mkvtoolnix/doc/development.md

File system layout

There's no dedicated include directory. Each include file is located in the same directory its accompanying cpp file is located in as well.

Directories in alphabetical order:

• ac: Macro snippets used for the configure script.
• contrib: Spec files for RPM creation and other contributed files.
• doc/guide: mmg's guide.
• doc/man: man pages. They're written as DocBook XML files and converted to HTML and man format automatically.
• examples: Examples for the end user.
• installer: Files for the Windows installer.
• lib: External libraries that are required for building. Should not be touched but can be used.
• po: Translation of the programs (not of the man pages).
• share: Image, desktop files etc.
• src/common: Common classes used by all tools. This library is linked to each program.
• src/extract: mkvextract's main files.
• src/info: mkvinfo's main files including its GUI.
• src/input: Readers (demux source files). Only used in mkvmerge.
• src/merge: mkvmerge's main files, e.g. command line handling, output control, file creating etc.
• src/mmg: mkvmerge's GUI.
• src/output: Packetizers (create the tracks in the output file). Only used in mkvmerge.
• src/propedit: mkvpropedit's main files.
• src/scripts and src/tools: Scripts and compiled tools that may or may not be useful during development.
• tests: Test suite (requires external data package not distributed freely)

Strings & encoding

Internally all strings are encoded in UTF-8. Strings must be re-coded during input/output.

Timecodes

All timecodes and durations are stored as int64_t variables with nanosecond precision.

Outputting messages to the user

There are basically four functions that output stuff: mxinfo(), mxverb(), mxwarn() and mxerror(). Each takes either a string or an argument produced by boost::format() (see the Boost::Format documentation if you're not familiar with it), and mxverb() also takes a verbosity level argument.

mxinfo() is supposed to be used for messages that are always output. Its messages must be translatable by using the _() macro, Example:

mxinfo(boost::format(_("Hello, %1%\n")) % user_name);

The same applies to mxwarn() and mxerror(). However, both prefix their messages with "Warning:" and "Error:" respectively. Also they modify mkvmerge's exit code -- "1" if mxwarn() has been used at least once and "2" if mxerror() is used. mxerror() also instantly exits the program. mmg shows such warnings and errors in separate areas as to draw the user's attention to them. Therefore mxwarn() should be used for stuff the user really should know, and mxerror() for stuff that prevents further processing altogether.

mxverb() is a debugging facility. mkvmerge starts out at verbosity level 1. Therefore mxverb(1, ...) is equivalent to mxinfo(...). You can raise the verbosity level by adding the "-v" or "--verbose" command line parameters multiple times. Messages output with mxverb() should not be translatable, meaning you shouldn't use the _() macro.

Classes

General

Most classes are nameed *_c, and the corresponding counted pointer implementation is called *_cptr. A counted pointer is a reference counter implementation that automatically deletes its instance once the counter reaches zero.

memory_c (from common/memory.h)

Stores a pointer to a buffer and its size. Can free the memory if the buffer has been malloced.

Allocating memory:

memory_cptr mem = memory_c::alloc(12345);

Copying/reading something into it:

memcpy(mem->get_buffer(), source_pointer, mem->get_size());
file->read(mem, 20);

mm_io_c (from common/mm_io.h)

Base class for input/output. Several derived classes implement access to binary files (mm_file_io_c), text files (mm_text_io_c, including BOM handling), memory buffers (mm_mem_io_c) etc.

mm_file_io_c (from common/mm_io.h)

File input/output class. Opening a file for reading and retrieving its size:

mm_file_io_cptr file(new mm_file_io_c(file_name));
int64_t size = file->get_size();

Seeking:

// Seek to absolute position 123:
file->setFilePointer(123, seek_beginning);
// Seek forward 61 bytes:
file->setFilePointer(61, seek_current);

Reading:

size_t num_read = file->read(some_buffer, 12345);

track_info_c (from src/merge/pr_generic.h)

A file and track information storage class. It is created by the command line parser with the source file name and all the command line options the user has passed in.

The instance is then passed to the reader which stores a copy in the m_ti variable. The reader also passes that copy to each packetizer it creates which in turn stores its own copy (again in the m_ti variable).

Readers

A reader is a class that demuxes a certain file type. It has to parse the file content, create one packetizer for each track, report the progress etc. Each reader class is derived from generic_reader_c (from src/merge/pr_generic.h).

An example for a rather minimal implementation is the MP3 reader in src/input/r_mp3.h.

Constructor & virtual destructor

The constructor usually only takes a track_info_c argument. A virtual destructor must be present, even if it is empty. Otherwise you'll get linker errors.

The constructor must open the file and parse all its headers so that all track information has been processed. This will most likely be split up into a separate function in the future.

read(generic_packetizer_c *ptzr, bool force = false)

Requests that the reader reads data for a specific track. mkvmerge uses a pulling model: the core asks each packetizer to provide data. The packetizers in turn ask the reader they've been created from to read data by calling this function.

If the reader supports arbitrary access to track data (e.g. for AVI and MP4/MOV files) then the reader should only read data for the requested track in order not to consume too much memory. If the file format doesn't allow for direct access to that data then the reader can simply read the next packet regardless which track that packet belongs to. The packetizer will call the reader's read() function as often as necessary until it has enough data.

The reader must return FILE_STATUS_MOREDATA if more data is available and FILE_STATUS_DONE when the end has been reached.

Each data packet is stored in an instance of packet_c. If the source container provides a timecode then that timecode should be set in the packet as well.

create_packetizer(int64_t track_id)

Has to create a packetizer for the track with the specific ID. This ID is the same number that the user uses on the command line as well as the number used in the call to id_result_track() (see below).

This function has to verify that the user actually wants this track processed. This is checked with the demuxing_requested() function. Example from the MP3 reader (as the MP3 file format can only contain one track the ID is always 0; see the Matroska reader for more complex examples):

// Check if the audio demuxer with ID 0 is requested. Also make
// sure that the number of packetizers this reader has created is
// still 0.
if (!demuxing_requested('a', 0) || (NPTZR() != 0))
  return;

// Inform the user.
mxinfo_tid(m_ti.m_fname, 0, Y("Using the MPEG audio output module.\n"));

// Create the actual packetizer.
add_packetizer(new mp3_packetizer_c(this, m_ti, mp3header.sampling_frequency, mp3header.channels, false));

A lot of packetizers expect their codec private data to be constructed completely by the reader. This often requires that the reader processes at least a few packets. For a rather complex case see the MPEG PS reader's handling of h264 tracks.

get_progress()

Returns the demuxing progress, a value between 0 and 100. Usually based on the number of bytes processed.

identify()

File identification. Has to call id_result_container("description") once for the container type.

For each supported track the function must call id_result_track(...) once.

Both the container and the track identification can contain extended information. For an extensive example see the Matroska reader's identification function in src/input/r_matroska.cpp.