81 Todo

Yann Collet edited this page 2019-10-17 14:45:14 -07:00

Table of Contents

• Research
• Unlikely

• lz4io : check for write error on close (/dev/full)
• Move default block mode to 64 KB linked blocks
  • definitely future (v2?), as it would break 3rd party implementations which only support independent blocks.
• replace ERROR_GENERIC with more specific error messages within lz4frame
• implement error codes within lz4.c
  • similar to lz4f or zstd
  • but seems very hard without breaking API. Might be a future (v2?) topic
• remove or reduce parsing restrictions in LZ4 decoder
  • but not in LZ4 encoder yet
• Add cygwin test on Appveyor
• add simple "one-pass" lz4frame decoder
• Fullbench : should work "out of the box", using synthetic data (datagen)
  • with configurable generated size

Research

• universal U16
• lz4 hc : special mode, to use less memory;
  • especially for external dictionary mode
  • end dictionary on 64KB border
• lz4frame decompression context : parameter to limit max possible memory ?
• Consider intermediate compression mode, like Carlo Medas 4mc
• Multi-threaded compression (lz4io) ?

Unlikely

• More efficient FCS content size format => 1 & 2 bytes
  • breaks format
  • start by enabling decoder, then a few versions later, the encoder as it is supposed to be triggered automatically