================================ VERSION 3.0 =================================

================================ VERSION 3.1 =================================

---------

From Michel Nizette <mnizette@ulb.ac.be> [9/1/06]:

Would there be any value in implementing pending conversations via
regular AgendaItems, like we just did for the boredom handling?
Again, the advantage would be simply to provide more control over the
order of priority of an actor's various pending activities.  Also,
should this prove useful, we could then think of adding an
Actor.scheduleConditionalInitiateConversation method, which would do
the same thing as scheduleInitiateConversation, but would additionally
specify an extra condition to be checked as part of the AgendaItem's
isReady property.  The condition could be specified simply as a
callback function that returns nil or true. 

---------

From Michel Nizette <mnizette@ulb.ac.be> [9/1/06]:

Currently, the method Actor.noteConversation is invoked each time a
TopicEntry is triggered, no matter if the topic is conversational or
not (i.e., marked with isConversational = true or nil).  The
noteConversation method does a number of things like marking the actor
as having conversed on the current turn, and resetting the boredom
count.  But I'm wondering if this is really what we want in the case
of a non-conversational topic, whose typical purpose is to describe a
refusal to engage in a particular conversation.  If the conversation
doesn't actually take place, should there really be any effect on the
NPC's behavior?  It would seem a little odd to see an NPC delaying a
pending conversation for several turns just because he was involved in
a sequence of conversational commands that all got blocked in the PC's
mind for motivational reasons and didn't result in a real dialogue,
for example. 

In contrast, noteConversation also does a few things that
unquestionably make sense for non-conversational topics, like updating
pronouns and establishing a default interlocutor: these are interface
features independent of what happens in the game world. 

So, my suggestion would be to distinguish between the situation where
a conversation actually takes place (in which case it makes sense to
call noteConversation) and the situation where the conversation is
only attempted, which we could handle via a new
noteAttemptedConversation method that would only set pronouns and the
default interlocutor.  The library could then simply invoke
noteAttemptedConversation everywhere it currently invokes
noteConversation, and could additionally call noteConversation in the
few places where a real conversation takes place, like the triggering
of a conversational topic or a call to Actor.initiateConversation.  If
maintaining backwards compatibility is important, we could have
noteAttemptedConversation call noteConversation by default, but we
could make this controllable via a switch. 

---------

From Krister Fundin <fundin@yahoo.com> [9/3/06]:

> bob, follow me
"Okay, I will follow you."
> g
"Okay, I will follow you."

The library comments suggest that the latter should not happen. The
problem is that Actor.nonIdleTurn() cancels following whenever the
actor receives an order. Thus the check for already being in following
mode is always negative.

[mjr: I actually think the current behavior as shown is fairly
reasonable.  It would be marginally better to have a message like "I'm
already following you."  But short of repeating the command ten times
in a row for comic effect, I'm not sure players would ever be bothered
by the present behavior in practice - if they're repeating the command,
they're presumably not sure that Bob was already in follow mode, so
they're not going to be bothered by another acknowledgment.]

I had already removed this cancellation from one of my projects, since
it wasn't appropriate there, but that seems to cause a different
problem instead:

> bob, follow me
"Okay, I will follow you."
> g
Nothing obvious happens.

In this case, Bob can already see the player and thus only makes a
call to checkMovingActorInto(), which does nothing since he's already
where he's supposed to be.

[mjr: This part is worth looking into whatever the disposition of the
first part, since the nonIdleTurn() comment suggests that it's okay to
override and not cancel follow mode, but apparently that's
problematic.]

---------

From Krister Fundin <fundin@yahoo.com> [9/3/06]:

just a thought (feel free to ignore it):

> bob, follow me
"Okay, I will follow you." Bob stands on the main platform.

versus:

> bob, follow me
Bob stands on the main platform. "Okay, I will follow you."

Doesn't the latter sound marginally better? 

[mjr: I'm not sure about this; Bob acknowledging first actually strikes
me at the moment as the better sounding version]

I guess in practice there would be a command separator between the two
reports, but that could actually improve things when two or more
implicit actions are needed:

> bob, follow me
"Okay, I will follow you." Bob stands on the main platform.

Bob stands on the red platform.

versus:

> bob, follow me
Bob stands on the main platform.

Bob stands on the red platform.

"Okay, I will follow you."

---------

From Krister Fundin <fundin@yahoo.com> [9/8/06]:

How about adding a command joiner report and corresponding pseudo-tag?
Any combination of joiner and separator or separator and joiner would
just remove both. I can think of a few other places where I could use
one of those.

[i.e., something like <.commandjoin> to cancel out <.commandsep>]

---------

From Krister Fundin <fundin@yahoo.com> [9/13/06]:

   > push television into white box
   (first standing on the main platform)
   Okay, you're now standing in the white box.

Perhaps it's not serious enough that it needs to be fixed now. In the
long run, one could imagine a "first trying to push the television
onto the main platform" type of implicit action here, but that would
take some work.

[I'm not even sure what would be right here.  I guess some
fine-grained variation of PushTravel for pushing into nested rooms or
something.]

---------

The ThingState mechanism should be extended to allow a single object
to have multiple current, active states.  For example, it should be
possible for an object to be simultaneously worn, lit, and open.  As
it stands, an object can only have one state at a time.

On the parsing side, this is straightforward: we simply have to
iterate over the list of active states and consider each one's token
selectivity in turn.  All active states must rule positively on the
token list to allow a match.

On the output side, it's slightly more work.  We'll have to list
multiple states in messages like "(providing light)", which isn't
hard: "(worn, open, providing light)".  We'll also have to list
multiple states in sublists: "one worn, open, and providing light; one
closed."  This is also relatively straightforward except that we'll
have to go to semicolons in the overall list when the sublists start
expanding out to the point that they need commas.

(Ideally, ThingState probably ought to be the *only* state mechanism
for any sort of state for which it's used.  For example, there really
shouldn't be a separate isOpen property - an object's open/closed
status instead ought to be determined by the presence of an openState
or closedState object in the object's state vector.  Opening an object
would then be a matter of calling obj.setState(openState).  This would
require that each ThingState know its set of states, for mutual
exclusion when a new state is set (which would simplify things
slightly in that we'd be able to dispense with the per-object
allStates list.)

(Also ideally, ThingState sets would implicitly and automatically
create distinguishers.  Any time we have a pair of objects with
different active states from a given set, we should automatically be
able to distinguish those objects on the basis of that ThingState
set.)

---------

In announceAmbigActionObject, announceMultiActionObject, and
announceDefaultObject, it'd be ideal to use distinguishers to decide
on the object name.  This would ensure, for example, that we generate
"(the lit candle)" when litness would distinguish the candle we chose
from others present.  

I think we'd basically have to do the same work that
BasicResolveResults.ambiguousNounPhrase does to choose a distinguisher,
but in this case we want to tell apart the object in question from
everything else in scope that has the same name.  So, probably we want
to go through the objects in scope and build a list of everything
with the same disambigName; then we want to search for a distinguisher
that can tell apart the object in question from all of those other
objects; failing that, we could settle for a distinguisher that can
tell apart the object in question from any of the other objects, as
this would provide at least some greater specificity.  Failing that,
we just use the disambigName.

---------

Consider removing several modules from adv3 and putting them in the
"system" library instead:

  banner.t
  menusys.t
  modid.t
  numbers.t

---------

Add a property setting that makes the contents of an item listed
immediately after the item's specialDesc, if it's using a specialDesc.
The contents would be listed there rather than in the miscellaneous
portable list paragraph.

---------

Add static File.deleteFile(filename), to delete a file from disk.
This operation should require WRITE privileges (as determined by the
file safety level setting) for the given file.

---------

Add a version of saveGame() and restoreGame() that use a ByteArray as
the saved state medium.  save() would create and return a new
ByteArray object containing the saved state (which must, of course,
exclude the new ByteArray itself); restore() would take a ByteArray
object and restore its contents as the current game state.

For interface purposes, we could simply let saveGame() without an
argument return a ByteArray, and restoreGame() could take a ByteArray
as its argument.  Alternatively, saveGame() could take a pre-created
ByteArray as its argument, and we'd fill in that ByteArray with the
saved state (but we'd somehow have to make sure that this ByteArray
was excluded from the saved state).

(Implementation note: this is fairly straightforward, but some work.
We first have to refactor CVmFileSave::save() and restore() to take
CVmStream objects instead of CVmFile objects; this is fairly simple,
but requires changes down through the whole save/restore stacks, which
touch a lot of objects.  Then, where we currently call save/restore
with CVmFile objects, wrap the CVmFile in a CVmFileStream and make the
same call.  Finally, we'd have to create a new CVmStream subclass that
operates on ByteArray objects, which should be a fairly easy mapping.
It'd be worth making sure that changing CVmFile to CVmStream in
save/restore doesn't balloon up the executable too much - this would
introduce a ton of new virtual calls, because all of the CVmStream
methods are virtual while the corresponding CVmFile methods are
non-virtual.  It'd also be worth measuring any degradation in
save/restore performance from all the new virtual calls.)

---------

Workbench: add a "ZIP source files" command that creates a .zip file
containing all project source files and all necessary directories to
compile.  The idea is that this file can be unzipped on another machine,
and the project can then be built there without any further fuss.

---------

Store ALL objects as potential future pronoun antecedents when
processing a multi-object command.  For example, when executing UNLOCK
DOOR WITH KEY, store *both* the door and the key as potential
antecedents for 'it'.  On using 'it' in a future command, look at all
of the possible antecedents, and choose the most logical one -
essentially treat the set of possible antecedents as though it were
the set of matching objects for an indefinite noun phrase.  If there's
more than one equally logical choice, choose one arbitrarily, the same
way we would if the player typed TAKE ANY BOOK:

>unlock door with key
Unlocked.

>drop it
Dropped.           // refers to the key, since the door is illogical for DROP

...but if the next command were this instead:

>open it
Opened.            // refers to the door, since the key is illogical for OPEN

...and if it were this instead:

>x it
(the door)             // both door and key are equally logical, so choose
It's a massive...      // one arbitrarily, and report the choice

---------

Debugger: display the current method (the top of the stack trace) in
the status line.

---------

Debugger: add an option to skip over library code when stepping.  This
could be a mode, or it could be a new 'step' command (alongside 'step
in', 'step over', 'step out').  This would make it easier to isolate a
bug in game code, and it would also clarify the control flow through
the sequence of game-provided entrypoints.

The main question in implementing this is how exactly to define
'library' code for the purposes of this new form of stepping.  Library
code could be defined as anything from a .tl file, or as any source
file under the system directory tree, or as files designated as such
manually (with a right-click on a file in the project tree, say).

---------

Can we rephrase the UNDO command results so that they reflect the logical
command rather than the actual text typed in?  Something like:

 >undo
 Taking back: attack the troll with the sword

(For commands involving multiple objects, this could get cumbersome, since
we probably wouldn't be able to easily reconstruct the original ALL or
plural usage, but would instead list all of the objects.)

---------

Maybe use <.Q> rather than <Q> in msg_neu and en_us, to allow customization
of quote styles.

---------

Add a global compiler options setting to workbench?

---------

The compiler should flag some types of errors at the location of the
particular token causing the error, rather than at the location of
the enclosing source line.  Krister Fundin <fundin@yahoo.com> pointed
this out with respect to an expression like this, where the 'method'
name is undefined:

   property.method(new function(x)
   {
        code;
   });

The error ought to be reported at the exact line containing the
undefined token 'method', rather than at the statement level - which,
in this case, reports at the *end* of the nested function.

To accomplish this, the compiler would have to (a) save the source
location for each individual token, in addition to the location of the
current statement; and (b) use the token location rather than the
current line when reporting any error that can be associated with a
token.  [The second part is probably pretty easy - almost all such
cases can probably be caught by changing
CTcTokenizer::log_error_curtok() to report at the current token
position.  Saving the token position probably isn't too hard, either,
but there might be some subleties, such as with respect to the
preprocessor.]

---------

Add a general "warning check" mechanism to the library.  This would
be an add-on module that would encapsulate a bunch of checks for
common pitfalls, using reflection mechanisms to display warnings.
Authors could optionally compile with this module from time to time
to check for common errors.  This would only be of interest to
authors during development, of course - it would always be removed
for the final build for release.  (This came up in the context of
checking for default Container objects when using the liquid add-on
library.  It would be nice to have a generic mechanism, or at least
some coding patterns to follow, that would simplify adding this kind
of warning check.)

---------

For ResolvedTopic, it might be nice to ask for disambiguation for
in-scope objects if there are more than one, especially if this were
parameterized.

---------

Menu-based conversation framework

---------

For travel, add a check to see if the actor fits through the passage.
This should ideally be implemented using a precondition that would
check the actor's bulk to ensure it's below a maximum.  The actor
would have a chance to reduce its bulk via a tryReducingBulk call,
perhaps - this would let the actor do something obvious like close
an umbrella.

---------

For travel, add a check for excessive encumbering bulk or weight.
This could be used for things like climbing ropes or walking across a
rickety bridge.  For excessive encumbrance, the actor could use the
standard tryMakingRoomToHold mechanism to reduce held bulk.  For
excessive weight, this is probably just a connector-specific check
and message on failure, so there might not be much we can do in the
library other than make sure the framework is in place.

---------

Weapons (attack with)

---------

flammability as a mix-in (for paper, wood, and the like)

---------

Liquids and bottles

---------

Drinkables (special kind of liquid)

---------

Probably need a variation of objHeld for liquids that lets us specify
that we need to be holding the container of the liquid (for 'pour',
for example).  containerHeld?  Or, we could do like we do with keys
on keyrings, and consider a liquid within a container being held to
be itself held.

---------

If we have a bucket or some other container filled with liquid, and
we try to put another item into the bucket, we should run a special
check for wetting the item.  The default should probably just be to
say "you don't want to get that wet" and refuse the command.
Similarly, "fill bucket" when the bucket contains non-liquid objects
should run the same checks.

---------

Burning and liquids: wetting a burning item with a non-flammable
liquid should extinguish the flame.

---------

Burning and liquids: provide a flammable liquid class.  This should
consume the liquid at a particular rate while burning.  The basic
flammable fuel consumption mechanism should be used here if possible.

---------

Provide an extensible "chemistry" framework for combining different
kinds of liquids.

Perhaps we could define an object to represent the operation of
combining each pair of liquid classes - this class doesn't represent
the mixture, but rather the operation of mixing.  So, for example, we
define an object for the operation of mixing water and liquid
nitrogen.

Each of mixing object must define a method that is called when its
pair of liquids are mixed; this method encapsulates the action of
mixing the liquids.  Use a global lookup table of these objects,
keyed by liquid class pairs; when we want to mix two liquids, we look
up the pair of liquid classes to be mixed and find the mixer object,
then call its mixer method. 

---------

Putting a container with liquid into another container in inventory
should have some conditions attached.  First, an openable liquid
container should be closed whenever moving it into a sub-container
(so a bottle of water should be closed before putting it in your
backpack) to avoid spilling.  Second, a container that can't be
closed should probably not be allowed to be moved to a sub-container.
Third, taking or otherwise moving any item should probably be
disallowed if the item contains an open container of liquid.

---------

Pouring a liquid onto the floor, into a container that isn't set up
as a container for liquids, onto a surface, or onto an arbitrary
object should all do something reasonable.  It's unclear what to do
in general; the easiest thing would be for the liquid evaporate
immediately, but that's not all that realistic.  Pouring liquid onto
an arbitrary object should probably call the object's
notify-getting-wet method, just as though the object were in a
container being filled (although we might want to differentiate
between submerging and dousing, maybe just via a flag to the
notify-getting-wet method), but then something needs to happen to the
liquid, even if the thing that happens is simply evaporation.

---------

Leaky containers: when a leaky container has liquid added, it should
set up a daemon to cause it to leak slowly.  The liquid leaking out
should do whatever liquid does when poured into the leaky object's
enclosing location.

---------

Absorbent materials: these should retain a quantity of liquid as
though they were a liquid container, allowing the absorbed liquid to
be recovered by squeezing the object.

---------

Ropes, to the extent they can be generalized

---------

In two-object actions (including topic and literal actions), when
both objects are missing, we might want to consider some alternative
way to specify which object we ask for first.  This might vary
by language, so it might be better to have an action property that
specifies which goes first when both are missing.  Or maybe this is
the same as resolution order, or opposite of the resolution order,
or something.

---------

OBJECTS command - list all portable objects that have been seen, and
the last location where they were seen

---------

It seems like it should be possible to talk to an actor in situations
when the actor is audible only, such as in the dark.  But how do we
initiate contact if we don't see the actor?

1 - broadcast a greeting:  

>hello
Bob says, "Is that you?  I can't see anything."

2 - try talking to a specific actor:  

>bob, hello

This doesn't seem very good because we need to have a way of knowing
Bob is present in order to talk to him.

3 - the actor hears you come in and says he's present:

>n
In the Dark
It's pitch black.

In the darkness, you hear Bob.  "Is that you?" he asks.

---------

Add new syntax:

  x.delegated y.prop(args)

This syntax explicitly targets 'y' with self as 'x'.  This differs from
the regular 'delegated y.prop()' in that the latter can only set 'self'
in the delgatee to 'self' in the delegating method.

Or, perhaps we could take this one step further, by allowing the
target object, self, *and* defining object to be independently specified.
Something like

  x.delegated y.inherited z.prop(args)


This would set self to x, target object to y, and defining object to
z, then try to find the method in z.  If the method doesn't exist in
z, we should probably throw an error rather than try to inherit from
there, since this is so fully specified already.

We would need a new opcode for each case.  The first would require an
opcode with the new self, the target object, and the target property
as operands; the second would require those operands plus another for
the defining object.  We could have two variations of each opcode
(one with a fixed property ID and the other with a property pointer
operand from the stack), or we could presume that the operands would
be rarely used and define only the property pointer version (it would
be less efficient for cases with a static property ID, but if the
opcode is rarely used, this extra cost shouldn't be important).

---------

Add a Set intrinsic class?  (For unordered lists, optimized for inclusion
testing: essentially a LookupTable with only the keys.)

---------

Soft linking:

   local x = ifdef(obj);  // if obj is linked in, return obj, else nil

   replace ifdef obj ...  // same as a new object def if obj isn't linked in

   modify ifdef obj ...   // ignored if obj isn't linked in

---------

List methods: add compare funcs to more methods:
   - intersect
   - getUnique
   - appendUnique

likewise:
   - Array.getUnique
   - Array.appendUnique
   - Vector.getUnique
   - Vector.appendUnique

---------

Compiler: equivalent property list syntax: 

  a, b = foo
  c, d(x) { bar }

---------

Add items that are by themselves ambiguous?  This might be interesting
for things like PO Boxes:

>look in box
Which box do you mean, the cardboard box, or a mail box?

>mail box
Which mail box do you mean?  They're numbered from 23000 to 23999.

>23001
Box 23001 seems to be locked.

This could be accomplished with an isAmbiguous flag in the object,
which would be similar in effect to isEquivalent.  Whenever the object
shows up in a list with multiple objects, we use an indefinite article
with it, as though it were one of a bunch of equivalent items.  When
the object shows up by itself, we ask a single-object disambig
question, perhaps by calling a method on the object itself -
disambigPrompt?

Ultimately, matchName() and matchNameDisambig() would be responsible
for coming up with a specific object based on the noun phrase.

---------

repeated messages: gather up and group defaults?

---------

Sound defaults.  Add a background sound resource property to each
object.  On each turn, the library scans for all objects in hearing
range and plays the associated background sound for each one.  Could
have near/far/obscured versions, or maybe simply adjust the volume,
or whatever.

---------

Graphics defaults.  Add a graphic resource property to each room and
item.  On "look" and "examine", display the associated graphic as
part of the default room/item description.  It would be especially
nice to have a choice of presentation styles, maybe even including a
Legend-style split-screen interface (but that might be more
appropriate as an add-on; even so, it would be good to keep it in
mind so that the necessary hooks are there for creating a
Legend-style presentation add-on).

---------

Add a music soundtrack scheme that simplifies management of background
music.

---------

Look into formally structuring things into try/check method pairs
(checkHolding/tryHolding, etc), to make precondition implementation
more consistent and clear.  This would apply most to the
preconditions.

---------

Krister Fundin <fundin@yahoo.com> proposes: The list of end-of-game
options currently specified as the 'extra' parameter to the
finishGame() and finishGameMsg() functions could instead be bundled up
into a FinishType object, so that the message and extra options could
be specified with just one object parameter.  This would make
game-ending code more concise and also make it easier to reuse an
option list.

[We'd probably need a new common function that takes a FinishType
object as its sole parameter; the existing interface functions would
simply create a dynamic FinishType object to encapsulate the
parameters and call the new common function.]

---------

Krister Fundin <fundin@yahoo.com> proposes allowing a list parameter
in the middle of a parameter list, rather than only as the last
parameter, as in foo(a, [b], c).  The compiler could generate function
prologue code that moves the trailing arguments into locals, then
truncates the list created via PUSHPARLST.  It's a little inefficient,
since we just throw away the first list and have to move the locals
around, but the case is probably sufficiently rare and non-critical
that this doesn't much matter.

---------

Andreas Sewe <2003-10-01.as.int-fiction@web.de> proposes that
'inherited' in templates should NOT implicitly expand to nothing.  As
it is, 'inherited' in a template expands to each inherited superclass
template, but also expands to nothing.  Andreas suggests that this
should be changed to eliminate the empty expansion, AND the empty
template should be allowed, so that it's possible to make inheritors
explicitly pick up the empty case.  [It's not entirely clear to me
that the empty case really needs to be inherited in the first place,
but Andreas's point is that it *could* be there to ensure backwards
compatibility with the existing template structure, which implicitly
does inherit the empty case.]

