-------------------------------------------------------------------------
-- ChangeLog
-------------------------------------------------------------------------

3.8 released February 17, 2009

- added Wietze Brandsma to THANKS
- revised config.vc
- revised NEWS
- restored missing mkmcpat.dsp, regress.cmd

3.8rc1 released February 6, 2009

- revisions to montecarlo.texi
- revised .dsp files
- Lee Fisher added to AUTHORS
- WINDOWS doc and install.texi revised per Lee Fisher's changes
- mkmcpat puts 16 items on a line instead of 8
- (unsigned char) and (unsigned short) explicit typings
- missing files added to EXTRA_DIST
- copyrights updated to 2009

3.7.13 released December 23, 2008

- restore cmake files that were missing from the 3.7.12 tarball
- bring copyright date to 2008
- doc revision
- be forgiving with replacement moves conflicting with unconditional analysis
- accept --enable-socket-support even if configure test fails
- add globals.c to mkmcpat for OS/X build

3.7.12 released April 7, 2008

- cmake configuration files for generating MSVC project files
- minor doc revision
- new options --mc-patterns, --mc-list-patterns, and --mc-load-patterns
- tunable local patterns for Monte Carlo simulations
- try harder to find the right move to make seki in find_moves_to_make_seki()
- better heuristics to avoid calling semeai reading in
  find_moves_to_make_seki() in hopeless cases
- never amalgamate dragons over three-liberty strings and sometimes not
  over two-liberty strings
- send_two_return_one() moved from utils.c to board.c and reimplemented
- new file engine/montecarlo.c
- new options --monte-carlo and --mc-games-per-level to use Monte Carlo UCT
  search on 9x9 and smaller boards
- call check_boardsize() from play_gmp.c to avoid crashing on bad board sizes
- new function check_boardsize()
- bugfix in handling of foreground color with ncurses
- orientation invariant board hashes added
- tuning
- add STS-RV semeai regression test suite
- new and revised tests

3.7.11 released November 1, 2007

- gtp-commands.texi is not built automatically
- updates texinfo.tex
- minor doc revision
- revised configure.in, aclocal.m4 and *Makefile.in for recent autotools
- removed acconfig.h
- don't play send-two-return-one in tactical reading attack moves
- bugfix in reevaluate_ko_threats() to avoid a crash
- more efficient aftermath moves
- version 3 of GPL and update copyright for 2007
- new static function remove_eye_filling_moves() in owl.c
- semeai reading revisions
- look more ambitiously for back-capturing moves in attack1()	
- generate cut move reasons for cutting moves found during find_connections()
- fix compilation with MAX_BOARD set to low boardsizes (>= 7 tested)
- performance tuning in do_owl_analyze_semeai()
- detect unconditionally meaningless moves
- get_lively_stones() revised to be valid also for stackp>0
- bugfix in compute_owl_escape_values()
- moves can be marked as known_safe_moves by patterns
- bugfix in compute_surroundings()
- owl tuning
- new and revised tests

3.7.10 released May 21, 2006

- gnugo.dsw and uncompress_fuseki.dsp revised
- configure.in revised to improve socket support detection on windows
- gameinfo_play_sgftree_rot() revised to fail gracefully
- new static really_do_trymove() and is_superko_violation()   in board.c 
- board hashes stored in the move history
- is_legal() no longer depends on allow_suicide
- new is_allowed_move() to determine legality for current suicide and ko rules 
- new global variables suicide_rule and ko_rule
- illegal moves replaced with passes near the end of do_genmove()
- reevaluate_ko_threats() revised
- under superko rules look for ko threats also if best move illegal
- new options --forbid-suicide (default) and --allow-all-suicide
- new options --simple-ko (default), --no-ko, --positional-superko, 
  and --situational-superko 
- is_allowed_move() used instead of is_legal() in play_gtp.c, play_ascii.c
  and play_solo.c  
- identify_thrashing_dragons() called earlier in make_dragons()
- do not play self-atari while splitting eyespaces in aftermath_genmove()
- in do_owl_attack(), don't consider send-two-return-one defender moves
- send-two-return-one moves not tried in find_more_attack_and_defense_moves() 
- send-two-return-one moves not tried in   find_worm_attacks_and_defenses()
- string connection bonus for ko captures reduced in move_connects_strings()
- maximum bonus from secondary move values limited to 0.4 points
- more legality checking in fill_liberty()
- detect_tactical_blunder() revised to avoid finding illegal defense moves
- semeai.tst reordered
- new option --limit-commands to regress.pike
- new testfile 13x13c.tst
- do not try to add an sgf circle for pass moves in replay_node()
- new enum same_dragon_value in owl.c
- new owl_defendpats.db classification "a" for connecting all friendly 
  stones to the goal
- owl tuning
- influence tuning
- tuning
- new and revised tests

3.7.9 released March 10, 2006

- new GTP command advance_random_seed <games>
- make update_random_seed() deterministic
- fix assertion failure in semeai()
- semeai attack and defense codes added to struct dragon_data2
- in estimate_territorial_value(), let stones have influence strength also
  when their safety depends on ko	
- unify internal handling of limit_search and restricted_genmove
- new function play_lib_n() in utils.c and autohelper functions
  oplay_lib/xplay_lib
- new static function modify_eyefilling_move() in owl.c
- do not reset permeability around dead stones
- do not consider moves playing into snapback in
  do_find_more_owl_attack_and_defense_moves()
- new function playing_into_snapback() in utils.c
- only call do_find_more_owl_attack_and_defense_moves() when stackp==0
- use .strategic_size in estimate_strategical_value()
- new field .strategic_size in dragon_data2, includes bonus for weak neighbors
- use gg_normalize_float2int() in valuate_combinable_pattern_chain() to
  address platform dependencies
- carefully select new owl goal origin
- allow excess strategical value for some single-worm dragons
- include played move in atari-atari blunder analysis
- score.c removed from sources
- owl tuning
- tuning
- new and revised tests

3.7.8 released January 30, 2006

- new global variable handicap
- chinese scoring adds the number of handicap stones in addition to the
  komi to white's score
- include handicap in sgf headers
- dead code removed from influence_evaluate_position() and the rest moved
  to game_status()
- metamachine enabled in ascii mode (as in GNU Go 3.6)
- configure option oracle renamed metamachine (as in GNU Go 3.6)
- new testfile gifu05.tst
- don't try large scale attack if original owl reading was expensive
- teach influence_score() about chinese rules
- copyright notices updated for 2006
- bugfix in unconditional status analysis
- let --capture-all-dead and --play-out-aftermath be effective also when
  playing out a game for scoring (e.g. --score finish)
- semeai analysis turned off when there are a huge number of dragons
- MAX_STRINGS value corrected
- new assertions that the next_string pointer hasn't overflowed
- hash flag bits and -H option removed
- default size of reading cache set to a fixed number of entries instead
  of a fixed amount of memory, improving platform independence
- update information about valid boardsizes in gnugo.pod
- bugfix for drawing 1x1 board in play_ascii.c
- regress.pike can run multiple engines in parallel (-j option)
- correction of color in order_moves() call in reading.c
- doc revision, updated instruction for setting up HTML regression views
- find_connection_moves() revised
- GTP command dragon_data fixed to give well-formed output also on empty
  board
- update regress.pl and regress.plx to count connection nodes instead of
  life nodes and expect reg_genmove instead of gg_genmove
- remove gnugo_play_sgftree(), gnugo_genmove(), gameinfo_load_sgfheader(),
  and gameinfo_play_move() from engine/interface.c
- 1D conversion of gnugo_play_move()
- new function get_sgfmove()
- let semeai code pass if no riskless move was found
- do not hardwire owl node limit for large scale attacks
- remove many unneeded configure tests
- READLINE compile (warning) fixes
- complete rework of clock.c, using level adaption logic from play_gtp.c
- use more tactical moves in do_owl_analyze_semeai()
- unused function transform2() removed
- report_worm() made static in worm.c
- hashdata_to_string() revised for consistency between 32 and 64 bit platforms 
- bugfixes in do_aftermath_genmove()
- adjust readconnect node limit when starting search at higher stackp
- new function get_depth_modification()
- removed functions set_temporary_depth_values()/restore_depth_values()
- always set dragon status to ALIVE if owl_status is UNCHECKED
- correct C-code footer in uncompress_fuseki.c for all values of
  NUM_HASHVALUES 
- play_replay() and replay_node() partially rewritten
- tuning
- new and revised tests
	
3.7.7 released October 9, 2005

- FSF address corrected in COPYING
- large scale improvements
- use signed char instead of char most of the time in the engine
- string_to_location(), debug_influence_move(), and report_worm()
  converted to 1D
- get_level() removed from genmove.c
- documentation update
- remove long deprecated command hurry from play_ascii
- use signed char in connected_components() in owl.c
- rotation support for --decide_* options removed
- remove trivial wrapper functions from engine/interface.c
- patterns/transpat.c, patterns/patlib.c, and patterns/patlib.h removed
- unused game records in regression/golois removed
- makevcdist.pl updated and re-run
- all Makefile.in regenerated
- debugboard traces from Makefile and configure removed
- endgame1.tst removed from distribution but not from CVS
- cutstone.tst, semeai2.tst, and *_rot.tst removed
- debugboard removed	
- support for pattern based reading (EXPERIMENTAL_READING) removed
- support for tree based pattern matching removed
- view.pike: support multiple tests as argument
- test revisions
- consistently initialize global variables in globals.c (not main.c)
- large scale code reorganized
- fullboard support (-f option) remove from mkpat
- atari_atari tuning
- endgame tuning
	
3.7.6 released September 21, 2005

- a few missing game records restored
- revised patterns.dsp and new uncompress_fuseki.dsp

3.7.5 released September 20, 2005

- revision to patterns.dsp (but more revision needed before VC will compile)
- send usage messages caused by bad input to stderr	
- new, larger fuseki database added as compressed files
- new files patterns/compress_fuseki.c and patterns/uncompress_fuseki.c
- use a fullboard hash for fuseki database patterns
- new array comulative_goal in local_owl_data
- don't count former owl goal dragon stones towards escape measure
- removed PROFILE_MATCHER
- handle CRLF newlines correctly in view.pike	
- new static function squeeze_moves() in reading.c
- fix komaster problem
- do not trust connection moves to be safe in
  estimate_territorial_value() 
- add random move value contributions also for pure ko threats
- new and revised tests

3.7.4 released June 12, 2005

- copyright notices updated for 2005 and new FSF address
- bugfix in find_backfilling_move in filllib.c
- make fast_defense() aware of under-the-stones tesuji
- make superstring_break_chain_moves() attack 3-liberty chains
- unused code to compute genus removed from make_dragons()
- owl tuning
- atari-atari tuning
- optics tuning
- tuning
- new and revised tests

3.7.3 released March 22, 2005

- special_rescue_moves() allows self atari moves which may set up a snapback
- increase depth after filling liberties in endgame_find_backfilling_dame()
- try harder to find safe backfilling moves in endgame_find_backfilling_dame()
- revised heuristics to find a font file in view.pike
- avoid crashing on bad engine paths when starting a new engine in view.pike
- move_connects_strings() revised
- outcommented patterns removed
- ignore off board vertices in openregion() in fuseki.c
- Revise diagonal value in topological_eye() to account for damezumari
- owl tuning	
- new and revised tests

3.7.2 released January 20, 2005

- gnugo.el: Grid display implemented
- gnugo.el: SGF handling improved
- gnugo.el: Undo and Redo related enhancements
- gnugo.el: Primitive edit mode
- gnugo.el: Regression view mode
- Doc revision (Emacs mode in using.texi)
- use print_numeric_influence() for printing escape influence
- take neighbouring half/false eyes into account when estimating lunch value
- do not trust strategic attack and defense moves to be safe when scoring
- thrash_around_helper() returns false if fuseki is disabled
- value_territory() revised to disable certain reductions of the territory
  value when scoring
- regress.pike revised so that previous output can be used as input
  with -f to rerun tests with unexpected results
- connection_value() revised
- use influence-based score estimate instead of Bouzy's algorithm
- new global variable white_score, black_score
- should_resign() now depends on optimistic_score instead of average_score
- added support for GTP over TCP/IP sockets on Windows
- new static function all_own_neighbors_inessential() in aftermath.c
- do_aftermath_genmove() revised
- breakage2tst.py now produces command line for regress.pike
- further revision of thrashing dragon heuristic
- try harder to find backfilling moves in fill_liberty()
- find_backfilling_move() in filllib.c revised
- test_attack_either() revised
- fixed anchor building of owl_attackpats.db repaired
- thrashing dragon heuristics revised
- readline support in ASCII mode
- liberties and neighbors list broken out of string_data array in board.c
  for speedup
- new functions remove_double_blocks(), check_double_block() to improve
  territorial evaluation at places where an intrusion is possible
- stop using worm data at stackp==0 in does_attack() and does_defend()
- concerns_noncritical_dragon() revised
- MIN_BOARD reduced to 1
- free memory allocated in a failed read of an sgf file
- extract_fuseki revisions: statistics on outcomes,
  popularity cutoffs based on unique players added to command line
  parameters, tweaks and bugfixes in game selection
- atari_atari moves considered blocking in estimate_territorial_value()
- MAX_LIBERTIES reduced to 8
- popgo() bugfix to repair sgf traces	
- do_owl_analyze_semeai() revised
- emacs variant of ascii interface removed
- give less bonus to connection moves when scoring
- influence tuning
- owl tuning
- tuning
- new testsuite tiny.tst
- new and revised tests
	
3.7.1 released November 23, 2004

- new function send_two_return_one() in utils.c
- don't consider send-two-return-one moves in
  do_find_more_owl_attack_and_defense_moves() 
- does_attack() made ko aware
- new GTP commands does_attack and does_defend
- sgfnode.c (readsgffile): Do not fclose stdin.
- gtp.texi (GTP command reference): Add @cindex.
- gtp-commands.sed: Use @verbatim. Tighten output. 
- gtp-commands.sed: Move `Function' field into @item line, sans field name.
- doc/ Makefile.am (gtp-commands.texi): Eliminate redundant use of cat(1).
- revision of gtp-commands.sed
- bugfix in main.c: remember to sgfFreeNode(gameinfo.game_record.root)
- allow PASS_MOVE in trymove() and tryko()
- do_trymove() and popgo() revised
- 'printfsgf -' supported from GTP	
- copyright notices added to xpm files
- gnugo.el brought up to Thi's gnugo.el-2.2.8
- new gtp_move_history() for use with gnugo.el
- minor doc revision
- capture_non_invincible_strings() and unconditional_life() moved from
  utils.c to unconditional.c
- resignation option reverted from GTP commands genmove_black and
  genmove_white
- new heuristic to reduce branching factor in connection reading.
- resign rather than pass when all own stones seem dead
- lively_dragon_exists() based on status rather than crude_status
- potential_moves[] array converted to 1D
- distance measures converted from floating to fixed point arithmetic
  in connection reading code.
- owl threats disabled as a command line option
- EXIT_SUCCESS and EXIT_FAILURE used in transpat.c and sgfgen.c
- revised interface to all genmove() type functions
- new global variable depth_offset
- only cache safe_move() results when reading depth is unmodified
- revision of interface/emacs.el
- new interface/xpms/*xpm, make-gnugo-xpms.el and gnugo-xpms.el
- break_chain2_moves() revised
- special_attack4_moves() revised
- defend_secondary_chain_moves() revised and split into two functions
- check_self_atari() in readconnect.c revised
- devalue tactical defenses of owl critical dragons when there are
  better owl defenses elsewhere 
- find_moves_to_make_seki() extended to examine the case of dead stones
  inside the eye of single-eyed dragons
- added support for GTP communication over TCP/IP socket;  GNU Go can
  both connect and listen
- bugfix in break_in_goal_from_str(): check pos+delta on board
- bugfix in sgfStartVariant
- constrain thrashing dragons more often
- hashing cleanups
- missing files added to EXTRADIST in Makefile.* in doc/ and regression/
- new file logo-36.pdf
- DFA micro optimizations
- Test results updated: cvs tag rel-3-7-1-pre-1
- use int arrays instead of char arrays to mark eyes in
  owl_determine_life() and dependent functions
- added .eps-version of new logo
- help string revised
- warnings fixed
- doc revisions
- TODO revised
- basic strategies to break mirror go implemented
- matcher_check verbosity refined
- new function are_neighbors()
- 9x9 opening moves tuned
- view.pike enhancement: switch to new test case
- debugging enhancement: owl goal marked in sgf trees
- new move reasons POTENTIAL_SEMEAI_ATTACK/DEFENSE to defer analysis of
  possible semeai winning moves until move valuation
- infrastructure to cut owl goal dragons
- new owl pattern class C: threaten to cut owl goal dragon
- view.pike enhancement: owl_does_defend/attack
- crash in owl_hotspots() fixed
- view.pike enhancement: directly start sgf viewer
- remove special case of CLASS_C in matchpat.c
- influence tuning
- owl tuning
- endgame tuning
- tuning
- implement persistent semeai caching
- don't crash on some .sgf-files from kgs
- NEWS updated
- refer to 3.6 instead of 3.4 in docs
- new and revised tests

3.5.10 released August 7, 2004

- move initalization of ..._vital_points to initialize_dragon_data()
- persistent caches reorganized to share common code

3.5.9 released July 18, 2004

- bugfix in read_eye()	
- bugfix in fast_defense()
- improved printing of hashvalues
- bugfix in atari_atari_blunder_size()	
- new static function dump_board_sgf() in printutils.c
- abortgo() no longer unwinds the stack but prints an sgf file
- new and revised tests
- doc revision

3.5.8 released June 8, 2004

- make `mkpat' detect duplicate pattern names
- duplicate pattern names renamed
- reduce DFA_MAX_MATCHED 20 times 
- calculate actual maximum dfa with new dfa_calculate_max_matched_patterns()
- remove unused configure option --experimental-semeai
- add missing output to --options
- new static function special_rescue4() called from defend2() in reading.c
- fast_defense() now heuristically looks for neighbors to capture
- don't restore random state in hash_init() and keyhash_init()
- major `view.pike' revision, added new "single window" mode
- break_chain2_moves() revised to generate more backfilling moves
- atari_atari_blunder_size() now returns multiple defenses, better verified
- remove %r support from gprintf, use result_to_string() instead
- double_atari_chain2() revised to generate more moves
- new argument recompute_dragons to owl_analyze_semeai_after_move()
- owl_mark_dragon() and init_owl() revised
- detect_tactical_blunder() revised
- value of tactical blunders doubled to twice the effective size of the string
- not a blunder if a string alive in seki is saved when outer liberties filled
- normalize apos, bpos in do_owl_analyze_semeai
- find_cuts() called from make_dragons() instead of make_domains()
- new compute_new_dragons() and new static join_new_dragons() in dragon.c
- new global array cutting_points[]
- cut field removed from struct eye_data
- INHIBIT_CONNECTION bit no longer used in the type field of struct eye_data
- find_cuts() no longer allowed to modify eyespaces
- bugfix in do_owl_analyze_semeai()
- tuning
- new and revised tests

3.5.7 released May 27, 2004

- minor revisions in score.c
- regress.pike measures uncertainty of moves
- global variable hashdata renamed to board_hash
- hashdata_compare() function removed
- fields in struct stats_data updated for new cache
- initialization of Zobrist random numbers in hash.c and cache.c unified
- new function hash_init_zobrist_array() in hash.c
- hashdata_remainder() never uses more than the 32 lowest bits of hash value
- hashdata_NULL, hashdata_clear(), and hashdata_init() macros removed
- any number of MIN_HASHBITS can now be used
- ROUTINE_COSTS revised
- find_connection_moves() revised
- always print copyright string on stdout
- do not include "Version" in the --version output
- old cache implementation removed
- ignore inessential critical dragons in in dilate_erode()
- TRACE_CACHED_RESULT* macros implemented for new caching scheme
- old cache implementation commented out
- new file engine/unconditional.c added
- break-in reading tuned
- safety checks improved for owl defense moves
- bugfix in find_neighbor_dragons()
- attribute "noreturn" for abortgo()
- don't add child to root node in sgffile_printboard()
- remove color parameter from examine_position(), compute_eyes_pessimistic()
- new arrays white_vital_points and black_vital_points
- read_eye() stores vital point in new arrays when add_moves == 1
- new functions find_lunches(), revise_inessentiality(), eye_computations()
  broken out of make_dragons()
- VITAL_EYE_MOVE almost retired
- fix bugs in recursive_break()
- owl tuning
- new and revised tests

3.5.6 released April 23, 2004

- new configure and runtime option large-scale
- new change_owl_nodes()
- gprintf supports %r
- cleanup of patterns/*.sgf
- revision of dragon_invincible()
- suppress copyright notice in ASCII playing mode or with --version
- copyright string shortened
- new GTP commands move_probabilities and move_uncertainty
- bugfix in unconditional_life()
- owl_update_goal() overrides same_dragon == 0 in obvious cases
- new and revised tests

3.5.5 released April 15, 2004

- moved TRACE_READ_RESULTS definition to hash.h
- initialize lunch data for single owl move
- bugfix in compute_eyes_pessimistic(): initialize chainlinks[]
- improved sgf trace in trymove: location of tried move noted
- complete conversion to new cache
- add priority to reading results in new cache
- kill unused hash_ng_init()
- minor statistics improvements
- enlarge owl goal when it has had lunch
- komaster and kom_pos made static variables in board.c
- edge_block_moves() proposes more moves
- don't try to undo when stackp is not zero (GTP mode)
- support for multiple engines in view.pike
- more info about semeais in dragon_data2
- double-check owl attack moves against dragons involved in semeai(s)
- be less optimistic about large, but bulky eyespaces
- support for 'make gnugo.pdf' and make 'gnugo.html' in doc/
- slowest move statistics from regress.pike
- update usages of time_report()
- `twogtp.pike' can now use third engine (arbiter) to score games
- breakin and readconnect fixes
- first implementation of owl threat combinator--intended for improving
  move ordering
- pattern attributes (shape, followup etc.) are now stored in lists, not
  in `pattern' structure
- there is now a possibility to define virtually unlimited number of new
  pattern attributes
- common code of shapes_callback() and joseki_callback() is broken into
  new static function handle_joseki_patterns()
- new function do_find_more_owl_attack_and_defense_moves() broken out
  of find_more_owl_attack_and_defense_moves()
- induce_secondary_move_reasons() tries to upgrade induced reasons to
  owl attacks/defenses with do_find_more_owl_attack_and_defense_moves()
- strategical attack move reason is induced for saving certain worms
  adjoining weak opponent dragons
- major improvement of analyze_eyegraph functionality in optics.c
- adjust min_level and max_level if --level is set outside their range
- do not malloc() 0 bytes in pattern_list_build_heap()
- aftermath_final_status() declaration moved from gnugo.h to liberty.h
- bugfix in find_connection_moves()
- bugfix in hashdata_init macro
- caching enabled for unbranched owl nodes
- comments added explaining heap in readconnect.c
- provide links to html regression views from regress.sh
- semeai analysis in make_dragons() postponed until after revision of
  inessentiality 
- identification of thrashing dragons in make_dragons() postponed
  until after semeai analysis 
- avoid overflow in reading_cache_init()
- converted owl to use new cache implementation
- owl tuning
- new game records and test cases

3.5.4 released January 23, 2004

- copyright 2004 added to copyright notices
- new static function capture_non_invincible_strings() broken out of 
  unconditional_life()
- unconditional_life() revised
- new semeai_propose_eyespace_filling_move() tries to play inside eyeshapes
- number_of_dragons initialized in initialize_dragon_data()
- new option --gtp-dump-commands to dump all incoming GTP commands to file
- new function and autohelper owl_lunch()
- lunches_are_current field no longer reset in owl_safe_move()
- don't claim levels higher than 10 are unsupported
- new GTP command unconditional_status
- estimate_lunch_eye_value() revised	
- avoid calling owl_substantial() when not needed in aa_compute_status()
- value_move_reasons() revised
- special_rescue6_moves() revised
- find_connection_moves() revised
- computation of NUM_HASHVALUES corrected in hash.h
- hashdata_to_string reimplemented
- defend4() and special_rescue_moves() revised
- avoid ko captures when filling outer liberties in semeai reading
- adjust breakin_depth in modify_depth_values()
- atari_atari tuning
- endgame tuning
- eye tuning
- fuseki tuning
- influence tuning
- owl tuning
- tuning
- new test suite unconditional.tst
- new game records and test cases

3.5.3 released January 7, 2004

- new GTP command kgs-genmove_cleanup
- compute_crude_status() revised to consider dragons with big
  territorial moyo value as alive
- semeai moves considered safe in mark_changed_dragon
- compare number of eyes when both colors pass in
  do_owl_analyze_semeai() instead of always consider it a seki
- do not use the attack point from owl if owl considers a dragon dead
  but semeai revises it to critical
- bugfix in break_through_helper() in reading.c
- new static function close_enough_for_proper_semeai() in semeai.c
- liberty_of_dragon() transformed into neighbor_of_dragon() in semeai.c
- in semeai(), be more careful when considering two dragons involved
  in a semeai
- new static function max_lunch_eye_value() in value_moves.c
- when restraining thrashing dragons, do not bother connecting
  inessential strings which cannot provide an eye (in
  estimate_strategical_value()) 
- new autohelper function adjacent_to_stone_in_atari
- new static bamboo_rescue_moves() in `reading.c'.
- computation of safety and status values in make_dragons() revised to
  more robustly handle invincible dragons
- dragon_invincible() revised
- detect_tactical_blunder() revised
- owl stack management reimplemented
- consider semeai worms adjacent to the outside as important in
  owl_analyze_semeai_after_move()
- additional semeai debug output
- let stdout be unbuffered instead of line buffered in play_gtp()
- if one dragon is considerably smaller than the other in a semeai,
  prefer to get ko over seki in do_owl_analyze_semeai()
- in a semeai, consider also the eyespaces including opponent semeai
  worms in order to detect a definite lack of eyespace and to find
  certain vital moves
- off-by-one error in loops over semeai_worms[] corrected
- don't allow "safe" common liberties to depend on ko in
  do_owl_analyze_semeai() 
- new static function find_moves_to_make_seki() in semeai.c
- do semeai analysis for all pairs of critical/dead dragons except
  when one is inessential, other exceptional cases removed from semeai()
- defend4() generates superstring moves
- semeai_move_reasons() revised
- induce_secondary_move_reasons() revised; do not allow owl attack moves
  to induce strategic defense of neighbors if involved in a semeai
- connection_value() revised to never give points for connection to
  inessential strings
- get_influence() revised
- list_move_reasons() revised and made global
- new functions print_all_move_values() and prepare_move_influence_debugging()
- delta_territory cache can be disabled in estimate_territorial_value()
- new GTP commands all_move_values, eye_data, half_eye_data, move_reasons
- GTP command influence replaced by initial_influence, move_influence, and
  followup_influence
- GTP command clear_cache now also clears the non-persistent cache
- bugfix in corner_add_pattern() (mkpat.c): labels must be rotated too
- spread_connection_distances() now delays expensive constraint evaluation,
  storing pending evaluation on a heap
- new function signed_mark_string()
- computation of active area revised in store_persistent_connection_cache(), 
  store_persistent_breakin_cache(), and store_persistent_owl_cache()
- move_is_marked_unsafe() in move_reasons.h revised
- modify_stupid_eye_vital_point() revised
- influence tuning
- fuseki tuning
- eye tuning
- endgame tuning
- owl tuning
- tuning
- new test suite seki.tst
- new game records and test cases

3.5.2 released November 5, 2003

- Cosmic GNU Go disabled by default.
- report_dragon() revised
- gtp_worm_data() revised
- pattern tagging
- regress.sh revised
- regress.pike revised
- attack2() revised
- defend4() revised
- slow sorting in get_next_move_from_list() is replaced with heap sorting
- new static functions pattern_list_build_heap() and
  pattern_list_get_next_pattern() in `engine/owl.c'
- new command line option --no-cosmic-gnugo
- remove global variables score, lower_bound, upper_bound
- remove (level >= 8) condition for estimate_score() call
- pass our_score instead of score to the move valuation functions
- tuning
- influence tuning
- owl tuning
- joseki tuning
- new game records and test cases

3.5.1 released September 4, 2003

- semeai() revised to better handle certain cases of multiple simultaneous
  semeais 
- find additional semeai moves in semeai_reasons()
- added more robust handling of .dtr files
- removed option to build pre-rotated DFAs
- unconditional_status in struct worm_data changed to enum dragon_status
- final_status in the aftermath code changed to use enum dragon_status
- enum dragon_status extended with BLACK_TERRITORY, WHITE_TERRITORY,
  and DAME
- new GTP command last_move
- threaten_to_capture_helper() and pattern `Sente12' retired
- no buffering on stdout in ascii mode
- roadmap to engine files in overview.texi updated
- move DEBUG_func to showbord.c
- pattern tuning: remove "t" classification from various patterns
- WHITE_BORDER, BLACK_BORDER, GRAY_BORDER retired 
- colors and komaster state, safety values and dragon status, cache routine
  ids are now "enum"
- board library: remove all stuff that is not directly related to board
  (and move it to the engine)
- board library: new file board.h as public interface, new file boardlib.c
  to collect global variable of the board library;
- board library documentation updated to reflect changes
- new function: hashdata_to_string()
- small_semeai() and small_semeai_analyzer() removed
- support for !experimental_semeai removed
- old semeai() removed and new_semeai() renamed to semeai()
- analyze_semeai(), add_appropriate_semeai_moves(), and liberty_of_dragon()
  removed
- old amalgamation (!experimental_connections) code and patterns removed
- dragon_eye() removed
- basic_cut_helper() and ugly_cutstone_helper() removed
- komaster schemes 1-4 removed
- komaster_is_empty() retired
- komaster_to_string() revised
- KOMASTER_TRACE functionality removed
- new GTP command combination_defend
- revised hash table scheme
- new COSMIC_GNUGO function choose_best_strategy() sets tuning parameters
- new global tuning parameters: minimum_value_weight, maximum_value_weight,
  invasion_malus_weight, strategical_weight, territorial_weight,
  attack_dragon_weight;
- remove unneeded configure stuff, add -Wundef to SOME_WARNINGS
- replace setlinebuf by setvbuf
- add safety_hash field to delta territory cache, reenable this cache
- cgi interface in interface/html directory removed
- owl tuning
- new game records and test cases

3.4 released July 31, 2003

- breakage2tst.py and gnugo-db.el added to EXTRA_DIST
- wariuchi fuseki: invade between shimaris
- new sgmp mode
- regress.pike added to EXTRA_DIST
- revision of copyright notices
- warnings fixed on some platforms
- doc revision

3.3.23 released July 15, 2003

- bugfix in show_dragons()
- redistribution consistency checking disabled for stable release
- gg_genmove replaced by reg_genmove almost everywhere in the regressions
- twogtp.py and twogtp.pike added to gtp_examples/README
- twogtp and twogtp-a made GTP version 2 compliant
- break-in code turned off below level 10; other level cutoffs revised
- support for ranges in sgf files
- time summaries reported with --showtime in gmp mode
- TODO revised.
- clock_init_autolevel() retired
- new global variables min_level and max_level
- new options --min-level and --max-level
- time handling implemented in play_gtp.c
- speedup: move some functionality of update_liberties() to do_remove_string()
- bugfix in do_play_move()
- doc: delete most lists of functions
- doc: remove move_generation.texi overview sections
- doc: reorder
- doc: Consistently use hierarchical structure instead of next/previous/up
- tuning
- owl tuning
- eye tuning
- doc: complete rewrite of first sections of overview.texi
- doc: subsection on 1D board array moved to board.texi
- doc: references to cavities in dragon amalgamation removed
- new game records and test cases

3.3.22 released June 28, 2003

- break-in code turned off below level 9
- VPATH modifications in regression/Makefile*
- new function thrash_around_helper()
- thrashing patterns in patterns.db
- configure --enable-cosmic-gnugo produces a center-oriented engine
- added cache for results of approxlib() and accuratelib()
- new static functions do_approxlib() and do_accuratelib() in board.c
- finish_sgftrace() will not free the trace tree if not written successfully
- remove tabs from twogtp.py
- twogtp.py checks that board is not empty before indexing it
- active area for persistent breakin cache shrinked
- new HIGH_LIBERTY_BIT2 in persistent caches, used by breakin cache
- new function break_chain4_moves()
- owl moves for non-critical dragons discarded
- do_owl_analyze_semeai() revised
- detect_owl_blunder() also looks for semeai defenses
- estimate_strategical_value() revised
- new function defend_secondary_chain_moves() in reading.c
- new parameter be_aggressive to break_chain2_moves(),
  break_chain2_defense_moves() and break_chain3_moves()
- new depth value: break_chain_depth
- defend2() no longer plays on self atari liberties
- superstring_breakchain_moves() renamed superstring_break_chain_moves()
- fix "no moves found" logic in various readconnect.c functions
- new debug flag DEBUG_BREAKIN
- "smaller goal" in breakin.c revised
- be more careful when erasing territory after successful break-in
- break-in move generation bug fix
- endgame_find_backfilling_dame() revised
- new global array forced_backfilling_moves[]
- find_backfilling_move in filllib.c revised
- disable revise_semeai() when doing_scoring
- no bonus for attack or defense of dead potential cutstones when doing_scoring
- reevaluate_ko_threats() revised to avoid playing losing ko threats
- RTRACE and VTRACE retired
- corrected format of AP property in sgf_write_header()
- tuning
- owl_tuning
- new game records and test cases

3.3.21 released June 3, 2003

- revision of owl_strong_dragon()
- extra care is now taken not to waste points on tiny ko threats
- ko threats with large followups always preferred to large reverse followups
- twogtp.py fixed to handle jigo in endgame contests
- made --enable-experimental-break-in the default
- compute_smaller_goal() makes the goal even smaller in the block_off case
- persistent caching for break_in()/block_off()
- inessential dragons handled differently in estimate_territorial_value()
- use smaller goal area when trying to break into territories
- semeai_trymove_and_recurse() propagetes back changes to &owla, &owlb
- copy over old eye_data in do_owl_defend() if escape > MAX_ESCAPE
- new configure option --experimental-break-in
- load_and_score_sgf_file() revised
- detect_tactical_blunder() revised
- ko result codes no longer discarded in owl_analyze_semeai()
- new_semeai() revised
- decide_semeai() in sgfdecide.c revised
- GTP command owl_analyze_semeai renamed to analyze_semeai
- new function owl_analyze_semeai_after_move() and corresponding GTP
  command analyze_semeai_after_move
- gtp_owl_analyze_semeai() revised
- new static function semeai_trymove_and_recurse() in owl.c
- all owl_analyze_semeai regression test cases changed to analyze_semeai
  and revised for new result format
- bugfixes in breakin.c and block_off()
- new helper functions goal_to_hashvalue(), xor_hashvalues()
- new function get_read_result_hash_modified to enable calling functions
  to use a modified hash value when storing/looking up results in cache
- results for recursive_break() and recursive_block() now get cached
- cast to avoid warnings
- documentation revisions

3.3.20 released June 3, 2003

- hoshi.sgf split into hoshi_keima.sgf and hoshi_other.sgf
- new functions recursive_break(), recursive_block() in readconnect.c:
  decide whether a string can be connected to an arbitrary goal area
- new functions break_in(), block_off(), frontends to the functions above
- corresponding GTP commands break_in, block_off
- new file breakin.c evaluates whether territories can be broken in
- new command line --experimental-break-in to evaluate territory break-ins
- retire move_safety check in value_moves.c for territorial move valuation
- read_eye() revised
- `num_defends' field of half_eye_data strtucture renamed `num_defenses'
- maximum 15 points of territorial_value will get shape bonus
- change of semantics for --seed option
- new GTP command printsgf
- debugboard repaired
- TODO revised
- new function partition_eyespaces()
- new function owl_find_relevant_eyespaces()
- cpu time support for regress.pike
- broke new spread_connection_distances() out of compute_connection_distances()
- new spread_connection_distances() tuned
- t moves devalued to value 16
- new function edge_distance() in board.c
- CPU time reported by twogtp.py
- new static function endgame_find_backfilling_dame() in endgame.c
- new static function endgame_find_liberties() broken out of
  endgame_analyze_worm_liberties()
- new owl_mineye() and owl_maxeye() and corresponding autohelper functions
- move_connects_strings revised not to try connecting invincible strings
- ALL_MOVE move reasons no longer considered to guarantee a safe move
- never give cutstone bonus for an inessential string
- new test suite joseki.tst
- do_push_owl() doesn't copy unnecessary fields of local_owl_data anymore
- retire special meaning of owl pattern values 99 and 100
- bugfix in revise_semeai()
- find_half_and_false_eyes() bugfixed
- improve_lunch_attack() revised
- D219 pattern retired
- new pattern action good_attack_threat overrides heuristics in value_moves.c
- new global known_good_attack_threats[][]
- new functions register_good_attack_threat(), is_known_good_attack_threat()
- make delta territory cache aware of base influence recomputation
- owl_find_lunches() revised
- owl_determine_life() bugfixed
- owl.c made completely 1D and other revisions
- new GTP command analyze_eyegraph
- new functions analyze_eyegraph(), evaluate_eyespace(), tactical_life(),
  tactical_life_attack(), and tactical_life_defend() in optics.c
- new functions set_/get_/update_/reuse_random_seed()
- save and restore random_seed where it is set to zero in deterministic
  GTP commands
- adjust conection and semeai depth parameters according to levels
- don't forget to switch exchange sgf_messages in order_moves()
- influence tuning
- tuning
- joseki tuning
- owl tuning
- major eye tuning
- atari_atari tuning
- new tests and test revisions

3.3.19 released April 28, 2003

- change USE_BDIST to 1 in owl.c
- who_wins() invokes silent_examine_position()
- bugfix in do_owl_analyze_semeai()
- global variable urgent retired
- num_stones pointer in do_corner_matchpat() replaced with macro
- joseki tuning
- new game records and test cases

3.3.18 released April 22, 2003

- don't even try 'wrong' liberties in simple_ladder_attack()
- major revision of do_owl_analyze_semeai()
- do_owl_analyze_semeai tracks ko results and result certainty
- semeai tactical phase deprecated
- semeai_trymove() eliminated
- semeai worms redefined
- new r format character supported by vgprintf
- uncertain semeais treated as critical
- new function sgf_trace_semeai() in cache.c used by SGFTRACE_SEMEAI() macro
- new dragon2 fields semeai_attack_certain and semeai_defense_certain
- owl_estimate_life() revised for use in semeai reading
- don't attack unconditionally dead stones, even if thrashing
- never play on vertices which are unconditional territory for either player
- bugfix in attack_either()
- back-capturing added in attack1() and code restructured
- new regression script regress.pike
- hashtable_clear() doesn't clear already clean parts of the table
- `stackp' pseudofield in Read_result renamed `remaining_depth'
- endgame1.sgf removed from the distribution
- play_move_no_history() revised to speed up undo
- modify_eye_spaces(), remove_lunch(), and not_lunch_helper() removed
- new function estimate_lunch_eye_value() broken out of sniff_lunch()
- obvious_false_eye() moved from owl.c to optics.c
- lunch invalidating patterns removed from conn.db together with
  corresponding code in connections.c
- semeai moves stored in dragon2 array
- new functions collect_move_reasons(), semeai_move_reasons()
- new prevent_attach_threat_helper() adds reverse followup value
- DEBUG_TERRITORY made less confusing
- handle suicides in do_commit_suicide(), not in do_play_move()
- fix prisoner accounting in do_commit_suicide()
- DEBUG_INFLUENCE made less noisy
- bugfix in gnugo_estimate_score()
- reevaluate_ko_threats() revised
- new helper backfill_replace
- revisions of twogtp.py
- bugfix in play_move()
- improved output files from scoring mode
- traces added to find_double_threats()
- semeai tracing improved
- fix arraybound violation in corner_matchpat()
- tuning
- influence tuning
- joseki tuning
- owl tuning
- atari_atari tuning
- eye tuning
- new and revised test cases

3.3.17 released March 2, 2003

- most superstring related move generation broken into new superstring_moves()
- try (a bit) harder in attack_either()
[--snip--]
