Current Architecture Projects
Contents
|
PCGen Architecture work
6.3 Projects
Ability Cleanup
Ability Cloning Work
http://jira.pcgen.org/browse/CODE-2465
Tests
Need to develop a significant number of tests to ensure this transition takes place correctly - big risk in duplicate removal not working correctly
MULT:YES STACK:NO items taken multiple times, multiple locations or same location
MULT:YES STACK:YES items taken multiple times, multiple locations or same location
A bunch of this is already in the integration tests (characters), but may want utests to check this (or really itest in the sense of doing tokenmodel tests)
Specifically what needs tests now is bonuses in Abilities - since the bonus system specifically grabs associations, this requires some significant testing to ensure it is correct
In Working Patch
CNAbility Relationship
Factory to build CNAbilities provides the right mix
Simplifies CNAbility usage - build it when you need it (doesn't have to purely be one-way-in one-way-out)
Identity == Required to have separate identities to make removal of items appropriately unique (many facets store source in IdentityHashMap, can't break that yet)
"Useless Adds"
Need an appropriate way of handling these situations and NOT lose information in case the duplicate "useless at the time" is not actually useless
The case is: MyAbility MULT:NO. Add MyAbility|!PRELEVEL:1,Fighter=3. Add MyAbility again at level 2, but the first one is revoked at level 3. How do we know we still have MyAbility?? (we DON'T today - this is a bug) - we must do duplicate removal "later on"
Solution is in having CNAbilityFacet that is MULT aware, then another facet that is stack aware
Disambiguation
Much of this (and how to replace methods) is based on whether items are called as a driver (what made the CHOOSE selection) or the target (the entire ability).
A destructive exploration is here: assocbreakout.23247.patch
The tracker: http://jira.pcgen.org/browse/CODE-2473
Chooseable
Race, Template, CNAbility, Skill, Domain become Chooseable objects
Requires split of ADD/CHOOSE (complete)
Requires Ability cloning cleanup on PlayerCharacter to be completed to the point where the getAssociation related methods can take a Chooseable
Facet Reduction: Abilities are currently handled in two facets and needs to be reduced to one
Future Considerations
Sample Object Removal
Description.getDescription gets a sample object (choosedriver assumption)
BenefitFormatting gets a sample object (choosedriver assumption)
PlayerCharacter.getDescription gets a sample object (choosedriver assumption)
PRExxx source
PREVAR needs source (only reason it is passed through to PRE testers) and that is a problem for CNAbility - want to avoid a special case :/ ... NOT a problem, just have to expose variableSource in some way...
This is actually an abstract issue at this point - due to ADD and Template's FEAT being forced to act as if it is a user selection, the source is not terribly relevant
Source
Handling source (for conditional) and pipeline (disambiguating sources) is tricky, since the CDOMObject source is required to properly evaluate things like PRExxx while the facet source is required in order to ensure that tokens cannot pollute each other
Idea: Have a separate facet store the CDOMObject source (ScopeFacet)
ScopeFacet can then be called by the conditional facet in order to pull the source, thus the conditional facet doens't have to be the first facet in the chain
Associations
May need to move associations out of AssociationSupport and into a Facet
Issue is whether CNAbility is identity or not
Without identity, need to ensure associations are properly attached for the UI
Currently AssociationSupport does not store sources of the associations
Facet is SubScopeFacet with some modifications, specifically to handle source weighting
Quick update does not interact well with cloning, since Ability is used to store associations (change to first Map in SubScopeFacet may help?)
See: assocToFacet.23464.patch
Problem is that restoring items on PCG load is an issue - don't have the original ADDED_FEAT for virtual abilities
This interacts with AssociationListKey.ADD since that is necessary to avoid redoing ADDs (see skills)
Major Issues
Lost Info
Information was destroyed in the PCG - have to assume all ADDs as well as Template's FEAT are "user selections" when that is not true - so removal on a restored PC will not (ever) work correctly
Establish if there is a method of determining the source of an Ability in the PCG file (info was destroyed on save)
LST Editor
Weaknesses
CATEGORY token is not editor friendly
KEY token is not editor friendly
PC Class Level repeatlevel does not properly parse for edit context
Kits are not at all editor friendly
Ability Category is not at all editor friendly
Core View
Getting Abilities into core view requires the Ability Cleanup to be completed, since the system must have facets in the general form expected by perspectives
Get Skill Costs into Perspectives
This has some unique challenges since it is not one object type. Once you pass through certain objects (looking backwards through the event tree) you end up with the PCClass disappearing or other things dropping out, so the Perspective somehow has to display that.
See: skillcostmetafirstpass.23125.patch
Templates
Facet Flow
TemplateInputFacet -> TemplateSelectionFacet -> UnconditionalTemplateFacet -> TemplateFacet
TemplateSelectionFacet -> ChooseDriverFacet
Input knows of a PCTemplate (and forces the CHOOSE if necessary), the TemplateSelectionFacet has a Selection, not just the PCTemplate, and then Unconditional and the model (TemplateFacet) hold just the PCTemplate
Problem: The flow of facets changes object classes multiple times. This is going to be complicated to write into the coreview system and makes templates different from something like languages. I'd rather it be similar.
Problem: TemplateInputFacet, in order to know which Selection<> is related to a Template (so the remove can be symmetric) is "fooling" the TemplateSelectionFacet to say that the source is the original PCTemplate. This allows add/remove symmetry, but the REAL source cannot be stored. Right now this is somewhat academic (as addTemplate() in PlayerCharacter never asks for the source), but in order to get things into the core view system, we want to know that source, so we want to enhance addTemplate on PlayerCharacter, and then store the real source later on (And be able to know that for any Template)
Required Flow
TemplateInputFacet -> TemplateSelectionFacet -> ChooseDriverFacet
TemplateInputFacet -> UnconditionalTemplateFacet -> TemplateFacet
This way the unconditional flow matches how other objects will look, and the selection is solely responsible for the association on the Template. This also means TemplateSelectionFacet becomes an AbstractAssociationFacet (thus more aptly named for its behavior than an AbstractSourcedListFacet)
In the future Domains and Races will change flow to match template
Conditional Templates
These are done in a "pull" fashion, and we want to do more "push". This work is in ConditionalTemplateFacet.
Improve for consistency and for the Core View system. Right now we have visibility to conditionally granted but not conditional (but not granted).
See: CODE-2442.condtemplatename.23102.patch
See: template perspective.23082.patch
http://jira.pcgen.org/browse/CODE-2006
Conditional Skills
Need to get conditional skills broken out into separate objects
SkillsFilter
Remove SkillsFilter to be only for the UI and not impact the core
Issue here is how to do "forward" processing for skills
See: skill filter first pass.23117.patch
Discussion
Dev Proposal is here: https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3773
_exp formal proposal See: https://groups.yahoo.com/neo/groups/pcgen_experimental/conversations/messages/17226
Facets
Facet work
Bonus MONSKILLPOINTS/LOCKNUMBER needs a wrapping facet on adding a template
Template Facet perspective needs source of the Template to be saved
CompanionMods are still done in a loop in PlayerCharacter - need to put in a facet
Consistency
http://jira.pcgen.org/browse/CODE-2442
Want to reduce Spell Facet uniqueness - see MasterSpells in DataSetID-based
DataSetID based
Related to Context strategic issue
There are certain items that are campaign-related not PC related
Master Usable Skills (see Conditional Skills)
DatasetID / Spells
Master Spells
Make Master Spells use DataSetID facet.
Include a cleanup to make the spells facets sub scope facets
Include CODE-2501 to reduce the memory usage of "unused" spell facets
See: +masterspell.23393.patch
Note Side effect: The spells today list ALL places they COULD be achieved from - the patch above makes them only appear with the sources where the CHARACTER receives them from. This means two things: innate spells don't show up with ANY sourcelist AND spells that do have a sourcelist will only show sourcelists that the PC actually has access to... is this side effect appropriate??
Concurrency
All Facets need to be tested against concurrent modification - so have a recursive removal (which someone could write as an iterator) have it also remove another item of that type - causes CME if not careful
BUGS: Potential
It appears ADD/CHOOSE on MULT:YES abilities no longer respects choices that have already been made; introduced in 6.3
MoveResult.getBaseMovement seems to be wrong as it is ignoring the load
TemplateExportToken feats method seems to be buggy as it ignores the level and HD parameters
PersistenceManager methods are called (including clear, and checking for presence of campaigns). These are no longer used by gui2 and should be removed, see: http://jira.pcgen.org/browse/CODE-1891 http://jira.pcgen.org/browse/CODE-1890 http://jira.pcgen.org/browse/CODE-1889
Description seems to ignore the Feat name when %FEAT is used with a specific Feat ... Need to check if this feature is at all used or valuable. see https://groups.yahoo.com/neo/groups/pcgen_experimental/conversations/messages/17087
Compatibility tokens may be spitting information into ParseResult even if they fail, which is bad as they should silently fail
It looks like an item with BONUS that contains <this> will fail if it is subsequently .COPY='d
Class Loader
SUBSTITUTIONCLASS:A LEVEL:1 CLASS:B LEVEL:2 CLASS:a.MOD LEVEL:3
Does PCClass loader have an issue? with order of ops
Where is LEVEL:3 placed? may be CLASS:B??
We need a complete test pass to determine if all ways to add an Ability respect MULT and STACK. This involves finishing itest/tokenmodel.ability
ABILITY:CAT|NAT|Foo(%LIST)|PRExxx: seems to ignore the PRExxx (pretty much all tokens of this form need to be checked)
It appears that ADD:FEAT|Foo ADD:FEAT|Foo will not work (Foo is only applied once, even if MULT:YES)
PRE house rule checks seem to allow for all kinds of bad behavior. AUTO:LANG|Foo|PRExxx on a FEAT will be passed if FEATPRE rule is enabled - isnt' that wrong??
BONUS tokens that impact the PC that are on EqMods may get applied more than once?? See TEST-98
Looks like STACK:YES item when reselected will re-add the originally selected item (set returned by the chooser is ALL choices, not new ones)
Epic Check: 3.5e Loremaster get bonuses, but if those bonuses are based on class levels, they may be lost since the epic level swap is doing churn in order to get CL correct (if not Loremaster, find something that gets BONUS from an Ability granted by the PCClassLevel)
.FORGET interaction with Unconstructed References. Are false positives being displayed, or should there be some implied method of "FORWARDREF" so that .FORGET doesn't require a whole ton of cleanup when an item is forgotten?
Documentation
Need to doc facets explanation, see https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3721
Primitive and Qualifier Architecture (Gabriel taking the first pass on the Wiki): https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3682
Review adding a new token documentation, see https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3529
Need to document Visibility/View usage
Need to check the items in pcgen.cdom.* to ensure the docs are correct - the facets seem to have taken on a bit of copy/paste error over time as some old references were missed in the docs
Known Bugs
Misc
Error/Enforcement
MULT:NO behaviors
MULT:YES abilities can be given MULT:NO behavior and that probably deserves a warning. The main problem is that if a MULT:YES Ability provides something like ADD:FEAT|Dodge, then Dodge will only be applied once, not the number of times the Ability is taken. That's because ADD is a MULT:NO behavior of the Ability. There are other examples, but the point is that it should be possible to know MULT:NO behaviors and warn on usage
see: detect mult single conflicts.22912.patch
see https://groups.yahoo.com/neo/groups/pcgen_experimental/conversations/messages/17262
Ability Names
Need to have a check on Ability object names, if X is a mult:yes ability then X(a) must be prohibited. While we support this currently in the parsing, there are other later subsystems that end up with ambiguity - specifically if the item appers in ADD:FEAT as an example, the two cannot be distinguished.
See: https://groups.yahoo.com/neo/groups/pcgen_experimental/conversations/messages/17250
Consistency
Bracketed PRE: PCClass/PCClassLevel (ADDDOMAINS) - NEWTAG 431
PCClass SPELLTYPE needs to reject empty and NONE - NEWTAG-430
Epic Checks/BAB
Problem: We expect the code to magically know that in the case of BONUS:CHECKS, the term CL is limited by the classes under the epic limit.
Implementation is ambiguous for certain object. It causes a churn in the objects on the PC.
Counterexample: Should an Ability attached to an epic class level impact the non-epic BAB?
Introduce a method of identifying "non epic levels", something akin to: BONUS:CHECKS|BASE.Reflex|NECL/3
Required Steps
Deprecate BONUS:CHECKS. This forces folks to "look at it" in the transition to 6.3/6.4
Add a BONUS:CHECK as a replacement (which basically does the exact same thing)
Add the term/formula for NECL
Conversion
If a trivial parser can determine that the term CL is used, when BONUS:CHECKS is on a base class (CLASS: line in a Class LST file), then it is converted to BONUS:CHECK using NECL
else it is not automatically converted and requires user intervention.
Refactoring
Type safety
Make SubClassFacet Type Safe, see subclassfacet-typesafe.22718.patch
LoadInfo refers to sizes
http://jira.pcgen.org/browse/CODE-1929
Class Moves
Export Tokens
io.exporttoken.StatToken can be moved to a plugin
io.exporttoken.WeaponhToken can be moved to a plugin
io.exporttoken.TotalToken can be moved to a plugin
Static Method Moves
AttackInfo.getAttackInfo can be moved to an export token
CDOM
pcgen.util.enumeration.Visibility can be promoted to cdom.enumeration
pcgen.util.enumeration.View can be promoted to cdom.enumeration
PC Associations
Eliminate SPECIALTY association
See elim-specialty-assoc.22785.patch
Problem: This breaks Quasvin as redundant information was saved but the redundant data is no longer relevant to the PC: https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/topics/3688
http://jira.pcgen.org/browse/CODE-1908
General Cleanup
Refactoring for Clarity (Gabriel's Suggestions): https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3782
Would be nice to improve the use of generics in gui2 (they are ignored in many locations where they could be used), see improvegui2generics.23033.patch
Eliminate catching of Exceptions without further messaging of the underlying issue
removeAll (id, source) on SourcedListFacet should be removeAllFromSource
Unit Tests
Cleanup
TokenLibrary.addBonusClass is called widely in test cases (as a static call) and should be consolidated into the testcommon directory in some form
StringBuffers
These probably DO deserve a default size as per the tracker indicating as such
See: stringbuffer size.23000.patch
Strategic Projects
Localization
Type Safety
A number of items need to be made type safe to handle L10N
RacePantheon (Deity)
Incompatible Tokens
A number of tokens are incompatible with L10N because they cannot easily be matched (DESC, BENEFIT, generally any free-form String that is allowed to appear multiple times (Spells will also have a bunch)
These may not matter as much if we use PO files? But it will make the PO files very sensitive to tiny data changes. Is there a compromise?
Also, how would replacements be handled?
General Concepts
We must not make producing a data set materially more complicated than it is today (no requirement to put %L10NNAME% type gunk into data)
We should target an ability to tell us if l10n is complete for any given data set
I need to understand more about L10N than I do today - I need to understand this issue: http://wiki.pcgen.org/Localization
Proposal, see: http://wiki.pcgen.org/Internationalization
PO Files
What tools are available for PO files that make a difference?
Are there any tools to help create PO files in Java since we know what needs to be translated?
What is the advantage if we use PO files vs using a custom format?
Formula Parser
This work is significant since any changes we make in the Formula Parser can't be leveraged by something like the bonus manager until a subsequent revision
Initial work done in 2010 time frame by Nuance to get the Parser up and running
For JEP Replacement
Get new formulas into a set of plugins rather than direct injection into maps
Change operations to be able to handle Number, Number as the parameters and return Number. Allows integer math when possible (Eric is looking at this as an initial project)
Conversion of Terms/Formulas
Convert existing terms/formulas to the new formula parser. A representation of what has been done is in this destructive exploration: formulacompleted.patch
Goal is to avoid the use of pattern matching that is used by the current system. In many cases, there are only a limited number of choices that the pattern match worked to handle, and these can be explicitly built
General Concepts
Formulas
Pluggable
getFunctionName() plus an interface to define which form
loaded like we do the lst tokens today
Three forms
Paren formulas
e.g. today's count() function and the other items in plugin.jepformula
BracketFormula
The COUNT is the bracketfunction
Compatibility with the old COUNT system, e.g. COUNT[CLASSES].
Since the current proposal does not include terms, need to think abotu how COUNT doesn't become a huge IF statement - can we load formula components or some such?
DotEqual Functions
These are actually "derived" from terms as they are not separated by the parser
Things like CL=x or CL.x
PCGen will need a "special" TermResolver to break these out
Validation
Every formula needs to be able to indicate if it is valid or not. This includes both checking the function name (the system will probably do this) as well as all of the arguments of the function
Every formula needs to be able to indicate if it is static (for optimization at runtime)
Parsing does not indicate it is valid. IT could be CNT[CLAZZES], which is not valid because CNT is not a bracketfunction and CLAZZES is not a bracketterm
If not valid, it needs to return WHY it is not valid - much like tokens return ParseResult.FAIL, Formulas need to return a FormulaValidity that indicates success or the problem (pcgen.base.formula.error)
There are no terms
Main Goals
Isolated from PCGen
Isolated via various interfaces
Note pcgen.base.formula.base does NOT have PlayerCharacter as an argument - but womehow it will HAVE to (unfortunately). Probably generics, but the point is that things like count() eventually need to be able to access the PC, and thus the function will have to receive some type of resolution object during evaluate()
Have a system that can be fully unit tested independent of PCGen. Therefore, no references to CDOMObjects or (especially) PlayerCharacter can be in the formula parser. Legal in Formulas since those are plugins
Minimize customization to classes related to the parser - walk the tree when possible. (few svn stored classes in pcgen.base.formula.parse)
Have the ability to detect use of different types of formulas and deprecate them individually
Temporarily compatible with JEP (try new system first, then old)
See: http://wiki.pcgen.org/Formula_Parser_Equip_Vars_Proposal
Equipment
Equipment primitive behavior (e.g. TYPE) can be modified by an EqMod, but that modification must occur in context (Note: Difficult to do this with just a modifier if there is any complexity beyond one level of EqMods)
EqMods can trigger the removal of other EqMods (in context to parent Equipment)
Proposal for Equipment Variables
Equipment needs to be promoted to a "full" object, on the level with PlayerCharacter
Proposal for Equipment to be able to contain other Equipment, needs to handle this in terms of character possession
Does Equipment have a "contains" set such that a PREITEM has to search up the entire "tree" hierarchically? How is that handled? Alternative seems to be that if the PC possesses an item, it must be in the PC's item list. This may end up storing redundant information, however
Is there a way to control the size of Equipment items so that most primitive items do not take up large quantities of memory? (what is the transition threshold and is this done with a strategy of some form?)
Need to finish separating the system so it does not have primary/secondary heads (but numbered)
Equipment would have its own resolution system, facets, etc as necessary
Bonus System Redesign
Proposal posted, see: http://wiki.pcgen.org/Bonus_Subsystem_Design
Breakout of types of Bonuses: http://wiki.pcgen.org/Bonus_Subsystem_Thoughts
Bonuses need to "lose" ownBonuses to get away from ugly behavior
Tokens
Nee to provide special consideration to the tokens we have that are "bonuses in disguise" and whether they should just become pure bonuses
Needs to include a token rebuild that includes better type safety (if they refer to a skill, we need to know if that name is valid or not)
Need some special attention on duplicate bonuses, and deciding what the "long term plan" is - combat has a number of bonuses that are available in multiple ways
Is there work that needs to be considered for bonuses that really aren't? Meaning bonuses that are actually acting as locks. This makes the name "bonus" rather confusing since it is a lock not a bonus
http://jira.pcgen.org/browse/CODE-2502
Export
Token Rebuild proposal
http://jira.pcgen.org/browse/CODE-1904
see: http://wiki.pcgen.org/Dev_Meeting_Log_20091114
Freemarker direct object access
Questions: https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3795
Provides a way to avoid the complicated token rebuild - focus on core code
Addition abstraction advantages in that input tokens can directly write properties to output without core knowledge if the map functions of freemarker work as per my questions
Looks to be a better solution than the Token Rebuild proposed earlier - less code for us to maintain
http://jira.pcgen.org/browse/CODE-2419
Performance
Bonus System String processing is probably our major performance issue
CollectionToAbilitySelection got a cleanup for infiniteloop but the same code exists in AbilityRefFromChoiceSet
Token Updates
Type Safety
Deity RacePantheon needs type safety
Template/RaceType
Template/Genderlock
Skill/ACheck
Error/Enforcement
Sponsor ImageSmallToken should be deprecated as the code is never used
ChooseLst should probably check that CHOOSE is present on an object if any choice actors are present on an object
CHOOSE:ABILITY and CHOOSE:ABILITYSELECTION the categories need to be a parent category not a pool, see https://groups.yahoo.com/neo/groups/pcgen_experimental/conversations/messages/17129
MULT:YES Abilities can only be directly applied if an association is provided, why is that not true for templates? and how would we actually provide the association?
Need additional checks for .CLEAR tokens - need to ensure .CLEAR applies (utest), and also that .CLEAR and then <blah> on the same line/object results in the correct answer during unparse (itest) [e.g. not caught in fumblerange as of 3/6/14]
Consistency
%LIST Consistency
Deprecate TEMPLATE:x.REMOVE and implement REMOVE:TEMPLATE
Can PreMultParser be moved to a plugin??
CSKILL and CCSKILL use LIST not %LIST
MONCSKILL uses LIST not %LIST
New Token Todo
EquipmentModifier CHOOSE tokens need update to new subtoken format
AbilityCategory should be converted to a CDOMObject and the fields moved to the maps and the tokens updated: Specifically one value here is reuse of ObjectKey's default on getSafe for Visibility
Idealism
Token separation
CHOOSE
Primitive PObject.FEAT= needs to have CHOOSE_INFO reference moved to PlayerCharacter
LoadContext.validateAssociations makes specific reference to CHOOSE_INFO but that is tied to the CHOOSE plugin
PrerequisiteUtilities makes specific reference to CHOOSE_INFO, tied to CHOOSE, move to PlayerCharacter
Enumerations
Delete AbilityCategory.ALL
Delete Nature.ALL
Interfaces
Should LoadContext really be an interface?
Should UnconstructedValidator really be an interface?
Eliminate Token Magic
Template/Subregion has YES
Template/SubRace has YES
Template/Region has YES
PRE/REQ
Design Issues
We have some items that are "prerequisites" (tested when the object is added to the PC) and some that are "requirements" (continually tested). Need to be able to put both on object and distinguish them
Prerequisites are not currently parsed as other items, and as such are not catching typos and otherwise type-safe
There are a number of things that cannot be done in Prerequisites
Limitations to overcome
There are requirements that include things like: 2 Skills of TYPE=Foo of rank 5. The problem we have is that if we list TYPE=Foo twice, then the same skill can match
The processing system we have today has trouble distinguishing between success and accumulation. For things like total number of skill ranks adding up to a number, we have to hack the prerequiste system to handle that, and at that point there are complexities in making that work in PREMULT and also situations that can cause it to fail. The solution is to have Prerequisites return a more complex object that indicates total matches, success, etc. and allow a later decision maker to interpret as necessary.
Right now there is a tit-for-tat system with qualification. Meaning you can write a PRE, and then have a QUALIFY override it and then a PRE:Q: override THAT. Well, that becomes confusing over time and basically ends up in infinite escalation where one item requires more and more text each time. Can you imagine PRE:Q:Q:Q:xxx and QUALIFY:P:P:xxx trying to compete?
http://wiki.pcgen.org/Prerequisite-using_Token_rebuild
FreeMarker
What do we need to have in terms of objects?
Assuming we have a set of objects is there a way we can hide thigns from Freemarker? (annotation) or do we have to build a facade? [looks like facade as best I can tell]
What do we initially want to have available? Certainly NOT the PlayerCharacter object
How do we appropriately wrap the facets for consumption by FreeMarker
Do we want to build a prototype here as well?
Probably need to take OUT the objects that are passed into FreeMarker today so that we don't give anyone the impression we are allowing the use of PlayerCharacter?
Isolation
Core/Output Isolation
Special Ability processing to Strings is happening in PlayerCharacter
http://jira.pcgen.org/browse/CODE-1911
Core/UI Isolation
Stat rolling is happening in PlayerCharacter - isn't this a user like function and thus where should it be?
Events
Rather than subscribing to domainFacet, I think there should be a "display interface" for the core that lets you get a ListFacade<Domain>
It will NOT be an AbstractListFacade<Domain>, but rather one that basically decorates domainFacet and translates the events that way we can keep that translation code localized to that one class (must be event driven per James)
Does NOT duplicate what is in the facades today, just moves it (end up with two "facade sublayers" talking to each other, one that makes "one way in/out" of the core and the other that insulates the UI elements from the core classes
Prototype
Clean up Domain interaction with the UI as a test case for isolation
Make available Domains a list that is processed in the core (allowed perspective)
Also serves as a prototype interface for the Freemarker template use?
See: +DomainWatching.23366.nowork
Awaiting Feedback from James
http://jira.pcgen.org/browse/CODE-1913
Strategic Thoughts
Channels
Long term thoughts when dealing with things like numerical values
Example is something like HANDS on a PC
Modifications that are necessary: Lock, Modify (+-*/), Limit
Need to be able to define a channel, the modify the channel
Need to cleanly address order of operations without having tons of tit-for-tat situations (the PREREQ system of Qualified and :Q: to override that is a tit-for-tat system that has no end
List Channels may also be considered definition, modification
There may need to be a namespace of some form for channels - things like class skill lists for classes could be considered a list, but are in context to a class. Stat value is another one that has a context (the stat)
Spring
Right now we have a lot of hollow objects because Spring only allows single instantiation
How hard would it be to have duplicate - what does it change in XML
What effect would that have on debugging
Effect on cache storage: couldn't use class as key anymore - would need a string of what the facet was to help debug
Other effects?
Strategic Issues
Retroactivity
Retroactive behavior (or lack thereof) is causing issues (and worse, data design changes)
This causes "time travel" problems. What if a level 3 item is retroactive and causes a prereq at level 2 to fail? (or the parent of the retroactive item?)
Context
Globals.getContext is very overused and causes issues with testing
Persistence decodeChoice often requries a call to LoadContext and should be passed in not implicit and grabbed from Globals
This may be fixable with the DataSetID that has been implemented. Many items that have CharID access could use that to get DataSetID and then have a facet that allow conversion of DataSetID to the LoadContext. That would allow a major reduction in overall usage (although how to handle in plugins would be a question since you don't want them directly accessing facets)
Fragility
PCClass.inheritAttributesFrom is fragile, and implies certain behaviors, need a better method of handling this (can overlayCDOMObject work?)
Data Copying and Modification
Need a good strategic method for Variables and Bonuses to handle .COPY= elegantly
Domain Source
DefaultDomainSource
One of the challenges we have in Domains is there is an ability to assign a "default domain source" for the Domain - meaning that it is possible that the intended source is destroyed. It also means that if Domains are added to the "Core View" system that the source may not be correct.
The example as to why a default is a problem is a dual-class Cleric, Druid: If each gets a domain, both of the domains should not be assigned to the default source (whichever class was taken first) So it seems to me that just like some of the other UI elements, we should have a list of the classes (those that allow Domains anyway).
Requirements
Goal: eliminate the default and make Domain sources a full behavior.
We can "pre-select" the class if there is only one that allows domains to avoid another click by the user (this should be a key design criteria)
A class can have itself listed in the Bonus without hardcoding the name (This hopefully has a side effect/benefit on other BONUS tokens of the same nature, and over time we can look at removing some of the code in PCClassKeyChange.changeReferences())
Deprecate BONUS:DOMAIN|NUMBER
The problem is that we have BONUS:DOMAIN|NUMBER If this is always attached to a PCClass, it's not a big deal. If there are bonuses elsewhere, we're pretty much in trouble.
This would be replaced by: BONUS:DOMAIN|x| where x is a class
(1) Compatibility assumes there isn't a class called Number. If we can't make that assumption we can use something like BONUS:DOMAINCOUNT|x|
(2) For any Ability in which the old format is used, we need to add a CHOOSE:CLASS|x and make the BONUS: use %LIST in place of Number. None of the abilities today has another CHOOSE (well, one uses NOCHOICE, but that's no big deal to replace)
(3) Use in the classes themselves becomes a problem. I'd therefore propose we support: %THIS as something referring to itself.
Hardcoding
Magic
TYPE=Natural is magical for Equipment and we need to consider when and how this should drive filtering of objects
Spell isAllowed method hardcodes Potion (defaulting to false) and should be moved to GameMode
PlayerCharacter initializes both INNATE and KNOWN spellbooks by default
PlayerCharacter getSpellRange has hardcoded formulas for spell ranges CLOSE, MEDIUM, etc.
exporttoken.Armor has hardcoded sizes when looking @ additional move
SystemHP.isDNDMassive has hardcoded sizes, also may be buggy because it may be abbreviations that are actually returned?
Constants
Constants.FEAT is specific and should probably be removed
Equipment Locations in Constants need to be extracted to Game Mode
Campaign Enumerations
CampaignID (?) would also be an index in certain enumerations, so those enums (like RaceType) would probably belong in .campaignenum or some such package, and thus use the CampaignID to access the constants. Or perhaps LoadContext needs a .enum as well to hold these simple enumerations (not sure Loadable is worth the effort or has fields that make any sense?)
Major Systems
Primitive/Qualifier
AbilityRefChoiceSet appears to use what is a Primitive in the parenthesis, and has similar/complicated parsing code for the contents. Can/should this be delegated to primitive(s)
Using primitives outside CHOOSE, now part of a proposal, see : https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3760
GUI2
Speak Languages
pcgen.core.Dataset needs a better method to identify the speak language skill
http://jira.pcgen.org/browse/CODE-2014
Language Chooser facade can use the configured controller for skills too
Need to look at breaking the assumptions in language chooser facade of skill vs other
Visual Quirk: is there a reason we sort items after they are displayed?
Administrative
JIRA
Move PrettyList out of CODE-, see: https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3765
Recently Completed
March 2014
Leveling Up
add/remove/add class levels causes a major bug with epic characters, need to consider how this is addressed in 6.3
Template Choose/Level
A Template with a CHOOSE which is added by a PCClass level produces an interaction that causes PCGen to dump stack
Problem is interaction
CHOOSE on a Template is fired regardless of the source. For Ability objects, this is only done with directly selected by the user - otherwise it has to be done by passing in an association
ADD: was made properly symmetric in the 6.3 branch, so fixing a tiny piece CODE-3 is literally what caused this regression
When we add a class level, we increment the level, test if it passes the PREREQs, decrement the level, then re-add the level. This is a disaster from an object thrashing perspective and given how CHOOSE is fired on a template is a problem
Solution is non-trivial
Just fixing that the item can be null (the naive fix) will still cause the CHOOSE to be fired twice
Regressing back the CODE-3 fix just kicks the can down the road on the issue
Should probably justify the difference in CHOOSE behavior vs Ability in any solution
Strategic solution would be a better method of doing incrementing levels, but that is a "hard" problem, in that it is complex
On option is to take a page from the ADD: system when a Kit is being applied or a PC is being loaded and set a variable to say "don't process this stuff right now". That can be activated for two different situations: Both the Template Choose/Level problem as well as the epic interaction
Token Abstraction
AbstractBoolean
Global
Descispi
nameispi
Template
Removable
Stat
Rolled
Alignment
valid for deity
validforfollower
Skill
Useuntrained
Exclusive
Size
isdefaultsize
PCClass
IsMonster
memorize
spellbook
allowbaseclass
modtoskills
EqMod
AssignToAll
costdouble
cmod
usemasterskill
campaign
showinmenu
islicensed
ismature
isogl
Ability
Mult
Stack
AbstractString
Template
AppliedName
PCClass
ClassType
ItemCreate
Abb
EqMod
FumbleRange
Equipment
RateOfFire
Deity
Worshippers
Title
Symbol
Appearance
Campaign
Setting
PubNameWeb
PubNameShort
PubNameLong
Help
Genre
Ability
AppliedName
Global
Sortkey
Sorting Associations
PC.getExpandedAssociations needs Collections.sort(ret);
See: CODE-2533
Or is there a better place to ensure this (Description, Aspect, SA, output tokens all do this stuff - and some rely on ChooseInformationUtilities)
Implement association ordering, see https://groups.yahoo.com/neo/groups/pcgen_developers/conversations/messages/3897
April 2014
Conditional Skills Work
Facet Structure
NonExclusiveSkillsFacet: New, gives the non-exclusive skills into the CCSKILL list
MasterToSkillCostFacet: imports the costs on the global objects when a PCClass is added to the PC
SkillFacet (model) needs to meet the following criteria: Has a Rank, has a BONUS:SKILLRANK, non-exclusive and useuntrained
BONUS:SKILLRANK is the main problem in that the current Bonus notification system is not aligned to do an entire bonus type. The current patch does not detect BONUS:SKILLRANK
GUI2 does not properly listen for BONUS:SKILLRANK as it ignores TYPE= situations
Need test fix (Quasvin's PsiCrystal).