0.10.4

Changes in 0.10.4

• Update dependencies for security fixes

• Fix runners configuration

• Remove MacOS runner due to problems

0.10.3

Changes in 0.10.3

• Update dependencies for security fixes

0.10.2

Changes in 0.10.2

• Update dependencies for security fixes

0.10.1

Changes in 0.10.1

• Upgrade fuser to get rid of abandoned users dep

• Update dependencies

0.10.0

Changes in 0.10.0

• Upgrade to clap4

• Rework and simplify Cli

  We now use the clap derive module to simplify the Cli. Also, split and
  cat are now proper subcommands. Having them introduced as a non-optional
  flag before was a poor decision.

• Update dependencies

0.9.2

Changes in 0.9.2

• Update dependencies to get security fixes

• Avoid using deprecated fucntion mount_spawn

0.9.1

Changes in 0.9.1

Update dependencies to fix security issues. Furthermore:

• Replace dependency fuse with newer fuser

  fuser (https://github.com/cberner/fuser) is a more maintained and
  up-to-date fork of fuse (https://github.com/zargony/fuse-rs), which
  ensures smoother future development.

• Use running integer as file handle

  This way, we do not need to rely on the time package to give a pseudo
  unique value.

• Use running integer as inode

  This way, we do not need to rely on the time package to give a pseudo
  unique value.

• Remove unused time dependency

0.9.0

Changes in 0.9.0

• Check mirror and mountpoint for sanity

  • Mirror and mountpoint have to exist.

  • Mirror must not be in a subfolder of mountpoint, to avoid recursive
    mounts. This is also in accordance to EncFS.

• Run scfs by default to easen rapid development

  Since we build more than just one binary, cargo run does not know
  which one to call by default. For this case, there is the key
  "default-run", which tells cargo run which binary to use when no
  --bin flag is present.

• Notify main loop when filesystem is dropped

  The filesystem implements the Drop trait now, which makes it possible to
  run a function when the filesystem is unmouted in a way other than by
  terminating the main loop (most prominently by using umount directly).

  The previous situation was, when the filesystem was unmounted via
  umount, then the main loop would hang infinitely, because there was no
  way to notify the main loop. Now we send a quit signal when the
  filesystem is dropped, so the main loop can exit normally.

• Canonicalize paths

  Using absolute paths is necessary for a daemon, since a daemon usually
  changes its working directory to "/" so as to not lock a directory.

• Add daemon flag which puts program in background

  The daemonizing is done after the filesystem has been created, to let
  the initialization happen in foreground. This minimizes the time the
  daemon is running but the filesystem is not mounted yet.

• Add flag to create mountpoint directory

  The mirror will intentionally not be created, since the mount is
  readonly and a missing mirror directory is most likely a typo from the
  user.

• Add converter for symbolic quantities

  This converter will be used to calculate the blocksize for a SplitFS
  mount. The size can now be given as an integer or optionally with a
  quantifier like "K", "M", "G", and "T", each one multiplying the base
  with 1024.

0.8.0

Changes in 0.8.0

• Implement readlink

• Correctly handle symlinks

• Replace each metadata with symlink_metadata

  Symlinks should be presented as-is, so it should never be necessary to
  traverse them.

• Silently ignore unsupported filetypes

• Add convenience wrappers for catfs and splitfs

  With these wrappers, it is possible to mount the respective filesystem
  without explicitly specifying the mode parameter.

0.7.0

Changes in 0.7.0

• Make blocksize customizable

  It is now possible to use a custom blocksize in SplitFS. For example, to
  use 1MB chunks instead of the default size of 2MB, you would go with:

  scfs --mode=split --blocksize=1048576 <base directory> <mount point>
  

  Where 1048576 is 1024 * 1024, so one megabyte in bytes.

• Short circuit when reading a size of 0

• Do not materialize vector after each chunk

  This step was highly unnecessary anyway and it needlessly consumed time
  and memory. An Iterator can be flattened in the same way, but without
  the penalty that comes with materializing.

• Do not calculate size of last chunk to read

  By simply reading blocksize bytes and only taking size before
  materializing, we can save a lot of possible mis-calculation regarding
  the last chunk.

  We make use of two properties here:

  • Reading after EOF is a no-op, so using a higher number on the
    reading operation does not hurt.

  • The reading operations take only place once we materialize the byte
    array. So even if we issue to read much more bytes than necessary on
    the last chunk, it will not hurt, since we only take the correct
    number of bytes on materializing.

• Fix off-by-one error

• Correctly handle empty files

  Create at least one chunk, even if it is empty. This way, we can
  differentiate between an empty file and an empty directory.

• Add test suites to modules

  With automated tests we now can effectively check if new features work
  as intended and that they do not break existing code.