TODO file for gzip.

Some of the planned features include:

- Structure the sources so that the compression and decompression code
  form a library usable by any program, and write both gzip and zip on
  top of this library. This would ideally be a reentrant (thread safe)
  library, but this would degrade performance. In the meantime, you can
  look at the sample program zread.c.

  The library should have one mode in which compressed data is sent
  as soon as input is available, instead of waiting for complete
  blocks. This can be useful for sending compressed data to/from interactive
  programs.

- Make it convenient to define alternative user interfaces (in
  particular for windowing environments).

- Support in-memory compression for arbitrarily large amounts of data
  (zip currently supports in-memory compression only for a single buffer.)

- Map files in memory when possible, this is generally much faster
  than read/write. (zip currently maps entire files at once, this
  should be done in chunks to reduce memory usage.)

- Add a super-fast compression method, suitable for implementing
  file systems with transparent compression. One problem is that the
  best candidate (lzrw1) is patented twice (Waterworth 4,701,745
  and Gibson & Graybill 5,049,881). The lzrw series of algorithms
  are available by ftp in ftp.adelaide.edu.au:/pub/compression/lzrw*.

- Add a super-tight (but slow) compression method, suitable for long
  term archives.  One problem is that the best versions of arithmetic
  coding are patented (4,286,256 4,295,125 4,463,342 4,467,317
  4,633,490 4,652,856 4,891,643 4,905,297 4,935,882 4,973,961
  5,023,611 5,025,258).

  Note: I will introduce new compression methods only if they are
  significantly better in either speed or compression ratio than the
  existing method(s). So the total number of different methods should
  reasonably not exceed 3. (The current 9 compression levels are just
  tuning parameters for a single method, deflation.)

- Add optional error correction. One problem is that the current version
  of ecc cannot recover from inserted or missing bytes. It would be
  nice to recover from the most common error (transfer of a binary
  file in ascii mode).

- Add a block size (-b) option to improve error recovery in case of
  failure of a complete sector. Each block could be extracted
  independently, but this reduces the compression ratio.

- Use a larger window size to deal with some large redundant files that
  'compress' currently handles better than gzip.

- Implement the -e (encrypt) option.
   
Send comments to Jean-loup Gailly <jloup@chorus.fr>.