format is as follows
HINT#combo hold(blah|myhand)^until(blah|mygraveyard)^until(blah|opponentshand)^restriction{type(creature|mybattlefield)~morethan~2}^cast(blah) targeting(blah|mygraveyard)^totalmananneeded({g}{g}{r}{u}{2})
the ai can be told to hold more then one card, until as many condiations as you want are met, until( is a TC and can basically be used in a fasion of saying "hold arbor elf until you have a lord of atlantas in play and a gaint growth in you hand"
once the condiations are met you can later tell it to cast(gaint growth) targeting(arbor elf[fresh]|mybattlefield)...
I also included the whole of the games restrictions system...
so you can get really really creative with this so far.
the next thing I will do is ability targeting and card favoring.
- attempt at reducing loading times on the PSP: I merged a few graphics files together, removed some unused calls from the initialization functions, and moved some other ones to have a more lazy approach. The PSP version remains fairly slow in some parts (especially loading, but also entering the shop, or starting a new game), so I will try to reduce file access as much as possible in the days to come. Not a release blocker IMO though, but I4d sure love if it were faster.
- uppercased "Track1.mp3" to be in line with the actual filename. Most likely this had been broken forever on case-sensitive OSes
- I removed costly calls from the textscroller. I believe it wasn't very useful in its previous state. Now it's only "advertising" for unlockable stuff, which I think is ok (and allows to refresh it every time the menu is loaded)
- As a counterpart, added a "% complete" progress bar in the menu, something I wanted to add a while ago.
- Modified DuelLayers to not use a global MTGPhaseGame instance anymore
- Moved the reset of currentActionCard out of the ActionLayer render function : that fixes the remaing problematic tests in the multithreaded testsuite
- Added a method in ActionLayer converting a card ability into a menu index
- Used this new method in the game observer to log correctly AI ability actions
- Added a DumpAssert method in the game observer, it can be used to dump the game and assert in order to easy crash reproduction
- Cleaned up TargetList properties access
- Added an optimisation in GuiMana to not compute update code if the rendering is not used (multi-threaded mode)
- Added a deadlock detection in the test AI vs AI multithreaded mode
- Fixed minor bugs in test AI vs AI multithreaded mode
- Added a games/second counter in the test AI vs AI rendering
- Added an "ACTION_LOGGING_TESTING" mode in the gameObserver. When this is defined, the game reloads itself in every update. I want to use that to track undo problems. Be aware that it kills performances and crashes with the testsuite if you want to activate it.
- Various cleanup/refactor of the game observer.
- Added a gameObserver == operator to compare two games
- Added player mode to the player serialization
- Added a multi-threaded mode to AI_CHANGE_TESTING. For the moment it's only useable with Qt. If you want to use it without, just defined a thread_count higher than 1.
- Refactored random generator class to use list intead of queue
- Defined a specific type for interrupt decision instead of int
- Reworked the testsuite to be able to work multithreaded. This is deactivated by default everywhere except in QT_CONFIG as one testcase still refuses to pass in multithreaded mode. On my 4 cores linux desktop, the 650 tests passes now in 4 seconds (1 fails).
- Replaced usage of CardSelectorSingleton by a card selector per game observer.
- Modified the resource manager to be optionnal and per game observer instance instead of being a singleton. Two reasons here : threading AND Open Gl access. I only updated the crashing parts called from the game observer, so most of the code is still using the single instance. Beware of copy-paste concerning resources ...
- Cleaned up the game observer constructors
- Fixed several problems in action logging code while testing proliferate decks
- Cleaned up Threading implementation based on QThread
- Decorelated the testsuite AI timer from the game timer to be able to have reproduceable results with AI tests.
- Created a random generator wrapper class
- Used two seperate instances of this random generator for AI and for the game
- Added methods to load randoms into AI from a testcase
- Fixed a probleme with undo and premade decks introduced in r4035
- Added basic test to test AI proliferate code
- Cleaned up goblin_artillery test
- Added AI tests into the testsuite test list
- Fixed looping bug into the multi target AI code
- Modified undo to stop at "next phase" action
- Added "muligan" and "force library shuffling" to the list of logged action
- Fixed random logging
- Fixed double logging of actions
- Merged all the "next game" functions into a single one
- Created a PlayerType type instead of using int
- Moved the player loading code into the GameObserver and out of GameStateDuel to avoid having player references in both and simplify the initialization and termination. Tweeked a bit the humanplayer class to be able to do that.
- Added a "load" menu available in testsuite mode, I use that to load problematique game. To use it, just copy-paste a game from the traces into Res/test/game/timetwister.txt. Game in traces starts by "rvalues:..." and ends by "[end]"
- Added some untested and commented out code in GuiCombat to use the mouse/touch to setup the damage on the blockers
- Broke the network game ... hoh well, I'll repair it when everything else works !!
- various code cleanup and compilation fixes on Linux
First version where I managed to finish a normal game while undoing several actions until the end. There are still some problems in direct damage spells and interruption management. I added several assert in the code to catch them.
- Added an undo menu using this code (beware, it's still very very alpha).
- Removed various warning
- Cleaned up avatar loading
- Added full random lists load/save including the deck shuffling (not sure if I could not replace that with seed load/save)
- Moved momir and Co rules configuration out of GameStateDuel
- Create a GameType type to avoid mixing int everywhere
this fixes 2 crashes I found, the first, 2 color random mode would crash on load.
2nd, ai vs ai testing would randomly crash, this should fix that also.
I noticed 2 color random mode is now trying to search for it's rules and sometimes flashes for a brief moment "error cant read file" or something like that....I could not find the source of that, it doesn't cause it to crash however it causes it to take a sec longer to load, this is before this commit btw, so the issue is still there.
it was trying to load the rules, flashed the error then crashes...i fixed the crash but not the rules error.
please review, i might have left in useless stuff...
I also did notice something, the way we are creating players is kind of all over the place. imo this is bad, it made this conversation extra hard becuase you create one player over here, another type over there, the human over in this direction, back track and create another somewhere else...this needs to be taken into account for a refactor, all player creation should happen in the same function, and at the same times...
the reason these 2 crashes existed was becuase players were being created before "gameobserver" in some modes, and in other modes, no player would exist at the time game was creating to set the player. but we then later call the same function when we actually load the player using the method specific to a mode.
this just leads to headaches, I mean no offense, just a general observation i made when converting this players array. unfortunately that kind of refactor is just a little beyond my coding ability.
important note: AiAction you always want to call "owner" becuase "player" here means single target spell has a target and the target is "player". I didn't come up with the variable name, sorry it is very confusing in this spot of the code.
- removed every references to the gameobserver singleton. This object can now be instantiated several times as it's needed for minmax. To be able to do that, I mostly added a reference to a gameobserver from any targetable object (cards, players, spells) and abilities.
- removed incorrect casts of MTGCardInstance into Spell objects.
- AI Test system now allows you to put decks in ai/bakaA and ai/bakaB instead of ai/baka. This allows to let AIPlayerBaka and AIPlayerBakaB play with specific decks
- Test suite speed improvement. Improved the card name cache. Test suite now runs in 850 seconds instead of 950 on my machine.
- minor code cleanup
- In order to clarify getEfficiency, started creating "getEfficiency" functions specific to each type of ability (see example with Damager). This won't reduce the file size but should make things bit clearer
- Minor cleanup of getEfficiency
- bug fixes in getEfficiency
-- This is a copy/paste and shouldn't have any impact on the logic. I just moved some functions from AIPlayer to AIPlayerBaka
- Added back the possibility to select a different Resource folder with file Res.txt
- Fix a crash when a token id does not exist
moved debug framerate message to be 10px higher. current position covers the credit count on the game shop screens.
changed hard coded numeric values for menu ids to Labels. ( this was done only to GameStateAward )
added abilities:
proliferate
ProliferateChooser:new targetchooser for cards with counter and poison counters "proliferation".
MenuAbility:new internal ability to create custom menus of abilities which can be activated in sequence one after another.
multikicker, syntax kicker=multi{b}
works with variable word "kicked", the amount of times it was kicked.
target=<number>tc,target=<upto:>tc,target=<anyamount>tc,target(<number>tc),target(<upto:>tc),target(<anynumber>tc);
multitarget is now supported with the exception of "devided any way you choose" which can not be supported becuase we allow detoggling of targeted cards with a "second" click....so you can not click the same card 2 times to add it to the targets list twice for example.
this is minor, as the bulk of multitarget is not "devided"
removed 's' parsing for multitarget, added a limit of 1000 to "unlimited" for easier handling; we currently can't handle activation of an ability on a 1000 cards very well on any platform(infact i don't suggest it)
Countershroud(counterstring), this MTGAbility allows you to denote that a card can not have counters of the type "counterstring" put on it.
"any" is for no counters allowed at all. this is a replacement effect. cards state that they can still be the targets of counter effects, however on resolve nothing is placed on them instead.
@counteradded(counterstring) from(target):,@counterremoved(counterstring) from(target):: these are triggers for cards which state "whenever you add a counter of "counterstring" to "target"; added counterEvents struct;
other changes:
added support for ai handling of multitargeted spells.
changed a few of delete( into SAFE_DELETE(, safed up a couple areas where they did not seem safe to me;
added better handling of menus presented to ai, it will try to select the best based on eff returns.
added varible lastactioncontroller for ai use, it keeps it truely from ever tripping over itself and brings ai more inline with MTG rules.
converted TC into a protected member.
added "abilitybelongsto" string to tc, and set "owner" of the tc. a tc should never belong to "no one" it should always have a owner.
abilitybelongs to string is solely for easier debugging, i found it was a pain to never know what ability created a tc while i coded multitarget. the owner of the tc is the only one that should be using it, if an ability needs to declare the opponent as the owner (choose discard which is currently unsupported for example) this will allow us to better handle that situation by setting the tc owner in the ability which called it.
rewrote the logic of "checkonly" in ai choose targets, the only time it is "checkonly" is when it is trying to see if it had a target for a spell before it cast it, i now set this in the actual function call instead, the old method was far to error prone.
wrote logic for ai checking of menu objects presented to it,
ai will now make better choices when a menu is presented to it based on what it already knows. this changes it from it's old method of "just click the first option".
taught ai how to use multi-mana producers such as birds and duel lands by adding a method for it to find it's mana for a payment. it can effectively use cards like birds of paradise and sol ring(without locking up). It's primary method of pMana searching was maintain for performance(no need to deep search if we have it in pMana).
added a vector to actionlayer to store mana abilities for pMana. this provides us with a dramatic improvement when mana lords are present by reducing the amount of objects that need checking when ai checks pMana.
with 80 mana objects and a ton of lords one instance i checked went from 8000ish checks down to 80<===big difference.
added "tapped" green coloring(sorry i missed that!)...added red coloring to current actionLayers current action card (usually the source).
changed "type(" restrictions second amount from atoi into wparsedint for more flexiable coding.
add "&" parsing to CD targetchooser, removed "iscolorandcolor" variables and functions becuase they were a hack the real fix was this.
cretaure[dragon&black&blue] a creature that is a dragon, and black and also blue.
changed some of the ai computeactions and
removed unneeded gaurds in ai chooseblockers, they did more harm then good.
-- zipFS has several limitations...
--- in a general way, seekg doesn't work... so getting a file's size needs to be done through JFileSystem.
--- getLine on files open with zipFS doesn't work so great. Not sure if it is a normal issue because files are open in binary or not... JFileSystem therefore offers a "readIntoString" function that needs to be used instead of the usual "getline" technique. However getLine can then be used on a stream connected to the string.
-- tested on Windows and PSP, I also made sure android still works, but haven't tested zip support on Android.
-- I tried to maintain backwards compatibility, but this might break on some platforms, if I broke some platforms and you can't find a way to fix them, please contact me and we'll figure something out
-- This removes wagic::ifstream. I didn't reimplement the securities that were involved in this, apologies for that. Might be useful to reimplement such securities in JFileSystem
-- I haven't tested options/profiles in a deep way, it is possible I broke that.
- rewrote some code dealing with text formatting for a card when rendering in text mode. I could swear my code has been reverted. If somebody has good reason to believe we should re-process the string formatting on every frame, please let me now. I believe my change can help rendering speed of text mode a lot.
- counters change to vector instead of array
this update requires you to update your rules folder files!!!
2nd
added 2 new vanguard game modes.
Stone Hewer Basic - when ever a creature enters play, a random equipment with a converted mana cost less than or equal to that creature is put into play and attached to it.
this mode is unlockable, requirement = win a match where 10 or more equipment were in the battlefeild at the moment you won.
Hermit Druid basic- in this game mode, during each of the players upkeeps, a random land card from their deck is placed into the battlefield, these do not count against your 1 land per turn limit.
to unlock this, win any match with less then 10 lands.
- Added a way to specify "no interrupt" in modrules.xml (I need to write a doc about this file) for a given game. This does not mean the stack does not show up (the stack is interesting to see what the opponent does), but that it only offers to continue.
- Fixed a Bug where AI would not correctly assign blockers if the first attacker is super strong.
- Added a hack to prevent AI from an infinite loop while choosing a target. There are edge cases where the AI gets to choose the targets for a TargetChooser that doesn't belong to it. I couldn't dig too long for the root cause, so I added a "return 0" when the case happens. Should probably open a ticket
- Added a "Hint" System in AI decks, to help the AI with its strategy. This is not really usable yet, it only works with abilities (not cards to play), and I only added some basic code for counters and tokens. This can probably be extended, but let's wait until we see it working on that other game I'm working on, before rushing into adding hints to all AI decks...
- minor cleanup of AI Code
comparing counters to p instead of isAi(), and hope to close off an edge case where ai would get stuck cycling through setting the target of a may ability. caught it stuck 2 times while watching demo. i watched demo for close to an hour without seeing any adverse effect, and without getting stuck, tho this is so hard to debug, since there is no call stack loaded for this case when it happens.
Also fixed a subtle memory pooling issue in the RenderCountersBig() routine: if you're in regular card display mode, the idea is that formatted text is only fetched if you flip into alternate render mode. However, in this function, if counters are being drawn, it would fetch the formatted text in order to determine where to draw the counters, EVEN IF the counters count was zero. So it had nothing to draw, but it meanwhile pooled the formatted strings into memory anyway.
I've marked all the ones I found with the following TODO comment:
TODO: C6246: <blah blah>
a few in particular are the ones related to "oneShot" and "_target". These are local variables that are declared that
mask either a method parameter or a member variable.
