Skip to content
This repository has been archived by the owner on Oct 16, 2024. It is now read-only.

Release 3.2

Compare
Choose a tag to compare
@kad kad released this 30 Mar 18:35
  1. Multi-stream bzip2 archives are now supported. These are usually created
    with the 'pbzip2' compressor.

  2. LZO archives are now supported too. Files with the following extensions are
    recognized as LZO-compressed: ".lzo", ".tar.lzo", ".tzo".

  3. Make 'bmaptool create' (and hence, the BmapCreate module) work with the
    "tmpfs" file-system. Tmpfs does not, unfortunately, support the "FIEMAP"
    ioctl, but it supports the "SEEK_HOLE" option of the "lseek" system call,
    which is now used for finding where the holes are. However, this works only
    with Linux kernels of version 3.8 or higher.

    Generally, "FIEMAP" is faster than "SEEK_HOLE" for large files, so we always
    try to start with using FIEMAP, and if it is not supported, we fall-back to
    using "SEEK_HOLE". Therefore, the "Fiemap" module was re-named to "Filemap",
    since it is now supports more than just the FIEMAP ioctl.

    Unfortunately, our "SEEK_HOLE" method requires the directory where the image
    resides to be accessible for writing, because in current implementation we
    need to create a temporary file there for a short time. The temporary file
    is used to detect whether tmpfs really supports SEEK_HOLE, or the system
    just fakes it by always returning EOF (this is what happens in pre-3.8
    kernels).

  4. Decompression should now require less memory, which should fix
    out-of-memory problems reported by some users recently. Namely, users
    reported that decompressing large bz2-compressed sparse files caused
    out-of-memory situation on machines with 2GB RAM. This should be fixed now.

  5. Reading and decompressing is now faster because we now use more parallelism:
    reading the data form the source URL is done in separate thread,
    decompressing happens in a separate process too. My measurement with Tizen
    IVI images from 'tizen.org' showed 10% read speed improvement, but this
    depends a lot on where the bottle-neck is: the USB stick, the network, or
    the CPU load.