Internally, commands are now stored a bit differently: some commands
which were at the end of a turn (like the players rack in duplicate
mode) are now at the beginning. But from a user point of view, there is
no visible difference, except some details in training mode (and the new
behaviour is arguably better :)).
Only (known) problem: after loading a game saved before this commit,
navigating in the game history may be less nice (in particular, racks
may not be fully restored).
This also fixes a very old (but minor) bug: when a word containing a
joker was played "manually" (i.e. not via a round generated by the
Results class but via the Game::checkPlayedWord() method), subsequent
search results using this joker on the board would display it as a
joker, not as a normal tile.
Knowing the actual player is not always possible, since potentially none
of the players has found the played move (for example in arbitration
mode). This ID is now only used to retrieve the rack from which the move
was formed, so any player ID will do.
The non-regression tests have been adapted accordingly.
Status:
It works well, but there are still a few details to improve/fix
More details about the changes:
- New dependency on Arabica and Libxml2 to parse the XML
- Loading the old format is still supported for this release, but won't be supported anymore in the next one
- Games are now only saved in the new format
- In training mode, the player is now created externally, like in the other modes
- Avoid using GameIO (the one from game/) whenever possible
- Do not use a FILE* argument anymore when loading a game
- Throw and catch exceptions correctly when a game cannot be loaded or saved
- The non-regression tests now use a new method to print the game history
This display string is used in the ncurses and Qt interfaces.
Display strings are types "wdstring" instead of plain "wstring" in a few places (not all of them) to help recognize them.
But we don't need to sort them all the time, and in general we don't even need to keep all the rounds.
This commit greatly improves the search performance by filtering the results in 3 different ways, depending on the context:
- A limit to the number of results can be given (useful for the training mode). The kept results are the best ones, not the first ones found by the search.
- When only the best round is needed (when the AI is playing with level 100, or when preparing the rack for an explosive game), we don't need to keep rounds with a lower score
- When the AI has a level lower than 100, it is still possible to skip many rounds
The search limit in training mode is configurable (defaulting to 100) and can be deactivated.
- Added a unit test to ensure it
- Fixed Game::back() (this fixes some unit tests broken by a previous commit)
- Added a unit test for a duplicate game with only one player
- Fixed a crash on Windows
- Do not authorize words if the cross-checks are invalid (it was possible when playing a joker)
- Fixed detextion of invalid moves
- Improved unit tests, and fixed a bug in one of them
Qt interface:
- Do not allow playing a word if the coordinates are missing
- Better focus
- regexp search results are now in uppercase, like other results
- the regexp search indicates if there are more results than the maximum
allowed
Qt interface:
- new window with some dictionary tools: check word, word+1 and regexp search
- New regexp parser using Boost.Spirit. Lex and yacc are now gone.
The main advantage of this new parser, apart from being purely C++,
is that it can handle wide characters.
Currently, the new parser does the same as the previous one, but the
code is not yet ready to use regular expressions with non-ASCII
dictionaries.
- Fixed a bug occurring in duplicate mode: sometimes the game ended too early,
while it was still allowed to go on.
Also updated the corresponding regression scenario
There are too many change to list properly, here is an overview of the main changes:
- the dictionary is now in C++
- the dictionary has a new format, where it is possible to specify the letters,
their points, their frequency, ... It is backwards compatible.
- Eliot now supports non-ASCII characters everywhere
- i18n of the compdic, listdic, regexpmain binaries
- i18n of the wxWidgets interface (now in english by default)
'regression.pl' is the script that runs the regression, and 'drivers' contains
the list of scenario files to be played.
By default, the complete regerssion is played, but you can play specific
scenarios by adding them to the command-line.
Example: ./regression.pl training_bag freegame_change
