The <ui:WixUI Id="WixUI_InstallDir" InstallDirectory="INSTALLFOLDER" />
element already defines WIXUI_INSTALLDIR (via the InstallDirectory attribute)
and ARPNOMODIFY (in the wixlib). Declaring them again causes WIX0091 errors.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Added WIXUI_INSTALLDIR property (required per WiX issue #7105)
- Changed RemoveFolder Id from "DesktopFolder" to "RemoveDesktopShortcut"
to avoid ID conflict with StandardDirectory element
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The WiX v4 extension path was hardcoded and didn't match the actual
installed location. WiX v4 allows referencing globally installed
extensions by name directly.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The installer directory already has files from the copy step, so
Expand-Archive needs -Force to overwrite/merge.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Changed WindowsTargetPlatformVersion from 10.0.22621.0 to 10.0 to
automatically use whichever Windows 10 SDK is installed on the build
machine. This fixes CI failures when the runner has a different SDK
version installed.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The solution file only has x64 configurations (Release-Full|x64,
Debug-Full|x64). The workflow was incorrectly trying to build with
Win32 platform which doesn't exist.
Changes:
- Platform=Win32 → Platform=x64
- Output path ./Release-Full/ → ./x64/Release-Full/
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Start new changelog section for unreleased changes. First entry is
the multi-page teletext extraction feature (#665) which allows
extracting multiple teletext pages simultaneously with separate
output files.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The page update logic at line 1029-1035 was incorrectly updating
tlt_config.page for all accepted pages, even in single-page auto-detect
mode. This caused the auto-detect logic at line 979 to be bypassed
because the first packet (even with an invalid page number like 0xFF)
would set tlt_config.page, preventing proper auto-detection.
The fix restricts the page update to multi-page mode only. In single-page
mode, tlt_config.page is set exclusively by:
1. User specification (--tpage option)
2. Auto-detect logic (first valid subtitle page found)
This fixes regression in SP Test 76 which uses sample
8c1615c1a84d4b9b34134bde8085214bb93305407e935edcdfd4c2fc522c215f.mpg
with --autoprogram --out=ttxt --latin1.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Implement support for extracting multiple teletext pages simultaneously,
with each page output to a separate file.
Changes:
- Support multiple --tpage arguments (e.g., --tpage 397 --tpage 398)
- Create separate output files per page with _pNNN suffix
(e.g., output_p397.srt, output_p398.srt)
- Maintain backward compatibility for single-page extraction (no suffix)
- Add per-page SRT counters for correct subtitle numbering
- Fix BCD to decimal page number conversion in telxcc.c
- Add --tpages-all mode support for auto-detecting all pages
Tested with 21 teletext samples from the sample platform, all passing.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Extends HEVC caption extraction support to MKV files.
Changes to matroska.h:
- Add hevc_codec_id constant for V_MPEGH/ISO/HEVC
- Add hevc_track_number field to matroska_ctx structure
- Add process_hevc_frame_mkv() function declaration
Changes to matroska.c:
- Detect HEVC tracks in parse_segment_track_entry()
- Modify parse_simple_block() to route HEVC tracks to HEVC processor
- Add process_hevc_frame_mkv() with is_hevc flag and store_hdcc() call
- Parse HEVCDecoderConfigurationRecord in parse_private_codec_data()
- Initialize hevc_track_number in matroska_loop()
- Update output messages to report HEVC tracks
Tested with HEVC MKV file - extracts 73 captions matching MP4 output.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
PR #1852 added HEVC caption extraction for MPEG-TS containers,
but MP4/MKV containers weren't supported. This adds HEVC support
for MP4 containers using GPAC.
Changes:
- Add HEVC subtype definitions (hev1, hvc1)
- Add process_hevc_sample() to parse HEVC NAL units and extract CC
- Add process_hevc_track() to iterate through HEVC track samples
- Detect and process HEVC tracks in processmp4()
- Add store_hdcc() call to flush buffered CC data after each sample
The key fix was adding store_hdcc() after processing each sample.
Without this, CC data was being parsed but never output because
store_hdcc() is normally called from slice_header() which is
AVC-only.
Closes#1690 (for MP4 containers)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Move all changes made after the 0.95 version bump (commit ee232b5)
to a new 0.96 section marked as "Unreleased".
This separates the released 0.95 content from ongoing development
work that will be included in the next release.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
OCR build fix:
- linuxdeploy was failing with "Invalid magic bytes in file header"
because it was passed the wrapper script instead of the actual binary
- When OCR is enabled, ccextractor is renamed to ccextractor.bin and
a wrapper script sets TESSDATA_PREFIX before executing the binary
- Now correctly passes ccextractor.bin to linuxdeploy when it exists
HardSubX build fix:
- Add libavdevice-dev to FFmpeg dependencies in CI workflow
- rusty_ffmpeg requires libavdevice which was missing
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Rewrites the AppImage build script to support three build variants
matching the Docker build options:
- minimal: Basic CCExtractor without OCR (smallest size)
- ocr: CCExtractor with OCR support (default)
- hardsubx: CCExtractor with burned-in subtitle extraction
Changes to build_appimage.sh:
- Add BUILD_TYPE environment variable to select variant
- Fix CMake options (was incorrectly using make flags)
- Bundle tessdata for OCR builds with wrapper script
- Create proper desktop file and icon handling
- Improve error handling and cleanup
New GitHub Actions workflow (build_appimage.yml):
- Builds all three variants on release
- Uploads AppImages as release assets
- Can be manually triggered for specific variants
- Caches GPAC build for faster CI runs
Usage:
./build_appimage.sh # Builds 'ocr' variant
BUILD_TYPE=minimal ./build_appimage.sh
BUILD_TYPE=hardsubx ./build_appimage.sh
Closes#1348🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When processing DVB subtitles from live streams or corrupted files,
the bitmap clipping operation can fail, resulting in a NULL pix object.
Previously, this would cause a fatal crash with "Failed to perform OCR -
Failed to get text" because the code continued to call TessBaseAPIGetUTF8Text
even when no image was set.
Changes:
- Handle cpix_gs == NULL by logging a message and returning NULL
(skip this bitmap) instead of continuing and crashing
- Change the fatal error when TessBaseAPIGetUTF8Text returns NULL
to a non-fatal skip, since this can happen with empty/invalid bitmaps
- Both cases now properly clean up allocated resources before returning
This allows CCExtractor to gracefully skip problematic subtitle frames
instead of crashing, which is especially important for live streams
where packet loss or discontinuities can occur.
Fixes#1010🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The plans/ directory is in .gitignore but these files were added
before that entry existed. Removing from tracking while keeping
files on disk.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
For clarity and consistency, use explicit i64 instead of c_longlong.
While c_longlong is 64-bit on all platforms, i64 is clearer and
follows the same pattern as the previous commit that removed c_long.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The C type 'long' has different sizes on different platforms:
- Linux: 64-bit
- Windows: 32-bit
This causes ABI mismatches when interfacing with Rust, since Rust's
c_long matches the platform's long size, but we were treating these
values as 64-bit throughout.
Changed the following fields from 'long' to 'int64_t':
- asf_constants.h: parsebufsize
- avc_functions.h: cc_databufsize, num_nal_unit_type_7, num_vcl_hrd,
num_nal_hrd, num_jump_in_frames, num_unexpected_sei_length
- ccx_decoders_608.h: bytes_processed_608
- ccx_demuxer.h: capbufsize, capbuflen
- lib_ccx.h: ts_readstream() return type, FILEBUFFERSIZE
- file_functions.c: FILEBUFFERSIZE definition
- ts_functions.c: ts_readstream() implementation
Also updated Rust code in common.rs to remove c_long casts, since
bindgen will now generate i64 for these fields.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The extern declaration for ccxr_add_current_pts used c_long, but the
actual implementation in time.rs uses i64. This caused an ABI mismatch
on Windows where:
- c_long = i32 (32-bit)
- i64 = 64-bit
On Linux both are 64-bit so it worked, but on Windows the type
mismatch could cause incorrect parameter passing.
Changes:
- Change extern fn declaration from c_long to i64
- Remove unnecessary cast (FRAME_DURATION_TICKS is already i64)
- Remove unused c_long import
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When using --goptime, timestamps were compressed to 00:00:01-02 instead
of actual GOP times (17:56:40-47). This was caused by conflicts between:
- GOP timing set from GOP headers (wall-clock time, e.g., 17:56:40)
- PES PTS timing (stream-relative time, e.g., 00:00:02)
The sync detection saw these as 64,598-second "jumps" and kept resetting
timing, corrupting the output.
Fixes:
1. Guard video PES timing in general_loop.c - skip set_current_pts and
set_fts when use_gop_as_pts == 1 to prevent PES PTS from overwriting
GOP-based timing
2. Disable sync check in ccextractor.c when use_gop_as_pts == 1 since
GOP time and PES PTS are in different time bases and sync detection
is meaningless
Test results:
- Before: 00:00:01,231 --> 00:00:01,729
- After: 17:56:41,319 --> 17:56:43,084
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
On Windows, c_long is i32 while on Linux it's i64. The function
ccxr_print_mstime_static expects i64, so casting to c_long caused
a type mismatch error on Windows builds.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The Rust FFI functions were using c_long for PTS/FTS timestamps, but:
- C code uses LLONG (int64_t, 64 bits on all platforms)
- Rust c_long is 32 bits on Windows, 64 bits on Linux
This caused timestamp truncation on Windows when PTS values exceeded
2^31 (~24 days at 90kHz), resulting in wrong subtitle timestamps.
For example, a file with Min PTS of 23:50:45 (7,726,090,500 ticks)
would have its PTS truncated, breaking the teletext delta calculation
that normalizes timestamps to start at 0.
Changes:
- ccxr_add_current_pts: pts parameter i64
- ccxr_set_current_pts: pts parameter i64
- ccxr_get_fts: return type i64
- ccxr_get_visible_end: return type i64
- ccxr_get_visible_start: return type i64
- ccxr_get_fts_max: return type i64
- ccxr_print_mstime_static: mstime parameter i64
- fts_at_gop_start: extern static i64
Fixes tests 18 and 19 on Windows CI which showed raw PTS timestamps
(23:50:46) instead of normalized timestamps (00:00:00).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>