layout.addSpacing(10)
# CRC Checksum
crc_label = QLabel("CRC Checksum:")
right_layout.addWidget(crc_label)
self.lbl_crc_info = QLabel("Page: ----\nCalc: ----")
self.lbl_crc_info.setStyleSheet("font-family: monospace; font-weight: bold;")
right_layout.addWidget(self.lbl_crc_info)
right_layout.addStretch()
self.layout.addLayout(center_layout, 1)
@@ -346,6 +356,30 @@ class MainWindow(QMainWindow):
# Menus
self.create_menus()
def update_crc_display(self):
if not self.current_page:
self.lbl_crc_info.setText("Page: ----\nCalc: ----")
self.lbl_crc_info.setStyleSheet("font-family: monospace; font-weight: bold;")
return
calc_crc = self.current_page.calculate_crc()
stored_crc = self.current_page.get_stored_crc()
stored_str = f"{stored_crc:04X}" if stored_crc is not None else "----"
calc_str = f"{calc_crc:04X}"
# Highlight if match
if stored_crc is not None:
if stored_crc == calc_crc:
style = "font-family: monospace; font-weight: bold; color: green;"
else:
style = "font-family: monospace; font-weight: bold; color: red;"
else:
style = "font-family: monospace; font-weight: bold;"
self.lbl_crc_info.setStyleSheet(style)
self.lbl_crc_info.setText(f"Page: {stored_str}\nCalc: {calc_str}")
def update_language_label(self):
idx = self.canvas.subset_idx
if 0 <= idx < len(self.language_names):
@@ -490,8 +524,9 @@ class MainWindow(QMainWindow):
self.language_overrides[key] = idx
# Patch Row 0 packet data to persist language selection to file
# Language bits are in Byte 8 (Control Bits 2): C12, C13, C14
# We need to preserve C11 (bit 3 of encoded 4-bit val) which is "Inhibit Display" usually 0
# Language bits are in Byte 7 (Control Bits C11-C14)
# Byte 7 encoded structure: bit 0:C11, bit 1:C12, bit 2:C13, bit 3:C14
# National Option index corresponds to (C14 C13 C12)
# Find Row 0 packet
header_packet = None
@@ -500,36 +535,23 @@ class MainWindow(QMainWindow):
header_packet = p
break
if header_packet and len(header_packet.data) > 8:
if header_packet and len(header_packet.data) >= 8:
try:
old_val = decode_hamming_8_4(header_packet.data[8])
# Encoded nibble structure: D1(b0), D2(b1), D3(b2), D4(b3)
# D1 maps to C12
# D2 maps to C13
# D3 maps to C14
# D4 maps to C11
# Byte 7 contains C11, C12, C13, C14
old_val = decode_hamming_8_4(header_packet.data[7])
# io.py logic for reading:
# language = ((c_bits_2 & 1) << 1) | ((c_bits_2 & 2) >> 1) | (c_bits_2 & 4)
# i.e. Lang Bit 0 comes from D2, Lang Bit 1 comes from D1, Lang Bit 2 comes from D3
l0 = (idx >> 0) & 1 # C12
l1 = (idx >> 1) & 1 # C13
l2 = (idx >> 2) & 1 # C14
# So for writing:
# D1 = Lang Bit 1
# D2 = Lang Bit 0
# D3 = Lang Bit 2
l0 = (idx >> 0) & 1
l1 = (idx >> 1) & 1
l2 = (idx >> 2) & 1
d1 = l1
d1 = (old_val >> 0) & 1 # Preserve C11
d2 = l0
d3 = l2
d4 = (old_val >> 3) & 1 # Preserve C11
d3 = l1
d4 = l2
new_val = d1 | (d2 << 1) | (d3 << 2) | (d4 << 3)
header_packet.data[8] = encode_hamming_8_4(new_val)
header_packet.data[7] = encode_hamming_8_4(new_val)
self.set_modified(True)
self.status_label.setText(f"Language set to {self.language_names[idx]} (saved to header).")
except Exception as e:
@@ -712,6 +734,7 @@ class MainWindow(QMainWindow):
# Force redraw
self.canvas.redraw()
self.canvas.update()
self.update_crc_display()
self.status_label.setText(f"Pasted {modified_count} rows.")
self.push_undo_state() # Push state after paste? NO, before!
# Wait, usually we push before modifying.
@@ -815,6 +838,7 @@ class MainWindow(QMainWindow):
self.canvas.set_page(self.current_page)
self.canvas.redraw()
self.canvas.update()
self.update_crc_display()
def populate_list(self):
self.page_list.clear()
@@ -848,9 +872,21 @@ class MainWindow(QMainWindow):
self.subpage_combo.clear()
for i, p in enumerate(pages):
# Display format: Index or Subcode?
# Subcode is often 0000. Index 1/N is clearer for editing.
label = f"{i+1}/{len(pages)} (Sub {p.sub_code:04X})"
# Try to find the clock in Row 0 (last 8 characters)
clock_str = ""
for pkt in p.packets:
if pkt.row == 0:
# Bytes 32-39 of the 40-byte data are the clock
raw_clock = pkt.data[32:40].decode('latin-1', errors='replace')
# Strip parity from each char and filter non-printables
clock_str = "".join([chr(ord(c) & 0x7F) if 32 <= (ord(c) & 0x7F) <= 126 else " " for c in raw_clock])
break
label = f"{i+1}/{len(pages)} "
if clock_str.strip():
label += f"[{clock_str.strip()}] "
label += f"(Sub {p.sub_code:04X})"
self.subpage_combo.addItem(label, p)
self.subpage_combo.blockSignals(False)
@@ -875,6 +911,7 @@ class MainWindow(QMainWindow):
self.canvas.update()
self.update_language_label()
self.update_crc_display()
self.canvas.setFocus()
def insert_char(self, char_code):
@@ -899,6 +936,7 @@ class MainWindow(QMainWindow):
self.hex_input.setText(f"{val:02X}")
mode_str = "Graphics" if is_graphics else "Text"
self.mode_label.setText(f"Mode: {mode_str}")
self.update_crc_display()
def on_hex_entered(self):
text = self.hex_input.text()