xz/TODO

LZMA Utils TODO List
--------------------

Major missing features

    Memory limits in the command line tool apply only to compression.

    Threading support in the lzma command line tool is still primitive.
    It cannot split a file in pieces yet.

    The --list mode isn't implemented in the command line tool.

    Handling of Multi-Block Stream information should be separated
    from Stream encoder and decoder. Those would be useful to implement
    multi-threaded coding in applications.

    Buffer to buffer coding is not implemented in liblzma. Probably
    a naive version should be written first, which would simply wrap
    things around lzma_stream. Later, there should be separate buffer
    coding functions, that are slightly faster (less memcpy()) and
    have smaller memory usage than the functions using lzma_stream.

    libzfile is not implemented.

    LZMA filter doesn't support predefined history buffer.


Security

    Search for bugs, especially security related issues. Security is
    important in every piece of code in LZMA Utils, but it is extremely
    important in the decoder part of liblzma.

    Subblock: If there is LZMA as a Subfilter but without EOPM, can it
    trigger infinite loop when Subblock's "Unset Subfilter" flag is hit?

    Similarly, can LZ decoder get stuck in infinite loop if the next
    filter in the chain returns LZMA_STREAM_END but the decoded data
    doesn't allow finishing the LZ decoding?


Reliability

    Create a test suite to be run with "make check".

    Should we use strlimit() and getrlimit() for memory usage limitting?


Performance

    Benchmark the CRC code on non-x86 CPUs. Won't have huge effect on
    overall speed, but it would still be nice to know what algorithm
    variant is the best on different CPUs.


Third party support

    Add support for LZMA to various applications. This naturally requires
    cooperating with the authors of the specific applications.
      * GNU grep and GNU diffutils: BSD grep already uses zlib directly
        instead of ugly shell scripts. It would be nice to get similar
        feature into relevant GNU tools. With libzfile, multiple
        compression formats would be easy to support.
      * kioslave for KDE
      * Magic for the `file' command
      * GNU Midnight Commander
      * GNU Texinfo
      * The `man' command
      * Package managers

    Test the patches already written. The patches should be sent to
    upstream developers _once_ LZMA Utils APIs are stable enough (so
    people don't need to fix those patches all the time).

    Mandriva has quite a few patches. Some of them are OK, some need
    adapting for new LZMA Utils.


Documentation

    Revise the man page of lzma command line tool.

    If the Doxygen docs aren't enough, write good Texinfo manual for
    liblzma. It's been a long time I've even tried to build the Doxygen
    docs, so they may look quite bad at the moment.

    Document LZMA as an algorithm. It would be great to have detailed
    description of the algorithm in English. Many people think, that
    reading the source code is not the optimal way to learn how LZMA
    works.


Other

    Some things return LZMA_PROG_ERROR with invalid options, some
    LZMA_HEADER_ERROR. These must be checked carefully and made so
    that LZMA_HEADER_ERROR is used only when the given option could
    make sense in future version of libzma.

    lzma_restrict vs. restrict

    Usage of LZMA_RUN vs. LZMA_FINISH with Metadata coders.

    Port the Deflate implementation from 7-Zip into liblzma. 7-Zip's
    Deflate compresses better than zlib, gzip or Info-ZIP. I don't
    know if Deflate will be included in .lzma format (probably not),
    but it's still useful once we also add support for .gz file format.