-- in Restore.pm:
      # TODO: inefficient!  we don't want to download the chunk from the
      # target multiple times.  better to cache it locally, or at least
      # only fetch a region from the target (but that's still kinda inefficient
      # and pushes complexity into the Target interface)

-- Optionally only use in-memory files when using encryption.

-- <lj user=grahams>: if a file which existed when Brackup was
  "discovering files" subsequently goes away while Brackup is working
  it's magic (like a vim .swp file) Brackup tosses an error and bails:

-- don't skip files ending in whitespace.  figure out why GPG barfs.  would
   the metafile also barf, having a trailing \r or \n?

-- figure out how files with intenal \r or \n in filename get written
   to metafile.  need to be escaped?

-- FUSE script to mount a *.brackup file

-- tool to clean digestcache, based on prefixes, looking for files that no longer
   exist or have new mtimes, etc?  or keep track of "last used" date
   field in the digestcache and just delete things that are too old?
   but then no longer just a dictionary.  SQLite would work, but we'd ideally
   like lots of dumber cache mechanisms.  maybe a ->clean method is optional?
   then a memcached backend/etc doesn't have to use it.

-- Tools to rebuild your inventory database from the target's enumeration
   of its chunks and the target's *.brackup metafiles isn't yet done, but
   would be pretty easy.

-- ionice stuff.  network nice stuff.

-- make tests pass without 'noatime' mount option

-- --ignore-debian-files option (if managed by a package management
   system (and not an unmodified conffile), don't back it up)

-- reuse tempfiles in Chunk.pm as Restore.pm does

-- restoring from existing config file.

-- better test coverage.  currently at 84%.  should ignore
   test coverage of test modules.  move Brackup::Test to
   t/lib/ probably

-- should do TODOs in code.  :)