v0.9.0 (April 3, 2024)

Generic N-in-the-bag pre-endgame, with N up to 6
- When a pre-endgame play shows up as a win for a given set of tiles in the bag, it is always a sure win in this case, provided we play our best endgame.
- For bag-emptying plays, it is always correct.
- For non-bag-emptying plays, it is almost always correct. A bug fix in the future will make it always correct.
  - The issue is if the bag is still not empty by the time it is our turn again. Macondo will not solve the inner/recursive pre-endgame exactly, but will instead optimize and find any win for ourselves. This is wrong.
- Overall it is very slow for N > 1, especially if we’re examining non-bag-emptying plays. See help peg for more options.
Endgame speedups of 2x or more in some cases:
- Endgame LazySMP implementation can now use 8-10 threads or so for best performance, due to aspiration search in multithreaded mode
- Added PVS/Negascout for an extra speed boost to the endgame
Many other bug fixes and improvements to memory usage, speed, etc.

v0.8.0 (Jun 26, 2023)

Made LazySMP implementation a bit better (can now use around 6 threads for best endgame performance)
Fully exhaustive 1-in-the-bag pre-endgame implementation.

Many bug fixes from the last refactor
Significant (can be 20X or more) speed increase in endgame engine for complex endgames.
Allow endgames to be multi-threaded. (Use up to 3-4 threads. Any more makes it slower).

Significant refactoring
Support multi-letter tiles (for Catalan)
Use KWG and KLV files instead of old gaddag/olv leave files
- Requires use of wolges for now (https://github.com/andy-k/wolges) for kwg/klv file generation
Speed up Monte Carlo sim by about 40%; make it zero-allocation

Add opp score to autoplay log
Add bingo tracking in autoplay log and stats
Create code for configurable pre-endgame heuristic values. Current heuristics are blank but can be iterated on.
Create utility functions to return all words created by a play, and to validate a play’s legality (not word legality but rule legality)
Fix a few GCG import and export issues

This is a code organization release with no new features.

protobuf files have been moved around for ease for importing by other projects.

Bug fix and enhancement release

Include NWL18 gaddag by permission
Changed the semantics of add by only allowing it to add on to a list of plays, but not commit the play
Added a commit command, which does what add used to do
Refactored game history with protobuf to fix various bugs and hacks. It is much cleaner now.
Added a lot of helpful options to autoplay command
Add a sim continue command to allow a sim to continue from where it left off
Add a sim trim command to delete bottom performers of a simulation.
Add an export command to allow saving to a .gcg file
add and commit commands are less picky about including the “through” letters in the play, as opposed to requiring a . character for these letters
Add a setlex command to set the lexicon of a game after loading it
Add color board display for terminals that support it
Fix bug where character encoding for GCGs containing CRLF characters was misparsed
Various other usability fixes

Bug fix release

Fixed shell sim command so we can do 3-ply and above
Display endgame results in a nicer format
Allow add exchange ABC and add pass commands to exchange and pass without move generation
Default to info-level logging (turn on debug with DEBUG=1 environment variable)
Keep text on board within a small horizontal range for smaller windows

The first pre-alpha version.