CU Amiga Super CD-ROM 23/CU Amiga - Super CD-ROM 23 (June 1998).iso/CreatingGames/GameCreators/TADS/Version.doc

Text File | 1996-11-22 | 159.2 KB | 3,535 lines

TADS 2.2.1.0
Unix Patchlevel 1
18-Oct-96

Mike Roberts has made source code to TADS generally available, so I have
merged my Unix changes into the general source distribution as of this
version.

This version also includes a bug fix:

Fixed a bug in the command parser that caused commands ending with numbers
(e.g., "footnote 1", "undo 5") to confuse the parser.

Also disabled swapping by default, since Unix machines will do better using
their virtual memory.  To force swapping, use the -t+ option.

See tadsver.dos for changes from 2.2.0.5 to 2.2.1.0 that apply to all
systems.

TADS 2.2.0.5
Unix Patchlevel 1
22-Nov-94

Fixed a bug that caused window sizes > 135 to crash the run-time.
Increased default buffer sizes.  The defaults that the compiler reports are
actually wrong.  You probably won't have to make any buffers bigger unless
you've got a *really* huge game.

See the general 2.2 changes below; there are many important new features.

TADS 2.1.2.2
Unix Patchlevel 2
17-Feb-94

Unix pacakges now include the standard documentation.

Fixed a bug that caused commands like

>X, give me the Y

to crash the run-time.


TADS 2.1.2.2
Unix Patchlevel 1
29-Jan-94

Unix-specific changes applied to latest TADS source.  Unix versions 2.1.0.0
pl2 through pl4 were never released, so please read the notes for those as
well -- this is the first release version that incorporates those changes.

The window-resizing code seems to have been broken.  Resizing the window
sends the run-time into space.  I'm looking into it.  For now, don't change
the window size while playing.

adv.t and std.t have been updated to 2.1.2.2 as well.


TADS 2.1.0.0
Unix Patchlevel 4
16-Nov-93

Changed the runtime and compiler so that the heapsize is always set to the
maximum.  An explicit -mh setting will override this.  Note that you cannot
make the heap bigger than 64K, since TADS has to run under DOS in real
mode, where data structures can only be 64K or smaller.


TADS 2.1.0.0
Unix Patchlevel 3
25-Oct-93

The adv.t and std.t files have been updated to 2.1.0.0.  The ones in the
earlier archives were actually the 2.0.13 ones.  (Doh!)

Now tadsr tries to save the game to file "fatal#####.sav" when it gets a
fatal signal, where ##### is the process ID.  Sometimes this file will be
unuseable; other times it'll work fine -- it depends on where the signal
occurs during execution.

Two environment variables now affect the behavior of the runtime:

    TADSSAVE: directory for save files
        TADSGAME: directory for game files

These pathnames will be prepended to any save or game file that does not
already have a slash in it.  For example, if

        TADSGAME=~you/games

and you type

     tadsr foo

the run-time will look for the file ~you/games/foo.gam.  Note that if you
want to prevent this, you have to explictly put a ./ in the filename, or
unset the TADSGAME environment variable.  The same goes for TADSSAVE.

Since some windowing systems clear the screen when the end-of-visual
terminal code is sent out, tadsr now waits for a keypress before it
terminates.  There is no way to disable this -- the run-time now ignores
the -p flag entirely.  You can always see usage (and game endings) now,
however, which is a plus!

The Emacs-style command line editing now handles ^D (delete character after
cursor) and ^K (delete to end of line).

Since most Unix boxes have loads of real memory and even more virtual
memory, I increased the size of the scrollback buffer eightfold.  It's now
256K.  This should be enough for marathon play sessions.

There are some known problems with the compiler -- core dumps on at least
one big game, and core dumps when dealing with precompiled headers and
modify/replace in another case.  I haven't been able to track these down,
but they aren't restricted to one architecture.

One thing that's been reported as a bug but isn't: If you have the margin
bell on in your xterm you'll get lots of rings as TADS updates the screen.
The solution: turn off the margin bell!


TADS 2.1.0.0
Unix Patchlevel 2

Output handling has been improved again.  Terminal settings are restored
properly upon suspend (^Z) or resume.  I improved a few other things that I
don't remember now.  :)


TADS 2.1.0.0
Unix Patchlevel 1

The programs should now be quite stable on all supported platforms.
Term handling has been vastly improved, so that the run-time should be
comfortable over a 2400 baud connection.  The compiler now avoids
all the screen handling stuff and just uses stdio for all output.

Starting with this release, Unix versions will be given patchlevel
numbers in addition to the standard High Energy version number.
There's one patch number of the general Unix sources, and another for
the specific platform the binaries were compiled for.

General Unix notes:

The .GAM files are now 100% portable across platforms.  You can compile
a game on a PC clone or Mac and then run the resulting .GAM directly
under Unix (and vice versa).  Kudos to High Energy for that!

There is no maketrx program for the Unix versions.  Now that TADS is
running on a much wider range of machines, creating binaries specific
to a particular machine seems rather foolish, so I have not kept
support for doing that in the this port.

The TADS run-time uses termcap routines for all output.  You will
therefore need to run it on a terminal with window scrolling capability
(i.e., most modern terminals).  If you find that the programs don't
work properly with your terminal, please let me know -- termcap hacking
is very error-prone, so I may have made a mistake in there somewhere.
Before assuming that the problem is with the TADS run-time, however,
try other programs on your system that use termcap, like vi.  (Emacs
has its own term handling code, so it's not a good test.)

There a few things special to the Unix versions:

    o (Limited) Emacs-style command line editing:

        ^F    ahead one character
        ^B    back one character
        ^A    beginning of line
        ^E    end of line
        ^P    previous command in command list
        ^N    next command in command list
        ^K    kill to end of line
        ^D    delete character after cursor

    o ESC enters scrollback mode (instead of F1).  Scrollback control
      keys are listed on the top line.

    o Press ^L at any time to redraw the entire screen (if you
      get a burst of line noise, for example)

    o Press ^Z (or whatever your SIGTSTP key is) to suspend the
      run-time.

    o tadsr recognizes SIGWINCH so that you can resize your
      terminal window.  Text above the current line won't be
      reformatted when you do this, but all subsequent text
      will be formatted to fill the screen.

      NOTE: Resize doesn't work in an OpenWindows cmdtool.

The output handling is still not absoultely optimal, but it's probably
as good as it's going to get since the machine-independent TADS source
imposes certain requirements on the behavior of the windowing functions.

Finally, please report bugs directly to the person listed as the
maintainer for the port in the banner, not to High Energy.  That person
is the only one (besides, perhaps, me) who's going to be fixing any
problems specific to the Unix ports.

Dave Baggett
[e-mail]

*-----------------------------------------------------------------------------*
General TADS notes (From the DOS version)
*-----------------------------------------------------------------------------*

This file contains a list of changes that have been made to TADS
since the initial 2.0 release.  Most of the changes are fixes to
bugs, so they don't change the documented behavior, but a few, as
explained below, add new functionality to TADS.  Releases are
listed with the most recent release first; each release incorporates
all new features and bug fixes of each prior release unless
otherwise stated.

2.2.1.0  10/01/96  minor corrections, numbered object enhancement

  - Note that there has never been an "official" version 2.2.0.5;
    some versions with that designation have appeared on a few
    platforms to expedite the release of some bug fixes, but 2.2.0.5
    was never universally released.  To eliminate any confusion, this
    release is designated 2.2.1.0, and should incorporate all of the
    bug fixes of and be upwardly compatible with any versions labelled
    2.2.0.5.

  - The player command parser has a new feature that allows you to use
    numbered objects without specifying in advance an object for each
    possible number.  For example, if you want to design an elevator
    with 100 buttons for floors, you can now create a single button
    for all of the 100 buttons.

    The new feature uses a special new adjective property setting, '#'.
    For example:

         elevatorButton: fixeditem
             noun = 'button'
             plural = 'buttons'
             adjective = '#'
             location = elevator
         ;

    The special adjective '#' tells the parser that the object can be
    used with any number.  The parser accepts the syntax "button 100"
    and "100 button" as equivalent, just as it would if you had specifically
    included '100' as an adjective with the object -- in other words,
    the adjective '#' behaves the same way that regular numeric adjectives
    behave.

    When the player refers to the button in a command, the parser requires
    a number to be used.  If the player refers to the button without a
    number, the parser displays error number 160, "You'll have to be more
    specific about which %s you mean" (where "%s" is replaced with the
    phrase the player entered that referred to the button).  You can
    customize message 160 with parseError as with any other message.

    If the player properly enters a number with the object name (by
    typing, for example, "button 99"), the parser calls a new method in
    the object:  newNumbered(actor, v, num) -- "actor" is the actor
    for the command (Me if the command was not explicitly directed to
    another actor), "v" is the verb object, and "num" is a number
    typed by the player.  "num" can also be nil:  if the player refers
    to the button with a plural ("look at buttons"), the parser calls
    newNumbered with nil instead of a number to indicate that the player
    wants to refer to all such objects collectively.

    The newNumbered method must return either an object or nil.  If it
    returns nil, the method should first display an error message -- the
    parser will simply cancel the command without any further messages
    if newNumbered returns nil.  If newNumbered returns an object, the
    parser uses this object (instead of the elevatorButton object itself)
    as the numbered object.

    adv.t includes a new class, numberedObject, that you can use to
    implement numbered objects.  This class provides a newNumbered
    method that creates a copy of the object, and sets its value
    property to the number the player used to refer to the object --
    for example, if the player types "button 99," newNumbered will
    create a copy of the button, and the copy's "value" method will
    return the numeric value 99.  The newNumbered method defined in
    numberedObject calls another method, num_is_valid(num), to
    determine if the number is valid; if this method returns true,
    newNumbered proceeds, otherwise it returns failure.  By default,
    num_is_valid(num) returns true; you can override this method if
    you want to restrict the range of valid numbers.  For example,
    you would define the elevator button's num_is_valid as follows:

        num_is_valid(num) =
        {
            if (num < 1 or num > 100)
            {
            "The buttons are numbered from 1 to 100.";
        return nil;
            }
            else
            return true;
        }

    The newNumbered method returns the original object by default if
    the player refers to the object with a plural; if you wish to use
    another object to handle plural references, override the method
    newNumberedPlural(actor, v) to return a different object, or nil
    (after displaying an appropriate error message) if you don't want
    to allow plural references at all.  You may want to handle plural
    references with a different object if you want plural references
    to work very differently -- in other words, if you want actions
    taken on all such objects collectively to have special handling.

    The numberedObject defines dobjGen and iobjGen routines that
    don't allow any actions to be taken on the object in the plural;
    they do this by checking to see if the value property is nil,
    which indicates that this is the original (plural) object.  These
    methods simply display "You'll have to be more specific about
    which one you mean" and otherwise ignore the command when the
    player uses the plural object.  You may want to override this
    behavior for some verbs.  For example, the elevator button's
    dobjGen should probably be defined like this:

        dobjGen(a, v, i, p) =
    {
        if (self.value <> nil or v <> inspectVerb)
            inherited.dobjGen(a, v, i, p);
        else if (v == inspectVerb)
        {
            "The boxes are numbered 1000 to 9999.";
        exit;
        }
    }

    This allows the player to get some useful information from "look
    at buttons," but doesn't allow commands such as "push all buttons."

    Note that the implementation of newNumbered in numberedObject
    automatically sets a fuse to delete the newly created object at
    the end of the command.  Therefore, you can't use the new objects
    to contain any state that you want to persist after the end of the
    command.

    For the elevator button example, the only thing you really need to
    remember is whether a particular button is lit up or not.  You could
    do this with a list.  When the game is first starting up, you could
    initialize this list like this:

        local i;
        elevator.button_list := [];
        for (i = 1 ; i < 10 ; ++i)
            elevator.button_list += [nil nil nil nil nil nil nil nil nil nil];

    Now you can define the button's doPush method like this:

        doPush(actor) =
        {
            if (elevator.button_list[value])
                "It's already lit.";
            else
                elevator.button_list[value] := true;
        }

    The parser uses one additional method in some circumstances.  When
    the player uses "any" with an object whose adjective list contains
    '#', the parser evaluates the property anyvalue(n) for the object.
    The parameter "n" is a number indicating how many of these objects
    has been requested so far for this command.  Currently, it's always
    1; in the future, a command such as "push any 3 buttons" may call
    this method three times, with "n" set to 1, 2, and 3, respectively,
    for the three calls.  For the time being, "push any 3 buttons" is
    treated by the parser as identical with "push buttons".  The
    default implementation of anyvalue in numberedObject in adv.t
    simply returns the same number "n" that it received as a parameter;
    if your valid number range doesn't start with 1, you should override
    this method to return a number in your object's valid range.

  - The compiler now generates more useful errors for undefined objects.
    In particular, errors for objects that are used in class lists but
    never defined now include the name of the undefined object and the
    source file location of the first use.  In addition, the compiler
    now lists all undefined object errors with a line number prefix giving
    the actual line of the error, rather than providing that information
    parenthetically in the error message as it did formerly.  In
    addition, the compiler now checks for undefined objects even when
    other errors are encountered, increasing the likelihood that you
    can find all errors on the first compile.

  - The player command parser now allows a word to start with a dash or
    apostrophe (these characters have always been allowed within words;
    now they're allowed at the starts of words, too).

  - The system allows format strings (such as %you%) to be used while
    the init() function is active, and uses Me as the current actor for
    translating the format strings.

  - Under certain circumstances, the compiler stored the same object/word
    relationship twice for a vocabulary word, causing the parser to list
    the same object more than once in a disambiguation question ("which
    book do you mean...").  This has been corrected.

  - The parser's handling of pronouns (it, him, her) has been changed
    slightly.  Previously, the parser always replied "I don't know what
    you're referring to with 'it'" if the antecedant was not accessible
    for the new command (the validDo/validIo routine returned a false
    indication).  The parser now instead uses the standard cantReach
    mechanism that would use if the command were typed with the object
    explicitly.

  - The runtime now calls preinit() prior to calling the restart
    callback function if one was provided and the game was compiled for
    debugging.  (The runtime previously called preinit() *after* calling
    the restart callback when the game was compiled for debugging, which
    meant that the game state as seen in the callback was different when
    a game was compiled for debugging than for the same game compiled
    with debugging off, because with debugging off, preinit is run by
    the compiler.  This change makes the state seen by the restart
    callback function the same for both versions of the game.)

  - Assignments through property pointers now work properly.

2.2.0  10/30/94  new features, enhancements, bug fixes

  - TADS now has support for reading and writing files.  This new
    feature is intended to let you save information independently
    of the game-saving mechanism, which allows you to transfer
    information between sessions of a game, or even between two
    different games.  The TADS file operations are not designed
    as general-purpose file system operations; in particular,
    these new functions don't have any provisions for creating or
    reading formatted files, or for exchanging information with
    programs other than TADS games.

    To open a file, use the fopen() function.  This function takes
    two arguments:  a single-quoted string giving the name of the
    file to open, using local file system conventions, and a "mode."
    (For  maximum portability, you should avoid using volume names,
    directories, folders, or other path information in filenames.)
    The mode argument is one of these single-quoted string values:

        r    open file for reading; file must already exist
        r+   open file for reading and writing; the file is
             created if it doesn't already exist
        w    create a new file for writing; the file is deleted
             if it already exists
        w+   create a new file for reading and writing; the file
             is deleted if it already exists

    The return value of fopen() is a "file handle"; this is simply
    a number that you you to perform subsequent operations on the
    file.   For example, this opens a new file called TEST.OUT for
    writing:

        fnum := fopen('test.out', 'w');

    To close an open file, use fclose():

        fclose(fnum);

    Note that the TADS runtime allows only a limited number of
    files (currently 10) to be open simultaneously, so you should
    close a file when you're done with it.

    To write to a file, use fwrite().  This function takes a file
    handle, and a value to write; the value can be a string, a number,
    or true.  The value can't be nil (this is because the fread() function
    returns nil to indicate failure; if you could write nil to a file,
    there would be no way to distinguish reading a valid nil from an
    error condition).  fwrite() stores the value, along with information
    on its type.

    The fwrite() function returns nil on success, true on failure.  If
    the function returns true, it usually means that the disk is full.

       if (fwrite(fnum, 'string value!')
           or fwrite(fnum, 123))
           "Error writing file!";

    If the file is open for reading, you can read from the file with
    the fread() function.  This function takes a file handle, and it
    returns a value it reads from the file.  The value returned is
    of the same type as the value originally written at this position
    in the file with fwrite().  If this function returns nil, it
    indicates that an error occurred; this usually means that no more
    information is in the file (you've read past the end of the file).

       res := fread(fnum);
       say(res);

    You can get the current byte position in the file with the ftell()
    function:

       "The current seek position is << ftell(fnum) >>. ";

    The ftell() function returns a number giving the byte position
    that will be read or written by the next file operation.

    You can set the file position with fseek() and fseekeof().  The
    fseek() function moves the file position to a particular byte
    position, relative to the beginning of the file.  For example,
    this seeks to the very beginning of a file:

       fseek(fnum, 0);

    The fseekeof() function positions the file at its end:

       fseekeof(fnum);

    Note that you must be careful with fseek().  You should only seek
    to positions that you obtained with the ftell() function; other
    positions may be in the middle of a string or a number in the
    file, so seeking to an arbitrary location and writing could
    render the file unusable by partially overwriting existing data.

  - You can now "capture" displayed text to a string.  This new feature
    allows you to examine and manipulate text displays, no matter how they
    are generated.  To activate capturing, use the new built-in function
    outcapture():

        stat := outcapture(true);

    This starts capturing text.  The return value is a status code that
    you use in the subsequent call to end capturing; you don't need to do
    anything with this status code except pass it to outcapture() when
    you're finished capturing output.

    While capturing is in effect, everything that your game attempts to
    display -- through double-quoted strings or through the say() built-in
    function -- is hidden from the user and instead stored in a string.
    Once outcapture(true) is called, no text will be displayed to the
    user until you call this function:

        str := outcapture(stat);

    The second outcapture() call turns off capturing, and returns a string
    that contains all of the text that was generated since the corresponding
    outcapture(true).

    Note that the system automatically turns off capturing any time it
    prompts the user for information.  The system turns off capturing
    when it starts a new command, or when it needs to prompt the user
    for information during a command, such as for disambiguation or to
    request a missing direct or indirect object.  When the system turns
    off output capturing, it clears the capture buffer, so any subsequent
    call to outcapture(stat) will return an empty string.  You can capture
    text only over the course of a single command line.

    The outcapture() function can be useful if you want to save some
    text for later.  It's also useful if you want to be able to examine
    the text that would be generated by a method such as sdesc or ldesc.
    Since these methods directly display text, you can't directly obtain
    a string representation of their values; using outcapture(), however,
    you can let them "display" their values into a string that you can
    then examine.

    Note that the status value returned from outcapture(true) allows you
    to nest calls to outcapture().  Since the outcapture(stat) call
    restores the capturing status that was in effect on the corresponding
    call to outcapture(true), you don't have to worry when using outcapture()
    about whether any of the methods you're calling are also using it.

  - TADS now allows you to create and delete objects dynamically
    at run-time.  This is done through two new operators:  "new"
    and "delete".

    To create a new object, use this syntax:

        x := new bookItem;

    This dynamically creates a new object whose superclass is bookItem.
    When this statement is executed, the runtime creates a new object,
    assigns its superclass to be bookItem, and executes the "construct"
    method in the new object; this method can perform any creation-time
    setup that's desired.  The default thing.construct in adv.t simply
    moves the new object into its location -- this is necessary so that
    the "contents" list of the location is updated to include the new
    object.

    A new object inherits all of the vocabulary of its superclass.

    To destroy an object you have created, use this syntax:

        delete x;

    This first calls the "destruct" method of the object to notify it
    that it is about to be deleted, then destroys the object.  Further
    references to the object are illegal, since its memory has been
    released (and thus may be given to another object).  The default
    thing.destruct in adv.t moves the object into nil, which removes it
    from its container's "contents" list -- this is necessary so that
    the reference to the object in that list is removed.

    Only objects created with "new" can be destroyed with "delete".
    Objects that are defined statically in your game's source file
    cannot be deleted at run-time.

    Object creation and deletion works correctly with the UNDO
    mechanism.  If the player uses UNDO after a move that created an
    object, the object will be destroyed; likewise, if a player uses
    UNDO after a turn that deletes an object, the object will be
    re-created with the same property values it had prior to deletion.
    Similarly, dynamically-created objects are preserved across SAVE
    and RESTORE operations.

    Note that TADS does not perform any garbage collection on
    dynamically-created objects.  The system is not capable of
    determining whether an object is accessible or not.  Hence, if
    you lose track of any objects you create with "new", they will
    remain in memory forever -- they will even be saved along with
    saved games and restored when the games are restored.  You must
    be careful to keep track of all objects you create to avoid
    filling all available memory (and the swap file) with unreachable
    objects.

  - It is now possible to dynamically add to and delete from the
    vocabulary words of an object.  You can also get the vocabulary
    words of an object at run-time.

    To add to an object's vocabulary, use the new "addword" built-in
    function.  This function takes three arguments:  an object, a
    vocabulary property pointer, and a word to add.  For example,
    to add 'red' as an adjective to the object myBook, you would
    do this:

        addword(myBook, &adjective, 'red');

    To delete the same word, you would write a similar call to the
    new built-in function "delword":

        delword(myBook, &adjective, 'red');

    You can add to and delete from the words of any object, including
    both static objects (explicitly defined in your source code) and
    dynamically-created objects (created with the "new" operator).

    Changes made by addword and delword are tracked correctly by the
    UNDO mechanism, and are saved and restored along with saved games.

    To get the words belonging to an object, use the new "getwords"
    built-in function.  This function takes two arguments:  an object,
    and a property pointer; it returns a list of (single-quoted)
    strings, which are the vocabulary words for the object.  For
    example, assume we define myBook as follows:

        myBook:  item
            sdesc = "small red book"
            adjective = 'small' 'red' 'tiny'
            noun = 'book'
            location = room2
        ;

    Also assume we haven't made any calls to addword() or delword() for
    myBook.  In this case,

        getwords(myBook, &adjective)

    would return this list:

        ['small' 'red' 'tiny']

    Note that the order of the words in the list is not predictable,
    so you shouldn't expect the words to be in the same order as they
    were when you defined them in the source file, or in the same
    order as they were added with addword().

  - We've added a new function that lets you get information on a verb.
    The new function is verbinfo().  This function lets you get the
    verification and action properties for a verb.  The new function
    takes one or two arguments:  the first is the deepverb object whose
    information you want to retrieve; the optional second argument is
    a preposition object.  If you call verbinfo() with only the verb
    argument, it returns the verification and action properties that
    are defined with the doAction definition for the verb.  If you
    also include the preposition argument, it returns the properties
    that are defined with the ioAction definition for that preposition.

    The value returned by this function is a list.  If you call verbinfo()
    with only the deepverb argument, the list has two elements:

      [1]   direct object verification property pointer (verDoXxxx)
      [2]   direct object action property pointer (doXxxx)

    If you call verbinfo() with both the verb and preposition arguments,
    the return value is a list with four elements:

      [1]   direct object verification property pointer (verDoXxxx)
      [2]   indirect object verification property pointer (verIoXxxx)
      [3]   indirect object action property pointer (ioXxxx)
      [4]   true if ioAction has [disambigDobjFirst] flag, nil otherwise

    In either case, if no matching doAction or ioAction definition
    exists for the verb, this function returns nil.

    Note that it is possible that additional flags (similar to
    disambigDobjFirst) may be added in the future; the returned list
    may be expanded to include information on any such added flags.
    So, for compatibility with future versions, we recommend that you
    don't write conditional code based on the length of the list.
    The lists will never shrink, but they may expand.

    For the removeVerb object defined in adv.t, you would get these
    results:

      verbinfo(removeVerb)
        = [&verDoUnwear &doUnwear]

      verbinfo(removeVerb, fromPrep)
        = [&verDoRemoveFrom &verIoRemoveFrom &ioRemoveFrom nil]

  - The ability to create new objects at run-time leads to some
    interesting problems involving indistinguishable objects.  Although
    you should generally use addword (see below) to make your newly-created
    objects distinguishable from one another, this will not always be
    desirable; for example, if you create new gold pieces that serve
    as currency, you will probably not want them to be uniquely named.

    To support indistinguishable objects, especially those created
    dynamically at run-time, the system now has a property that you
    can set to indicate to the parser that an object does not need to
    be distinguished from others of the same class.  The new property
    is "isEquivalent".  When isEquivalent returns true for an object,
    all other objects with the same immediate superclass are considered
    interchangeable by the parser.  When a player uses one of these
    objects in a command, the parser will simply pick one arbitrarily
    and use it, without asking the player which one.

    If a player uses a noun that is ambiguous with multiple equivalent
    items and one or more other items, the parser will need to
    disambiguate the objects as usual.  In such cases, the parser's
    question will list the distinguishable items only once.  For
    example, assume we have five gold coins that are all equivalent
    (in other words, they all have isEquivalent set to true, and they
    all are immediate subclasses of the same class).  Assume further
    that a silver coin and a bronze coin are also present in the room.

       Treasure Room
           You see a bronze coin, five gold coins, and a silver
       coin here.

       >get coin
       Which coin do you mean, the bronze coin, a gold coin, or
       the silver coin?

    Note that the objects which appear only once are listed with "the"
    (using the thedesc property), while the indistinguishable objects
    are listed only once, with "a" (using the adesc property).

  - The new property pluraldesc has been added to thing in adv.t.
    The definition in thing simply adds an "s" to the end of the
    sdesc property.  This new property is used by listcont(obj)
    when multiple equivalent objects are present in a list; see the
    information on the changes to listcont(obj) for details.

  - The adv.t functions listcont(obj) and itemcnt(list) have been
    changed to support indistinguishable objects.  To support this
    new functionality, the new functions isIndistinguishable(obj1, obj2)
    and sayPrefixCount(cnt), and the new property pluraldesc, have
    been added to adv.t.

    isIndistinguishable(obj1, obj2) returns true if the two objects
    obj1 and obj2 are equivalent for the purposes of listing.  The
    two objects are considered equivalent if both have the same
    first superclass (the return value of the new built-in function
    firstsc(obj)), either they are both being worn or neither is
    worn, and either both are lit lightsources or neither is.  This
    function doesn't test the isEquivalent property of either object,
    since it's assumed that it will only be called if an object has
    already been found whose isEquivalent property is set.

    sayPrefixCount(cnt) displays a number.  If the parameter cnt is
    a small number (from one to twenty), the spelled-out number will
    be displayed (for example, "five" will displayed if cnt = 5).  If
    the count is larger, the number will be displayed as digits ("35"
    will be displayed if cnt = 35).  This function is used by
    listcont(obj) to show the number of equivalent items when more
    than one equivalent item is being listed.

    itemcnt(list) now returns the number of distinguishable items in
    the list that are to be listed.  For each item in the list whose
    isEquivalent property is true, itemcnt(list) checks each other
    item in the list, and counts each set of equivalent items only once.
    Hence, if a list consists entirely of equivalent items, itemcnt(list)
    will return at most 1 (it will return 0 if none of the items are
    listable).

    listcont(obj) will list each set of indistinguishable items in the
    contents list only once, and will show the number of each such item.
    To display the number, the new function sayPrefixCount(cnt) is
    used.  For example, if a room contains a silver coin, a bronze coin,
    and five gold coins (all with the isListed property set to true),
    listcont(room) will display this:

       a silver coin, a bronze coin, and five gold coins

  - A new built-in function has been added to make it possible to
    implement the functionality of the new listcont method.  The new
    function, firstsc(obj), returns the first immediate superclass of
    the given object (or nil if the object has no superclass).  This
    function, along with the isEquivalent property, can be used to
    determine if two objects should be considered indistinguishable.

  - A new special word has been added:  ANY, which is equivalent to
    EITHER.  These words are at the end of the original specialWords
    list; for compatibility with past versions, a specialWords list
    that omits this position is still legal, and indicates that the
    default ('any' = 'either') should be used for this slot.

    This special word slot is used by the parser during disambiguation.
    Whenever the parser asks the player to choose an object from a list
    of ambiguous objects, it will accept ANY:

        >take coin
        Which coin do you mean, the silver coin, the bronze coin, or
        the gold coin?

        >any
        silver coin:  Taken.

    When ANY is used in these cases, the parser will simply pick one
    of the objects arbitrarily.  Note that it displays the chosen
    object in the same manner as it would if multiple objects were
    being used.

    In addition, the parser will accept noun phrases that start with
    ANY to indicate that any object matching the given noun phrase is
    acceptable; the parser will choose one of the objects arbitrarily
    in these cases.  For example:

        >take any coin
        silver coin:  Taken.

        >look at any of the coins
        bronze coin:  It's a valuable 1964 Tadsmid, worth over
        .0004 cents on today's scrap bronze market.

    The player can also specify the number of items to take.  When
    a number of items is specified, it must be applied to a plural
    noun phrase, and it means the same thing as "any," except that
    the parser (arbitrarily) chooses the given number of items rather
    than just one.  For example:

    >look at 3 coins
    >look at any 3 coins
    >look at 3 of the coins
    >look at any 3 of the coins

    As a special case, a count of "1" can be used with a singular
    noun phrase.  It means the same thing as "any."

        >look at 1 coin

  - A new convenience feature, similar to doSynonym and ioSynonym but
    somewhat easier to use, has been added.  You can now specify that
    a method in one object should instead be sent to another object.
    An example:

        desk: fixeditem
            noun = 'desk'
            sdesc = "desk"
            location = office
            doOpen -> deskDrawer
            doClose -> deskDrawer
        ;

    This specifies that doOpen, verDoOpen, doClose, and verDoClose
    calls should be sent to the deskDrawer object when received by
    the desk.  Note that this should only be used for standard verb
    handler methods, because it redirects both the method indicated
    and its verXoVerb equivalent.

  - The parser calls a new method, multisdesc, when displaying the
    name of an object that's part of a list of objects.  Previously,
    the parser simply used sdesc in these cases.  The default adv.t
    definition of thing.multisdesc simply calls the object's sdesc.
    This new method is intended to allow you greater control over
    the display in situations like this:

      >get all
      book:  Taken.
      rug:  That's much too heavy to carry.

    The object names listed before the colons are now displayed
    with multisdesc.

    For compatibility with old games, if an object being listed does
    not define or inherit a multisdesc property, its sdesc is used
    instead.  This ensures that games compiled with previous versions
    of adv.t will continue working properly.

  - A new user-defined function that the parser calls has been added.
    This new function is called preparseCmd(), and is similar to
    preparse().  Whereas preparse() is called once for an entire
    command line, and is called with the original, unfiltered text
    of the player's command line, preparseCmd() is called separately
    for each command on a command line if more then one command is
    entered.  Furthermore, preparseCmd() is called after the command
    has been "tokenized" (broken into individual words).  Whereas the
    argument to preparse() is a string with the entire command line,
    the argument to preparseCmd() is a list, each entry of which is
    a (single-quoted) string giving an individual word.  Using the new
    function, you can exert much greater control over how a command
    is parsed, including rewriting a command entirely.

    preparseCmd() is called immediately before Me.roomCheck() is called
    for the command.  This call is made prior to any disambiguation or
    object defaulting.

    If preparseCmd() returns nil, the command is abandoned (but no
    error message is displayed), and no fuses or daemons are run.  If
    the function returns true, the command proceeds as normal.  Any
    remaining commands on the same line are executed regardless of the
    return value from preparseCmd().  (If you need to cancel the entire
    rest of the command line, one approach would be to set a property
    in the global object to indicate to preparseCmd() that all commands
    are to be ignored; preparseCmd() would always check this property
    before doing anything else, and return nil if it were set.  You
    could clear this property in preparse() so that commands always
    start off enabled.)

    In addition, preparseCmd() can return a list of (single-quoted)
    strings, in which case the parser starts over parsing the list
    instead of the original command.  The list is limited to a maximum
    of 128 characters, with one additional character of overhead per
    word -- in other words, you can't make the new command longer
    than the original command, which is limited to the same maximum
    when entered by the player in the first place.  However, you can
    otherwise rewrite the command entirely.  If you want to include
    any special words in the new command, use the conventions described
    below (for example, if you want to include 'and' in the new command,
    use ',' instead).  After the new command inserted by preparseCmd()
    has been processed, the parser will resume processing any remaining
    commands on the player's original command line.

    The new command list returned by preparseCmd() can contain multiple
    commands.  Simply separate the commands with commas (',') in your
    list.

    If preparseCmd() returns a list, preparseCmd() will be invoked on
    the new command.  However, preparseCmd() is not allowed to return
    another new command in these cases -- if it does, the parser will
    assume that preparseCmd() is looping, and will generate an error.

    Several new parseError codes have been added to respond to error
    conditions that can arise from preparseCmd():

        32  Internal game error: preparseCmd returned an invalid list
        33  Internal game error: preparseCmd command too long
        34  Internal gmae error: preparseCmd loop

    The sample implementation of preparseCmd() below simply lists all
    of the words in the current command and displays a newline, then
    allows the command to proceed as usual.

        #pragma C+
        preparseCmd: function(cmd)
    {
        local i, tot;

        for (i = 1, tot = length(cmd) ; i <= tot ; ++i)
            "<<cmd[i]>> ";
        "\n";

            return true;
    }

    Note that the parser performs conversions of the special words.
    These conversions will show up in the list as follows:

        "and"   becomes  ","
        "all"   becomes  "A"
        "but"   becomes  "X"
        "it"    becomes  "I"
        "them"  becomes  "T"
        "him"   becomes  "M"
        "her"   becomes  "R"
        "any"   becomes  "Y"

    Here are some examples using the preparseCmd() function above.
    (The actual response of the commands has been removed -- only the
    text displayed by preparseCmd() is shown.)

        >look at all
        look at A

        >examine him; take everything except the box and the book and go north
        examine M
        take A X the box , the book
        go north

    preparseCmd() is called even for commands that the parser doesn't
    understand.  This allows you to rewrite commands that TADS wouldn't
    normally understand and put them into a format that's acceptable to
    the parser.  For example, the preparseCmd() example below will
    take sentences of the form "tell <actor> to <command>", and convert
    them to the normal TADS syntax, "<actor>, <command>".

        #pragma C+
        preparseCmd: function(cmd)
    {
        local i, tot, to_loc, actor, the_rest;

            tot = length(cmd);

        /* check to see if it starts with "tell" */
        if (tot > 3 && cmd[1] == 'tell')
        {
            /* see if there's a word "to" */
        for (i = 1, tot = length(cmd) ; i <= tot ; ++i)
        {
            if (cmd[i] == 'to')
                to_loc = i;
        }

        /* if there's a "to", convert the command */
        if (to_loc != nil && to_loc > 2)
        {
            /* find the parts before and after the 'to' */
            for (i = 2, actor = [] ; i < to_loc ; ++i)
                actor += cmd[i];
            for (the_rest = [], i = to_loc + 1 ; i <= tot ; ++i)
                the_rest += cmd[i];

            /* convert it to "actor, command" */
            return actor + ',' + the_rest;
        }
        }

            /* otherwise, process the command as normal */
            return true;
        }

    When the parser doesn't know how to handle a sentence, it calls
    preparseCmd with the entire rest of the command (which will include
    everything before the next THEN or period), and processing is the
    same as in any other case.  If preparseCmd() returns nil in this
    case, the command is cancelled as usual without any further
    processing by the parser.  If preparseCmd() returns true when the
    sentence is not recognized by TADS, the parser will display the
    usual message (parseError code 18, "I don't understand that
    sentence").  If preparseCmd() returns a list of strings, the
    command is replaced with the list, and the parser starts over
    processing the new command.

  - Several new parseError codes have been added.

    In order to produce better messages when actors are involved,
    the system now attempts to figure out whether an object in a
    command refers to an actor, and if so, to determine whether the
    actor should be called "him" or "her".  Previously, this
    one-size-fits-all message was generated:

       What do you want to <verb> it <prep>?

    for example:

       >hit bill
       What do you want to hit it with?

    Although it's not always possible to determine which object
    should be used in these cases (because disambiguation will not
    be possible until the indirect object is known), the system will
    make its best guess.  To do so, it looks at all of the objects
    that might be involved, based on the vocabulary.  If the player
    appears to mean multiple objects, " them " will be used -- this
    is message number 144 for parseError:

       >throw ball and bat
       What do you want to throw them at?        <- message 144
                                ------

    If only one object appears to be intended, the parser will try
    to figure out whether the object is male, female, or neuter,
    using the isHim and isHer properties (these properties are not
    new; they have been around since before TADS 2.0).  If all of
    the objects that match a single noun phrase have isHim and
    not isHer, " him " will be used -- this is parseError message
    number 145:

      >hit bill
      What do you want to hit him with?          <- message 145
                             -----

    If they all have isHer and not isHim, " her " will be used
    (message number 146):

      >hit jill
      What do you want to hit her with?          <- message 146
                             -----

    If both are set, the system equivocates with " them ", message
    number 147 (note that this is the same default text as message
    144, but it's distinguished as a separate message number in
    case a game author wants a more suitable word in this case,
    especially in a non-English language):

      >hit hermaphrodite
      What do you want to hit them with?         <- message 147
                             ------

    If not all of the objects involved have isHim and/or isHer
    set, the system uses "it" (message 141) as in previous versions.

    Another change to the processing for this message involves
    actors.  If an actor is specified in the command, the system
    now builds a replacement for message 140.  First, message
    148 is displayed, which has the default text "What do you want ".
    Then, the actor's thedesc is invoked to display the actor's
    name.  Finally, message 149 (default text " to ") is displayed,
    and the rest of the message is built as before.

      >guard, throw ball
      What do you want the guard to throw it at?
      -----------------         ----
         message 148             149

  - The functionality of the parseAskobj function has been extended
    so that you can generate the same sort of message that the parser
    now generates when a missing object is needed for a command directed
    to an actor.  To provide the new functionality, the parser now calls
    a function called parseAskobjActor.  This function is exactly the
    same as parseAskobj, except that it takes the actor as the first
    parameter, the verb as the second parameter, and the preposition
    as an optional third parameter (which is only present when asking
    for an indirect object).

    If your game defines parseAskobjActor, the system ignores parseAskobj
    and calls the new function instead.  If parseAskobjActor is not defined,
    but parseAskobj is defined, the system calls parseAskobj as it did in
    past versions.  If neither function is defined, the system generates
    the message itself as described above.  You should use parseAskobjActor
    rather than parseAskobj for new games, since it gives you more
    information.

    Here's an example of parseAskobjActor, which generates roughly the
    same message as the system would (but it doesn't do any of the checking
    of isHim and isHer to determine which pronoun to use -- this function
    simply uses "it").

    parseAskobjActor: function(a, v, ...)
    {
        if (argcount == 3)
        {
            "What do you want ";
                if (a <> Me) a.thedesc;
        " to <<v.sdesc>> it <<getarg(3).sdesc>>?";
        }
        else
        {
        "What do you want ";
            if (a <> Me) a.thedesc;
        " to <<v.sdesc>>?";
        }
    }

  - The parser has been changed slightly to allow a command to specify
    an actor in any command within a single line containing multiple
    commands.  Previously, if an actor was to be specified, the actor
    had to be specified at the very beginning of the command.  This
    restriction has been removed, which allows commands like this:

      >joe, north.  bob, south.  bill, east, take book, west.

    Note that, as in past versions, once an actor is specified, the
    actor remains in effect for subsequent commands.  Since you could
    only specify one actor for an entire command line in past versions,
    this meant that the actor was used for every command on the line;
    with this new version, an actor remains in effect until another
    actor is specified.  So, in the command above, "north" is directed
    to Joe, "south" is directed to Bob, and "east, take book, west"
    is directed to Bill.

  - The maximum number of notifiers has been increased to 200.  The
    maximum number of daemons and fuses has been increased to 100 each.

  - The maximum number of ambiguous words matching a particular
    vocabulary word has been increased to 200.  This should relieve
    problems that some people have reported with the error message
    "The word 'foo' refers to too many objects".

  - The compiler now supports most of the remaining C operators:

        a % b      - returns the remainder of dividing a by b
        a %= b     - sets a to a % b
        a != b     - equivalent to a <> b
        !a         - equivalent to (not a)
        a & b      - bitwise AND
        a &= b     - sets a to the bitwise AND of a and b
        a | b      - bitwise OR
        a |= b     - sets a to the bitwise OR of a and b
        a && b     - equivalent to (a and b)
        a || b     - equivalent to (a or b)
        a ^ b      - bitwise XOR of a and b
        a ^= b     - sets a to the bitwise XOR of a and b
        ~a         - bitwise negation of a
        a << b     - a shifted left by b bits
        a <<= b    - shifts a left by b bits
        a >> b     - a shifted right by b bits
        a >>= b    - shifts a right by b bits

    Note a slight complication involving the >> operator:  you can't
    use this operator from within an embedded string expression, because
    it would be confused by the parser for the end of the expression.
    It doesn't help to use parentheses, since the embedded string
    processing is essentially a textual substitution mechanism which
    happens without knowledge of the expression context (and is thus
    unaware of parenthesization).  For example, this would be illegal:

       myprop = "x divided by 128 is << (x >> 7) >>!"     // wrong

    You would have to code this instead as:

       myprop = { "x divided by 128 is "; x >> 7; "!"; }  // right

    Another slight complication arises from the use of the & operator
    in lists.  Since TADS allows list elements to appear without any
    separating punctuation (except whitespace), you can have a list
    that looks like this:

       mylist = [&sdesc &adesc &thedesc]

    This construct is still legal, and is still interpreted with the
    "&" operators as unary operators, not bitwise AND operators.
    However, the parser now warns when such a definition is used; see
    the description of the new warning TADS-357 below for details.

  - The compiler will generate a new warning when it detects a unary
    operator within a list that could also be interpreted as a binary
    operator.  For example, in this list,

       list2 = [5 -2 -6 -7]

    the "-" operators could be interpreted either as unary negation
    operators, which would result in a list with four elements (the
    numbers 5, -2, -6, and -7), or as binary subtraction operators,
    which would result in a list with only one element (the number
    -10).  This same problem arises with the operators "+" and "&",
    since these also have a unary and binary interpretation which
    depends on context.

    In these cases, the compiler interprets the operators as unary
    operators, and issues a warning to let you know that the usage
    was ambiguous.  The warning is TADS-357:

      TADS-357: warning: operator '-' intepreted as unary in list

    Note that this is a change from the previous version for the '+'
    and '-' operators.  If your game depends on building lists from
    calculated numeric constants, you will need to change your code;
    we don't expect that any games actually depend on the old behavior.

    If you really do want the operators in these cases interpreted
    as binary operators, use parentheses:

       list2 = [(5 -2 -6 -7)]

    The parentheses tell the compiler that the expression is to be
    interpreted as a single list element.

    If you want the unary interpretation, and you want to suppress
    the warning, use commas between the list elements:

       list2 = [5, -2, -6, -7]

    This doesn't change the interpretation, but it does suppress the
    warning, because it removes the ambiguity:  when the commas are
    present, there is no way the '-' operators could be interpreted
    as binary operators.

    You can suppress the TADS-357 warning if you wish using a new -v
    suboption.  Specify -v-abin in your compiler options to turn off
    the warning ("abin" is short for "ambiguous binary operator", which
    is the meaning of the warning).  You may want to suppress this
    warning when compiling code written prior to this version of TADS,
    since the warning is generated in case you wanted the new meaning
    of the & operator.

  - A new compiler command-line option has been added:  -C, a toggle
    option, which turns on and off C-language operator compilation.
    By default, C operator mode is off (-C-), which makes the compiler
    use the normal TADS operators.  Specifying -C+ turns on C operator
    mode; specifying -C- disables C operator mode.

    When C-language operator mode is in effect, two operators are
    affected:  the assignment operator becomes '=', and the equality
    comparison operator becomes '=='.  With normal TADS operator mode
    in effect, assignment is ':=' and equality is '='.  If you are
    a C programmer, and you're unhappy with the slight variation in
    operator notation between TADS and C, you can use -C+ to make TADS
    behave more like a real language.  Thanks to the #pragma C options
    (see below), C operator mode has no effect on your choice of
    header files -- you can use the same old adv.t and other header
    files unchanged, and still use C-style operators in your code.

  - A new preprocessor command has been added:  #pragma.  This special
    directive can be used to specify certain compiler options from
    within your source code.  Currently, the only #pragma option available
    is the C operator mode option.  Use #pragma C+ to turn on C operator
    mode, and #pragma C- to turn it off.

    The #pragma C+ and #pragma C- settings are local to a particular
    file.  If a file is included by another file, the #pragma C settings
    specified in the included file will be in effect only until the end
    of the included file; the including file's #pragma C settings that
    were in effect before including the other file will be restored at
    the end of the included file.  adv.t and std.t now start with a
    #pragma C- command -- this allows adv.t and std.t to be included
    from a file with C-style operators (and thus a #pragma C+ or
    command-line C operator mode setting).  Since the enclosing file's
    #pragma C option will be restored after the inclusion, files with
    different operator modes can be freely intermixed with #include, as
    long as each included file specifies its desired mode with a
    #pragma C directive.

  - The precedence of the comparison operators has been changed to
    be the same as that used by C.  Previously, all of the comparison
    operators were at the same precedence; starting with this version,
    == and <> (and thus !=) are at the same level of precedence, and
    associate left to right as before, but <, >, <=, and >= are one
    level higher in precedence.  The following type of expression will
    be affected:

          a > 1 <> b > 1

    Previously, this grouped as:

          (((a > 1) <> b) > 1   // obsolete

    This now groups as:

          (a > 1) <> (b > 1)    // current behavior

    This shouldn't affect any existing code, since the old interpretation
    should always have resulted in an error (because a truth value, true
    or nil, can not be compared in magnitude to a number or other type).

  - Limited conditional compilation and preprocessor text substitution
    (#define) support has been added to the compiler.  The following
    preprocessor directives are now available:

      #define symbol value
      #undef symbol
      #ifdef symbol
      #ifndef symbol
      #else
      #endif

    #define is used to assign a value to a preprocessor symbol.  The
    "value" is simply text that will be substituted verbatim for the
    symbol whenever it occurs in your file (other than within quoted
    strings).  For example:

      #define TEST say('hello from TEST!')

      myfunc: function
      {
         TEST;
      }

    The symbol TEST is replaced with its definition, so the function
    myfunc() displays "hello from TEST!" when called.

    Preprocessor symbols defined with #define are in a separate
    namespace from all other symbols in your program.  Unlike a
    standard C preprocessor, no arguments are allowed in #define
    macros.

    #undef deletes a previously #define'd symbol.  You can #undef
    the special symbols defined automatically by the compiler if
    you wish (see below).

    #ifdef tests to see if a preprocessor symbol is defined.  If it
    is, the lines following the #ifdef line, and up to the corresponding
    #else or #endif, are included; otherwise, they are ignored.

    #ifndef is the opposite of #ifdef:  #ifndef tests to see if the
    symbol is NOT defined.  If the symbol is undefined, the lines
    following the #ifndef line up to the corresponding #else or
    #endif are included; otherwise, they are ignored.

    #else indicates that the lines between the #else and #endif are
    to be included if an only if the corresponding #ifdef (or #ifndef)
    failed.  #else is optional.  At most one #else is allowed per
    conditional.

    #endif terminates a conditional block.  Exactly one #endif must appear
    for each conditional (#ifdef or #ifndef).

    You can use #ifdef to compile certain parts of your code
    conditionally.  For example, if you want to include a verb only
    for your debugging version of a game, but you want to remove it
    from the final version, you could do something like this:

      #ifdef __DEBUG
      magicVerb:  deepverb
         verb = 'xyzzy'
         action(actor) =
         {
         }
      ;
      #endif

    Note that __DEBUG is especially handy for this sort of thing,
    because the compiler automatically defines this symbol when
    debugging (-ds) is turned no (see below).

  - The compiler automatically defines several preprocessor symbols.
    These symbols can be used or tested within your code as needed.

    __TADS_VERSION_MAJOR is defined to a number indicating the major
    version number of the compiler (in the present system, 2).

    __TADS_VERSION_MINOR is defined to a number indicating the minor
    version number (in the present system, 2).

    __TADS_SYSTEM is defined to a string (single-quoted) identifying
    the operating system the compiler is running on.  In addition, the
    same identifier contained in the string is defined as a preprocessor
    symbol itself (its value is always 1; it is intended that it will
    be tested with #ifdef, and not otherwise used).  For example, if
    __TADS_SYSTEM is 'MSDOS', the symbol MSDOS will be defined to 1.
    If __TADS_SYSTEM is 'Macintosh', the symbol Macintosh will be
    defined to 1.

    __DEBUG is defined to 1 if debugging is turned on for this
    compilation (with the -ds compiler option).  Otherwise, this symbol
    is not automatically defined.  You can test the existence of __DEBUG
    with #ifdef to conditionally include code only when you are compiling
    for debugging.  This might be useful if you want to include certain
    commands only in the debugging version of your game, and want to
    remove them when you actually deliver the game to players.

    If this set of symbols were entered manually with #define statements,
    the definitions might look like this:

      #define __TADS_VERSION_MAJOR  2
      #define __TADS_VERSION_MINOR  2
      #define __TADS_SYSTEM         'MSDOS'
      #define MSDOS

    __DATE__ is defined as a single-quoted string giving the system date
    when the compilation began, in the format "Jan 01 1980".

    __TIME__ is defined as a single-quoted string giving the system time
    when the compilation began, in a 24-hour format, "13:40:50".

    __FILE__ is defined as a single-quoted string giving the file being
    scanned at the point where the __FILE__ macro is encountered.  Each
    time you use __FILE__, it will have the correct value for that point
    in your source code.

    __LINE__ is defined as a number giving the line number at the point
    where the __LINE__ macro is encountered.  Each time you use __LINE__,
    it will have the correct value for that point in your source code.

  - The compiler has two new options that give you further control
    over preprocessor symbols.  The -D option allows you to define
    a preprocessor symbol from the command line.  For example, to
    define TEST to 5 from the command line, you could do this:

      tc -i/tads/include -DTEST=5 mygame.t

    Note that if you omit the equals sign, the default definition
    of the symbol will be 1:

      tc -i/tads/include -DTEST mygame.t

    This defines TEST to 1.

    The -U option undefines a predefined symbol.  You can use this to
    undefine one of the symbols automatically defined by the compiler.
    You can also use it to undefine a symbol defined in a precompiled
    header, if you're loading one; -U is applied after the precompiled
    header is loaded, so it will undefine symbols loaded from the file.
    For example, to compile for debugging, but leave __DEBUG undefined,
    you could do this:

      tc -ds -U__DEBUG mygame.t

    The -U option is applied after all -D options, so you can also use
    it to undefine a symbol placed earlier on the command line.  This
    may be useful if you are using a configuation file (CONFIG.TC)
    that contains -D options for symbols you sometimes want to undefine.

  - The compiler has a new preprocessor directive, #error, which allows
    you to generate your own error during compilation.  If a #error
    directive is encountered, any text after the #error is displayed
    as a compiler error; an occurrence of #error is counted as an actual
    compilation error, so compilation will fail if #error is encountered.
    For example:

       #ifndef TEST
       # error TEST is not defined!
       #endif

    If the preprocessor symbol TEST is not defined at the point when
    this sequence is encountered, the compiler will display an error:

      mygame.t(181): error TADS-124: TEST is not defined!

  - The runtime has a new debugging feature that may help you track
    down problems with word definitions.  You can make the player
    command parser generate a number of status messages as it analyzes
    a player's command; these messages provide information on how the
    parser is interpreting the words in the command.

    To activate this new debug mode, use the debugTrace function with
    these arguments:

       debugTrace(1, true);

    To turn the debug mode off, call with nil instead of true.  This
    debugTrace function is always available, even when running under
    the normal runtime; the function returns no value when called with
    these arguments.

  - The built-in function incturn() has been extended to allow you
    to run a series of turns all at once.  You can now specify a numeric
    argument to incturn(); the argument gives the number of turns that
    should pass.  An argument of 1 causes incturn() to behave as usual.

    When an argument higher than 1 is given to incturn(), the function
    runs all of the fuses that are set to burn down within the number
    of turns specified, but not after that number of turns.  Note that
    the normal incturn() has never actually executed any fuses, but
    simply burns down all fuses by one more turn.

    For example, if you call incturn(2), the system will first run
    any fuses that are set to burn down after 1 turn, then will shorten
    all remaining fuses by one more turn.

  - A new built-in function, skipturn(), has been added.  This new
    function takes a numeric argument specifying the number of turns
    to skip; it must be at least 1.  skipturn(n) is similar to incturn(n),
    except that it does not run any of the fuses that burn down during
    the 'n' turns -- instead, it simply removes them without running
    them.

  - A new built-in function allows you to force capitalization off --
    this function, nocaps(), is the opposite of caps().  If you call
    caps() then call nocaps(), the next character is lower-case; if
    you call nocaps() then caps(), the next character is capitalized.

    Along with nocaps(), the special sequence \v has been added.  Using
    this sequence in a displayed string is equivalent to calling
    nocaps(); this sequence is analogous to \^, which is equivalent
    to calling caps().

  - A new built-in function has been added to clear the screen:
    clearscreen(), which takes no arguments and has no return value,
    simply clears the screen.  This routine may do nothing on some
    platforms; for example, in -plain mode on DOS, clearscreen() has
    no effect.

  - The parser is now capable of disambiguating direct objects before
    indirect objects.  By default, everything works as it always has --
    in a two-object command, the indirect object is disambiguated first,
    and then the direct object is disambiguated in the presence of the
    known indirect object.

    It is now possible, however, to specify that the reverse should be
    done.  To do this, you use a new special flags syntax when defining
    your verb:

       tellVerb: deepverb
          verb = 'tell'
          desc = "tell"
          ioAction(aboutPrep) = [disambigDobjFirst] 'TellAbout'
       ;

    The new special flags are placed in square brackets between the
    equals sign and the property template for the verb definition.
    The flags currently accepted are:

       disambigDobjFirst
       disambigIobjFirst

    Note that disambigIobjFirst is provided for completeness only; it
    is never needed, because it is the default setting.

    When the disambigDobjFirst flag is specified, it means that the
    command should have its direct object disambiguated before its
    indirect object.

    When the disambiguation order is reversed, the normal argument
    lists for verDoTellAbout and verIoTellAbout are interchanged.
    For our example, the prototypes for the verifier methods become:

       verDoTellAbout(actor)
       verIoTellAbout(actor, dobj)

    Normally, verIoVerb would not receive the direct object as an
    argument, because the direct object would not be known at the time
    of the verIoVerb call; and verDoVerb would receive the indirect
    object as an argument, because it would be known by the time the
    direct object was being tested.  When the disambiguation order is
    reversed, however, so are the prototypes to these functions.

    The actual action method, ioTellAbout(actor, dobj), remains
    unchanged.  All other methods also remain the same.

    When the direct object is disambiguated first, the player is not
    allowed to use multiple direct objects (or multiple indirect objects)
    in the command.  If the player tries to do so, the new parseError
    message number 28 is displayed:

       >tell bob and bill about gun
       You can't use multiple objects with this command.

  - Note that a .GAM file format change was required to support the
    extra information needed for the disambigDobjFirst flag.  The TADS
    file format is now format "C".  The compiler is still capable of
    producing formats A or B, in case you need to generate .GAM files
    that can be played with versions prior to 2.2; however, if you
    use a format prior to C, you will not be allowed to use the
    disambigDobjFirst flag (or any other similar flags that may be
    added in the future).

  - Two new methods have been added to disambiguate actors.  Previously,
    the system validated and disambiguated an actor by pretending that
    you were attempting to take the actor -- takeVerb.validDo and the
    actor's verDoTake were used to validate and disambiguate the actor,
    when all you wanted to do was speak to him.  This did not always
    produce satisfactory results, and in particular did not allow for
    such situations as talking over a radio to an actor in another room.
    To provide better control over actor validation and disambiguation,
    the system for testing actors has been enhanced.

    First, to validate an actor, the system now uses the method
    validActor in the actor object itself.  This method takes no
    arguments; it returns true if the object is valid as an actor in
    a command, nil otherwise.  This method is called before the verb
    or any of the objects involved in the sentence are known.  Its
    function is not to determine whether the actor wants to receive
    the command, or even if the object can be used as an actor
    (actorAction is the place to do both of these tests), but rather
    simply to determine if the object can be addressed by the player
    at all.  This method should return true if the actor is accessible
    by voice command (or whatever other means you want to provide for
    giving commands) to the player.  The default thing.validActor in
    adv.t returns true if the object is reachable by Me, which provides
    roughly the same behavior as the old takeVerb-based mechanism.

    Second, to disambiguate an actor, the new preferredActor method
    has been added.  This method is called if an actor is ambiguous.
    As with validActor, it doesn't take any arguments, and it returns
    true if the object is "preferred" as an actor, nil otherwise.  If
    exactly one ambiguous object's preferredActor method returns true,
    the parser will use that object as the actor without further
    questioning the player; otherwise, the system will ask the player
    to disambiguate the noun as normal.  In adv.t, movableActor defines
    preferredActor = true.

    An example of how these might be used:

       Sleeping Compartment
          You are in a sleeping compartment on a moving train.
       A pair of bunks is along each wall.
          There is a copper wire here.
          The train conductor is standing in the doorway, asking for
       your ticket.

       >look at conductor
       Which conductor do you mean, the copper wire, or the train
       conductor?

       >copper
       It's a piece of wire, about a foot long.

       >conductor, where is ivan?
       "Your ticket, please," is all the conductor has to say.

    In this sequence, when the command is addressed to "conductor", the
    parser matches both the copper wire and the train conductor.  It
    checks validActor in each of them -- both return true, since both
    are accessible to the player.  (As described above, even though the
    copper wire couldn't possibly be an actor, it is valid as an actor
    at this point -- it's not until its actorAction that we will decide
    that there's no point in talking to it.)

    So, the parser has an ambiguous actor.  The parser tries to
    disambiguate the actor by testing preferredActor in each object.
    The copper wire's preferredActor returns nil; the train conductor's,
    however, returns true, because the conductor is a movableActor
    object.  The parser now has only one object, and thus doesn't
    need to ask the player for further information.

    For compatibility with games compiled with previous versions of
    adv.t, the parser will continue to use the old mechanism (involving
    the takeVerb) if validActor is not defined in your game.  Each time
    an actor is used, the parser checks to see if the first object in
    the list of possible actor objects has a validActor method defined;
    if not, the parser uses the old mechanism.  If you use the new adv.t,
    all objects will at least inherit a validActor method (from thing),
    so testing for the presence of this method in any object lets the
    parser determine if the new mechanism can be used with the game.

  - A new parseError message has been added:  number 27, whose default
    text is "You can't repeat that command."  This message is displayed
    when the player types AGAIN, but the parser can't repeat the command;
    this is the case if one of the objects involved in the command is
    no longer accessible.

  - A new parseError message has been added, related to validActor
    (see above).  This new message, number 31, is used when none of
    the objects matching the vocabulary for an actor in a player's
    command can be used as an actor.  The default text of this message
    is "You can't talk to that."  For example, if there is no object
    matching the vocabulary "guard" whose validActor method returns
    true, but an object named "guard" is visible to the player, the
    message is used:

      Security Room
          You are inside a small cubicle.  A thick, laser-proof
      (as you now know from your ill-fated attempt) glass door
      (closed) is to the north.  Through the door you can see
      a guard standing watch.

      >guard, open door
      You can't talk to that.

    This message is used for actors in place of the standard cantReach
    processing done for direct and indirect objects under these conditions.

  - A new parseError code, number 29, has been added.  The default
    message is "I think you left something out after 'any of'", and
    is used when the player uses "any of" in a sentence, but doesn't
    follow it with anything.

  - Using the same word as both a plural and a noun works much better
    now.  The parser will first attempt to use such a word as a plural;
    if no objects match the plural usage, the parser will try to use
    the word as a noun instead.

  - "of" can now be used as a preposition.  In previous versions of
    the run-time, "of" was exclusively a special word that was embedded
    in noun phrases (such as "pile of paper").  The runtime will still
    allow "of" to be used in noun phrases as before, but it also will
    treat "of" as an ordinary word when, based on the context, it does
    not appear to be part of a noun phrase.

    The parser will still attempt to treat "of" as part of a noun
    phrase whenever it matches an object.  For example, if you have
    an object with nouns matching "pile of paper", then the following
    interpretations will apply:

        accuse bob of murder   ->  dobj = bob, iobj = murder, prep = of
        accuse pile of paper   ->  dobj = pile of paper

    Note that ofPrep, defining "of" as a preposition, has been added
    to adv.t.

  - All of the system verbs that use "abort" have been modified
    slightly in adv.t to make it easier to augment their behavior
    with the 'modify' statement.  All of the processing other than
    the "abort" has been moved out of the doVerb (or action) method
    in each case, and put into a new method.  For example, saveVerb's
    action routine now looks like this:

        action( actor ) =
        {
        self.saveGame(actor);
        abort;
    }

    The new method saveVerb.saveGame(actor) now performs all of the
    processing that the action(actor) method previously performed.

    The benefit of this change is that you can now modify the
    saveGame(actor) method, and inherit the original behavior,
    without having to worry about an "abort" interfering with the
    order of operations.  For example:

        modify restoreVerb
            restoreGame(actor) =
            {
                // restore the game as usual - check for success
                if (inherited.restoreGame(actor))
                {
                    // re-randomize the puzzle
            "The carnival hawker flashes a mischevious
            smile at you.  \"There's no use trying to
            guess the answer,\" he says.  \"I changed
            around the shells while you were busy
            restoring!\"";
            puzzle.answer := rand(100);
                }
            }
        ;


  - The format mask fmtMe has been added.  You can now use %me% in
    messages to refer to the actor.  For basicMe, fmtMe is set to
    the message "me"; for other actors, it is set to the actor's
    thedesc.  In adv.t, thing.ldesc has been changed to use %me%:
    "It looks like an ordinary <<sdesc>> to %me%."  This makes the
    default sentence somewhat more adaptable if you ask another
    actor to describe something:

        >guard, look at the card
        It looks like an ordinary card to the guard.

  - AGAIN, WAIT, and SLEEP are now darkVerb's in adv.t.

  - A new verb, breakVerb, has been added to adv.t.  The vocabulary
    words for this verb are 'break', 'destroy', and 'ruin', and it
    provides a single-object command "break <direct-object>".  The
    thing class has been adjusted so that verDoBreak(actor) validates
    breaking any object, but doBreak(actor) simply displays "You'll have
    to tell me how to do that."  If you want to make a breakable object,
    simply override doBreak(actor) so that it breaks the object (you may
    also want to override verDoBreak(actor) so that it doesn't allow a
    broken object to be broken again).

  - "there" has been added a synonym for "it" in the specialWords list
    in adv.t.  This allows sentences like this:

        >get box.  put ball in there.

  - lightsource now has a doTurnnon method in adv.t.  This method will
    show the room's description if the room becomes lit as a result
    of turning on the lightsource.  Note that lightsource, by default,
    has no verDoTurnon, so you can't turn a generic light source on
    and off.  However, if you add switchItem to the superclass list of
    a lightsource object, you will have a light source that you can
    turn on and off, and which will have this new behavior.  Note that
    you should put lightsource in the superclass list prior to
    switchItem, so that lightsource.doTurnon overrides switchItem.doTurnon:

        flashlight:  lightsource, switchItem
            sdesc = "flashlight"
            noun = 'flashlight' 'light'
            adjective = 'flash'
            location = tunnel
        ;

  - verIoGiveTo and ioGiveTo have been added to movableActor in adv.t.
    verIoGiveTo rejects the command if the actor is the same as the
    indirect object, otherwise the command is accepted.  ioGiveTo
    always rejects the offer.  In addition, ioGiveTo has been added
    to basicMe; the method always accepts anything from another actor,
    because this method will only be called in response to commands
    such as:

        >guard, give me the key

  - A small problem with thing.isVisible in adv.t has been fixed.  If
    the vantage is inside the object, and the object's contents are
    visible, isVisible returns true.  This is is needed in certain
    situations involving nested rooms.

  - moveableActor.travelTo(room) in adv.t has been corrected so
    that it does nothing when room = nil, which is the case when the
    actor can't travel in the desired direction.  Previously, the
    "noexit" message would be displayed, but then the actor would
    be moved into a nil location.

  - moveableActor now has a roomCheck method in adv.t.  The method
    is the same as basicMe's roomCheck method; its omission in
    previous versions was an oversight.

  - The showcontcont function in adv.t has been corrected so that
    the isqsurface flag is checked correctly.

  - A new class, seethruItem, has been added to adv.t.  This class is
    intended for objects such as windows or magnifying glasses which
    the player can look through.  Each object of this class should
    define an appropriate thrudesc method, which displays what the
    player sees when looking through this object.

    Note that a seethruItem is not the same as a transparentItem.
    The class transparentItem is intended for objects, such as glass
    bottles, whose contents are visible, whereas seethruItem is intended
    for objects that are not containers but which the player can look
    through.

  - The compiler now sets an error code on exit.  If you are using
    a MAKE utility or other program-building tool, you can use the
    exit code.  On success, the compiler uses exit code 0; if any
    errors occurred, a non-zero exit code is used.

  - The compiler no longer creates a .GAM file if errors occurred
    during compilation.  If a .GAM file of the same name exists
    prior to a compilation, and errors occur, the original .GAM
    file is unaffected (it is not deleted or overwritten).  Although
    the compiler previously produced a .GAM file even when an error
    occurred, this .GAM file was not generally usable; to avoid
    confusion, the compiler no longer produces a .GAM file at all
    when an error occurs.

  - The compiler performs more checking for invalid -m parameters.
    When -m parameters are entered that exceed the compiler's internal
    limits, it will report an error and abort the compilation.  The
    affected options are -mp, -ml, and -ms.  The runtime similarly
    checks its parameters more carefully now.

  - The DOS runtime no longer displays anything on the status line at
    startup.  It initializes both the left portion (which normally is
    used to display the location) and the right portion (which is
    normally used to display the score and turn count) to empty
    strings.  To ensure that a score is properly displayed at the
    start of the first turn, we added a call to scoreStatus(0, 0) in
    the "init" function defined in std.t.

  - The DOS runtime now has a plain ASCII mode.  In this mode, the
    runtime uses only DOS character input and output functions,
    performs no highlighting or cursor positioning, and does not
    display a status line.  Command editing and screen scrollback
    are also disabled (although normal DOS keyboard editing can be
    used, and any loaded command editor, such as DOSKEY, will be
    active).  This mode is intended for people with a special need
    to use direct DOS character I/O, such as persons using a voice
    synthesizer.  To invoke this mode, specify -plain in the runtime
    command options:

          tr -plain mygame

    Note that .EXE files produced by MAKETRX can be used in this
    manner as well; simply specify -plain with the game program:

          deep -plain

  - The DOS runtime now uses the control-left-arrow and
    control-right-arrow to move the cursor left and right by one
    word in the command editor.  This behavior is consistent with
    most other DOS command editors.

  - $$ABEND can now be used after a question from the parser, such
    as during disambiguation and when OOPS is permitted.

  - The runtime handles hyphenation better.  Multiple hyphens will
    no longer be split across lines -- so if you use two hyphens
    together for a dash, both hyphens will always be grouped on one
    line.  Furthermore, the runtime formatter will never put a dash
    at the very start of a line; if a line must be split at a dash,
    the formatter will put the dash at the end of the line, then
    break the line, and will back up and split at the previous word
    if necessary.

  - The runtime will now properly convert "\" sequences in askfile()
    prompts properly.  If you use a \n, \t, \', \", or \\ sequence
    in a prompt to askfile(), the sequence will be displayed as the
    appropriate character.  (For the DOS runtime, this only matters
    in -plain mode, because the DOS file selector in the normal full-
    screen mode doesn't use the prompt string.)

  - The DOS runtime's file selector dialog incorrectly interpreted
    the Alt keys for disks in the last version -- Alt-B selected the A
    drive, Alt-C selected the B drive, and so forth.  This has been
    corrected.

  - The MAKETRX command now takes an additional argument that lets
    you specify the run-time command options that should be used when
    the game is executed.  MAKETRX still accepts the original command
    formats; using one of the old-style command formats will not bind
    any command options into the resulting .EXE file.

    To specify command options for your game executable, you must
    first create a file.  Use the same format as CONFIG.TR -- simply
    enter your options into the file as you would on the TR command
    line; separate options by newlines or spaces.  For example, to
    specify a minimal cache size and a swap file of SWAP.DAT, you
    could make your CONFIG.TR file look like this:

        -m0 -tf swap.dat

    Once you've created a file with your command options, specify
    that file to MAKETRX by using it as the first parameter on the
    command line, prefixed with an "at" sign (@):

        maketrx @config.tr mygame

    The @config option can be used with any of the original command
    formats for MAKETRX.  Once the config file is bound into your
    executable, its options will be used every time a player runs
    your game's .EXE file.  Note that you may want to avoid specifying
    anything specific to your system, such as drives or directories,
    since that may prevent the game from working properly on someone
    else's system.

  - The new sentence parsing for the form VERB PREP IOBJ DOBJ
    (introduced in 2.1.1) causes a subtle problem:  if you create
    an object with an adjective that is also a preposition, the
    system attempts to interpret sentences with a single-word verb
    that refers to the object as VERB PREP IOBJ.  The result is
    this:

       >push off button
       What do you want to push?

    This is because the parser is interpreting the sentence as
    "push something off button", and needs to know the "something".
    This problem has been corrected.  Now, the parser will check this
    type of case to see if the preposition can also be used as an
    adjective, and if so, checks to see if a noun (possibly preceded
    by one or more adjectives) follows; if this test is met, the word
    will be interpreted as an adjective, as it should be.

  - The parser now correctly distinguishes between cases involving
    a word that is defined as both a preposition and an adjective.
    The parser previously did not accept sentences such as this:

      >enter south wall

    when 'south' was defined as an adjective.  Because 'south' was
    also defined as a preposition (in adv.t), the parser attempted
    to interpret this sentence as though 'enter south' were a verb
    (in the same manner as 'pick up' or 'put down').  The parser
    now checks to make sure that 'enter south' is a valid combination
    verb; if it's not, and 'south' is also defined as another part of
    speech, the parser assumes that 'enter' is the verb, and treats
    'south' as an adjective.

  - The parser previously asked to disambiguate a direct object twice
    if askio(prep) was used.  For example:

       >unlock door
       Which door do you mean, the large door, or the small door?

       >large
       What do you want to unlock it with?

       >key
       Which key do you mean, the silver key, or the gold key?

       >gold
       Which door do you mean, the large door, or the small door?

       >large
       The door unlocks with a satisfying click.

    This has been corrected -- the parser now only asks once about
    the door.

  - The parser did not previously accept a word during disambiguation
    if the word was defined as both a noun and an adjective.  For
    example, if you've defined objects "violet paper", "violet banana",
    and "paper towel", and the parser asked you this:

       >x violet
       Which violet do you mean, the violet paper, or the violet banana?

    then you couldn't respond with "paper".  This has been corrected.

  - The inputkey() function now clears the "more" line counter.  So,
    when inputkey() is called, a "more" prompt will not show up until
    another screenful of text has been displayed.

  - "local" statements that occur out of context (i.e., as other than
    the first statements after an open brace) now generate better
    diagnostics.

  - Dividing by zero is now flagged as a run-time error.

  - A problem involving the embedded string << >> notation has been
    corrected.  Under certain circumstances, if one expression used
    in an embedded string invoked another string with an embedded
    expression, an error occurred (usually "invalid type for built-in
    function").  This should no longer occur.

  - The system did not properly handle lists containing function
    pointers.  This has been corrected.

  - A problem involving pre-compiled headers and 'modify' has been
    corrected.  Several people have encountered problems that generally
    were manifested as "assert" failures in mcm.c (the cache manager)
    when using 'modify' and pre-compiled headers; these should no
    longer occur.

  - An internal cache corruption occurred under certain obscure
    circumstances involving 'modify'.  This has been corrected.

  - A note on roomCheck:  if you return nil from roomCheck, fuses
    and daemons are NOT run; in this sense, returning nil from roomCheck
    is equivalent to using abort in other methods.  (This is not a
    change -- it's always worked this way -- but this behavior is
    not mentioned in the TADS Author's Manual.)

  - A note on using << >> embedded strings:  If you put a newline
    immediately before the << of an embedded string, any spaces between
    the last non-space character preceding the << and the << will be
    lost.  If you want a space before the <<, do not put it on a new
    line.  Instead, you can put it at the end of the line, and put
    the embedded expression itself on the next line.  For example:

        embeddedString = "embedded string"
        sdesc = "This is a message with an
                 <<self.embeddedString>>!"     // wrong

    This will display as follows:

        This is a message with anembedded string!

    If you intend a space to precede the embedded string, you should
    write this as follows:

        sdesc = "This is a message with an <<
                self.embeddedString>>!";      // right

    (This is not a change, but simply a note for game developers who
    have encountered this problem.)


2.1.2  11/22/93  enhancements, bug fixes

  - You can now detect when the player uses multiple direct objects
    with a verb, and reject such commands.  Whenever the player uses
    multiple direct objects with a command (or uses "all", even if it
    results in a single object being used), the parser calls the
    verb object's rejectMultiDobj(prep) method.  If you don't wish to
    take any special action for multiple direct objects used with a
    particular verb, simply return nil from this method (or don't
    define the method at all for the verb).  If you want to prevent
    multiple direct objects from being used, however, you should display
    an appropriate message, and return true.  The parser will skip the
    command entirely.  Note that the parser doesn't display any additional
    message when rejectMultiDobj(prep) returns true; the method should
    display whatever message is desired.  The "prep" parameter is the
    preposition object used with the command; it will be nil if no
    indirect object is present.  An example:

        modify inspectVerb
            rejectMultiDobj(prep) =
            {
                "You can only look at one thing at a time.";
                return true;
            }

    The verb's rejectMultiDobj(prep) method is called immediately
    before the actor's actorAction method.  Note that the parser will
    continue processing any remaining commands on the line, and will
    then run daemons and fuses as normal, even if rejectMultiDobj(prep)
    returns true; if you want to stop the current turn altogether, use
    abort.

  - The player command parser now gives you greater control over object
    validation error reporting.  In previous versions, if an object was
    visible but did not pass the validIo/validDo test, the parser called
    the object's cantReach method to report the error (see the note below
    about an additional change to object.cantReach processing).

    Now, however, the parser will call verb.cantReach instead, if the
    command's deepverb object defines (or inherits) a cantReach method.
    If the verb does not have a cantReach method at all, the parser
    will use the old behavior instead.  The new cantReach method should
    be defined as follows:

       myVerb: deepverb
           verb = 'whatever'
           cantReach(actor, dolist, iolist, prep) =
           {
               // your code here
           }
       ;

    This method doesn't return a value; it simply displays the appropriate
    message explaining why the object can't be used with the command.
    verb.cantReach is used only when the objects are visible (that is,
    object.isVisible(actor) returned true for each object in the list).

    Only one of dolist or iolist will be non-nil.  If the direct object
    of the command refers to one or more objects that are visible but
    can't be used (according to validDo), dolist will be a list of all
    such objects, and iolist will be nil.  Otherwise, iolist will be a
    list of such objects used for the indirect object, and dolist will
    be nil.

    adv.t has not been changed to use verb.cantReach.  This change has
    been made to provide finer control for game authors implementing
    their own verbs and object validation procedures.

  - The player command parser had an odd quirk when ambiguous nouns
    were used with transparent items.  If the player used a command
    containing a noun that referred to multiple objects that were
    visible but were not valid for the verb (for example:  "take trophy"
    in a room containing a closed glass trophy case containing a bowling
    trophy and a tennis trophy), the parser asked the normal disambiguation
    question.  This was not really necessary, because the parser already
    knew that the objects were invalid.  This has been changed; the parser
    now simply uses the cantReach message for *each* object that is
    visible and matches the vocabulary, using the usual multiple-word
    format:
        bowling trophy:  You'll have to open the glass case first.
        tennis trophy:  You'll have to open the glass case first.

    This new behavior should have no effect on your game code.  Note
    that it is entirely irrelevant if you use the new verb.cantReach
    feature described above.

  - The compiler sports a new case sensitivity option.  By default,
    the compiler is case-sensitive, as it has been in past versions.
    However, the new toggle option allows you to change this.  Specify
    -case- to turn off case sensitivity (the default is -case+).  Note
    that this is a toggle option, so simply using -case will reverse
    the current case sensitivity (which is useful if you use a CONFIG.TC
    file that sets a non-default case option).  When case sensitivity
    is turned off, the compiler will treat upper- and lower-case letters
    in symbols (names of objects, properties, functions, and local
    variables) as equivalent.  Hence, foodItem = fooditem = FoodItem,
    and so on.

    If you create a pre-compiled header with -w, any compilation which
    reads that binary header with -l will use the same case sensitivity
    as was in effect when the header was pre-compiled.  The -case option
    is ignored when a pre-compiled header is loaded.  Likewise, the
    debugger uses the same case sensitivity that was in effect when the
    game being debugged was compiled.

  - The debugger's command set is no longer case-sensitive (hence,
    BP = Bp = bP = bp).

  - adv.t has a new darkVerb class.  This is a type of deepVerb that
    can be used in the dark.  The darkroom class has been changed to
    accept any verb of class darkVerb in its roomAction and roomCheck
    methods.  The travel verbs and system verbs have all been made
    darkVerb objects, so darkroom only needs to check the single
    verb type.

    In a related change, turnOnVerb and dropVerb have been changed to
    be darkVerb objects, allowing the player to turn on an object or
    drop it in the dark.  The verDoTurnon method in switchItem has
    been changed so that it checks to see if the player is in a dark
    room; if so, the item can only be turned on if the player is already
    carrying the object.  This allows the player to turn on a light
    source that's already being carried, but doesn't allow the player
    to pick up a light source in a dark room.

  - The compiler issues a new warning message (TADS-452) if you
    use the same verb with two deepverb objects.  The parser can only
    choose a single deepverb object for any verb typed by the player,
    so you should never define the same 'verb' vocabulary word in more
    than one deepverb object.  In past versions, the compiler did not
    flag this as a warning.

  - Several improvements have been made for numbers in player
    commands.  First, multiple numbers are now allowed in a single
    commands; for example, the player can now say "press 1, 2, 3,
    4 on keypad," and the numbers are set in numObj.value, one by
    one.  Second, the sdesc, adesc, and thedesc properties of
    basicNumObj have been improved to show the number's value.

  - Similar improvements to those for numbers have been made for
    strings.  Multiple strings are now allowed in a single command,
    and the basicStrObj properties sdesc, adesc, and thedesc have
    been improved to show the string's value.

  - specialWords has been enhanced.  First, in past versions, if a
    game contained multiple specialWords statements, the word lists
    were additive -- all specialWords lists were used in the game.
    This has been changed so that each specialWords statement replaces
    any previous list in effect.  However, you now can explicitly add
    to the specialWords list, without removing any of the previously
    defined words, by using "modify specialWords".  When you use
    'modify', you can use nil in any word slot if you do not wish
    to add any words for that slot.  Finally, you can use "replace
    specialWords" to make the replacement explicit; this is the default
    if neither 'modify' nor 'replace' is specified, but the compiler
    will now issue a warning (which is harmless) if you use specialWords
    without 'replace' or 'modify' and a previous specialWords list is
    already in effect.

  - The words "one" and "ones" (or their equivalent for your game,
    if you've changed them with specialWords) are no longer considered
    reserved words within normal commands.  This allows you to use
    objects such as a "one dollar bill"; previous versions rejected
    player commands containing "one" or "ones".  These words are now
    considered special only during the parsing of a response to a
    disambiguation question, when they can be used in place of a
    noun ("the red one" can be used in answer to "Which book do you
    mean...").

  - The hider class has been changed so that 'it' or 'them' (as
    appropriate) are set to the object or objects found when
    searching the hider.

  - The verDoPutIn and verDoPutOn messages in thing and surface
    (respectively) have been improved for the somewhat obscure case
    of attempted circular containment - that is, putting an object
    into a container, when the container is already in the first object
    (either directly or by virtue of being inside another object which
    is inside the first object, or inside an object which is inside an
    object which is inside the first object, and so on).  The new
    method thing.circularMessage(iobj) is called in these cases to
    display an appropriate message; the default implementation of this
    method displays the complete list of containers of the direct
    object out to the indirect object.  For example, if you have a
    crate which contains a box, and you try to "put crate in box",
    the message is "You can't put the crate in the box, because
    the box is already in the crate."

  - The default doTake method has been changed to include the weight
    of any contents of the item being taken, in addition to the item
    itself, to determine if the actor's inventory is too heavy.
    The old doTake method only included the weight of the object
    being taken, not counting its contents.

  - The Actor class has been changed to add a travelTo method.  You
    can now move any actor (Me included) with travelTo(destination).
    The default Actor.travelTo method moves the actor; it announces
    the departure of the actor if the actor was in the same location as
    Me before leaving (and the location is lit); and it announces the
    arrival of the actor if the actor is moving into the same location
    as the player (and the location is lit).  The departure message
    is generated with a call to self.sayLeaving, and the arrival
    message is generated with self.sayArriving.  The default versions
    of these methods simply display "Thedesc leaves the area" and
    "Thedesc enters the area", respectively; you can override these
    methods if a more specific message is desired.

  - When modifying a class object with 'modify', the modified object
    was not a class unless the 'class' keyword was included with the
    'modify' statement ("modify class foo" rather than "modify foo").
    This has been corrected; a modified class is still a class.

  - outhide(true) now returns a status indicator, which is a value that
    can be used in a subsequent call to outhide() to return output
    hiding to the state it was in before the outhide(true).  This
    allows you to nest text hiding.  When you use the nested form
    (which you do simply by using the return value of outhide(true) as
    the parameter - in place of nil - to the subsequent call to outhide()),
    the value returned by the second outhide() indicates whether any
    text output occurred ONLY BETWEEN THE NESTED CALLS.  For example:

     old_stat1 := outhide(true);
     "This is some hidden text.";
     old_stat2 := outhide(true);
     // write no text here
     new_stat2 := outhide(old_stat2);
     new_stat1 := outhide(old_stat1);

    Because outhide(old_stat2) indicates whether any output occurred
    during the NESTED outhide(true), new_stat2 = nil.  However, new_stat1
    = true, since output occurred after the first outhide(true).  Consider
    another sequence:

     old_stat1 := outhide(true);
     // write no text here
     old_stat2 := outhide(true);
     "This is some hidden text.";
     new_stat2 := outhide(old_stat2);
     new_stat1 := outhide(old_stat1);

    In this case, both new_stat1 and new_stat2 will be true, because
    hidden output occurred within both nested sections.

    The general form of a nested hidden output section looks like this:

         {
            local original_hide_stat;
        local nested_stat;

            hide_stat := outhide(true);
            // do whatever you want to do while output is hidden
        nested_stat := outhide(original_hide_stat);
     }

    Now nested_stat will indicate whether any output occurred during
    the nested outhide() - that is, between the outhide(true) and
    the outhide(original_hide_stat).  In addition, output hiding will
    be returned to the same state it was in prior to the original
    outhide(true).

  - The random number generator has been improved.  Many people have
    complained about the many undesirable properties of the old
    generator, especially when small upper limits were used.  The
    interface to the new random number generator is the same as
    before - call rand(upper_limit), which will return a uniformly
    distributed random number from 1 to upper_limit, inclusive.

    Note that the old random number generator will still be used if
    you don't call randomize().  This allows test scripts (which require
    a fixed sequence of random numbers in order to be repeatable) that
    were written with older versions to continue to operate unchanged.
    If you want numbers from the improved generator, be sure to call
    randomize().

  - When 'modify' was used on an object, the compiler sometimes did
    not correctly apply the original object's vocabulary and location
    to the new object.  This has been corrected.

  - restore() and undo() have been changed so that they always cancel
    all pending commands on the command line.  In the past, if the
    player typed several commands, and something happened (such as
    the player character dying) during one of the commands that led to
    an undo() or restore(), the remaining commands were still
    executed.  This has been fixed.

  - If you explicitly set an object's location to nil in its object
    definition, and the object inherited a location from a superclass,
    the system incorrectly placed the object in the contents list of
    the object named in the location inherited from the object's
    superclass.  This has been corrected.

  - "abort" can now be used within a daemon or fuse, and the
    expected behavior will occur.  In the past, "abort" within
    a fuse (or daemon) merely exited from the current fuse,
    but the remaining fuses and daemons were still executed.
    Now, "abort" will cause the entire turn to end; no more
    fuses or daemons will be executed on the current turn.


2.1.1  09/09/93  enhancements

  - You can now access objwords(1) while deciding whether to use a
    default direct object in doDefault.  This is useful mostly if
    you want to prevent players from being able to use "all" with
    certain verbs, but still want to generate a default direct object
    for the verbs.  To do this, you can detect when objwords(1) = ['A']
    ('A' is the parser's internal code for "all", which saves you the
    trouble of checking for "everything" and shorter abbreviations as
    well as "all"):

      doDefault(actor, prep, iobj) =
      {
          if (objwords(1) = ['A'])
          {
              global.allMessage := 'You can\'t use "all" with this verb.';
          return [];

      /* your normal default object code goes here */
      }

    If you wish, you can also suppress the default message that the
    parser will generate ("I don't see what you're referring to").
    To do this, you'll have to write your own parseError() function
    and detect when an "all" violation has occurred (cleverly using the
    global.allMessage, which we set above for this purpose):

      parseError: function(str, num)
      {
          // if there's an allMessage waiting, use it instead of the default
          if (global.allMessage <> nil)
      {
          local r;

          r := global.allMessage;
          global.allMessage := nil;
          return r;
      }
      else
          return nil;
      }

  - The compiler's error message format has been changed slightly
    to work better with editors and workbench programs that scan error
    logs to go to lines with errors.  The format is now:

        file(line): error TADS-xxxx: message

    For example:

        deep.t(1151): error TADS-300: expected colon


  - The parser now accepts sentences of the form VERB PREP IOBJ DOBJ,
    where the PREP is *not* part of the verb.  For example, GIVE TO THE
    MAN THE BALL.  This change has two benefits.  First, while this
    type of sentence is not common in English, some other languages
    allow this type of phrasing, so the parser is now somewhat more
    adaptable to non-English languages.  Second, this allows for
    object defaulting and command completion when specifying just the
    indirect object, which was not possible before.  For example, if
    the player types ASK FOR A BEER, the parser will be able to attempt
    to provide a default (if one is available), or at least ask for the
    direct object.  Previous versions would simply say "I don't understand
    that sentence."  Note that the parser still attempts to combine the
    verb and preposition into a single phrase; the new action happens
    only when the verb and preposition don't go together (that is, they
    haven't been defined together as a "verb =" property of a deepverb).
    For example, suppose that a line like this appears in a deepverb:

        verb = 'pick up'

    In this case, PICK UP THE BOX will use THE BOX as the direct object,
    just as in previous versions.  Only when the verb-preposition combination
    is not specifically defined in a verb will the new phrasing be used.

  - When no preposition is specified between the direct and indirect objects,
    the parser will now evaluate a new property, nilPrep, in the deepverb
    object.  This property should return the preposition object that should
    be used as the preposition between the objects.  Previous versions of
    TADS always looked for an object that defined the word 'to' as a
    "preposition =" property.  While 'to' is almost always the correctly
    the correct preposition to substitute in English, it's obviously the
    wrong word in other languages; furthermore, the correct word in other
    languages is sometimes a function of verb.  If no nilPrep property is
    defined for the deepverb, the parser will still use the object whose
    "preposition =" property matches the word 'to'.

  - The class transparentItem in adv.t has been modified so it works better
    when you define an object that inherits from both transparentItem and
    container or openable.  First, an ldesc has been added so that the
    contents of a transparentItem are listed by default with the ldesc.
    Second, the "look in" command now works on a transparentItem.  In
    addition, the openable class has been changed so that the "look in"
    command can be used when an openable is also a transparentItem, even
    when the openable is closed (because you should be able to see the
    contents of a transparentItem regardless of whether it's open or closed).
    Thanks to Ron Hale-Evans for pointing out this problem and finding the
    solution.

2.1.0  04/07/93  new features, enhancements, bug fixes

  - TADS finally has a way of changing objects and functions in
    adv.t without changing adv.t itself.  The new mechanism allows
    you to entirely replace a previously defined object or function
    with one of your own, and also lets you modify a previously
    defined object by adding or overriding properties.  Two new
    keywords have been added to the language to support these new
    features:  "replace" and "modify".

    Using these new features, it should be possible to make most
    of the changes to adv.t that are necessary while writing a game
    without actually changing the file adv.t itself.  This should
    make version upgrades much easier to apply, since you shouldn't
    need to reconcile any changes you have made to adv.t with the
    new version.

    High Energy Software requests that you advise us of any changes
    to adv.t that would facilitate modification and replacement of
    the objects defined in adv.t.  Examples would include common
    code fragments that could be moved into a function for easy
    replacement, and single deepverb objects that should be split
    into multiple objects.

    You can replace a previously-defined function with a new
    definition simply by prefacing your new definition with the
    keyword "replace"; for example, to replace adv.t's new
    scoreStatus function, you could do this:

       #include <adv.t>

       replace scoreStatus(points, turns)
       {
          setscore(cvtstr(pts) + ' points/' + cvtstr(turns) + ' moves');
       }

    You can do the same thing with objects.  For example, to replace
    a verb in adv.t, you could do something like this:

       #include <adv.t>

       /* we don't want "buckle", so replace adv.t's fastenVerb */
       replace fastenVerb: deepverb
          verb = 'fasten'
      sdesc = "fasten"
      prepDefault = toPrep
      ioAction(toPrep) = 'FastenTo'
       ;

    Replacing an object entirely deletes the previous definition of
    the object, including all inheritance information and vocabulary.
    The only properties of a replaced object are those defined in the
    replacement; the original definition is entirely discarded.

    You can also modify an object, which retains its original definition
    (including inheritance information, vocabulary, and properties), and
    allows you to add new properties and vocabulary.  You can also
    override properties, simply by redefining them in the new definition.
    The most common addition to an object from adv.t will probably be
    new verb associations; for example:

       modify pushVerb
          verb = 'nudge'
          ioAction(withPrep) = 'PushWith'
       ;

    Note several things about this example.  First, no superclass
    information can be specified in a "modify" statement; this is because
    the superclass list for the modified object is the same as for the
    original object.  Second, note that vocabulary has been added.
    The additional vocabulary does NOT replace the original vocabulary,
    but simply adds to the previously-defined vocabulary.  Further note
    that verb association pseudo-properties, such as doAction and ioAction,
    are legal in a "modify" definition.

    In a method that you redefine with "modify", you can use "pass"
    or "inherited" to refer to the REPLACED method.  In essence, using
    "modify" renames the original object, and then creates a new
    object under the original name; the new object is created as a
    subclass of the original (now unnamed) object.  There is no way
    to refer to the original object, except indirectly through the
    new replacement object.  Here's an example of "modify" and "pass":

        class testClass: object
        sdesc = "testClass"
    ;

        testObj: testClass
            sdesc =
        {
            "testObj...";
        pass sdesc;
        }
        ;

        modify testObj
            sdesc =
        {
            "modified testObj...";
        pass sdesc;
        }
    ;

    Evaluating testObj.sdesc results in this display:

        modified testObj...testObj...testClass

    However, you can override this behavior for a property by
    using the replace keyword on the property.  In the example
    above, we could do this instead:

        modify testObj
            replace sdesc =
        {
            "modified testObj...";
        pass sdesc;
        }
    ;

    This would result in the following display for testObj.sdesc:

        modified testObj...testClass

    The "replace" keyword before the property definition tells the
    compiler to completely delete the previous definitions of the
    property.  This allows you to completely replace the property,
    and not merely override it, meaning that "pass" and "inherited"
    will refer to the property actually inherited from the superclass,
    and not the original definition of the property.

  - It is now possible for the player to customize the colors used
    by the runtime.  A small new program, TRCOLOR, is provided to
    set up the runtime screen colors.  The program is self-explanatory;
    simply type TRCOLOR at the DOS prompt to invoke it.  Once you've
    selected your color scheme, TRCOLOR will create a small file
    called TRCOLOR.DAT (in the current directory); the runtime will
    read this file in subsequent game sessions.

    Note that you can use multiple TRCOLOR.DAT files, in much the same
    way you can use two CONFIG.TC files.  The runtime looks first for
    TRCOLOR.DAT in the current directory; if no such file is found, the
    runtime will then look in the directory where TR.EXE resides.  So,
    you can set up separate color schemes for each game you're playing,
    and have a default color scheme for games with no specific color
    scheme of their own.

  - The user interface of the MAKETRX program has been improved.
    For compatibility with existing makefiles, the old command line
    syntax is still allowed; however, you can now omit most of the
    arguments, and MAKETRX will use convenient new defaults.

    First, you can now omit the extensions on all the arguments.
    The extension assumed for the TR.EXE program is .EXE; for the
    game file it is .GAM; and for the executable output file it is .EXE.

    Second, you can now omit everything except the name of the game
    file.  If you omit the name of the TR.EXE program, MAKETRX attempts
    to find TR.EXE in the same directory as MAKETRX.EXE; so, if you
    simply keep all of your TADS executables in a single directory,
    you won't need to specify the location of TR.EXE when running MAKETRX.
    If you omit the name of the destination file, MAKETRX will use the
    same name as the game file, with the extension replaced by .EXE.

    The command line arguments to MAKETRX that are now understood are:

       maketrx gamefile
         Converts gamefile.gam into gamefile.exe, using TR.EXE from the
     same directory as MAKETRX.EXE.

       maketrx gamefile output
         Converts gamefile.gam into output.exe, using TR.EXE from the
     same directory as MAKETRX.EXE.

       maketrx \tads2\tr.exe gamefile output
         Converts gamefile.gam into output.exe, using \tads2\tr.exe.


  - The dobjGen and iobjGen mechanism has been changed slightly.
    In the original implementation, you could prevent the system
    from calling dobjGen/iobjGen by defining an appropriate verXoVerb
    property in the actual object, but NOT in a superclass.  This
    made it impossible to define a class that had exceptions to
    dobjGen/iobjGen except by explicitly testing for those verbs
    in the xobjGen routines.

    The change is that the system will now skip calling xobjGen
    if an appropriate verXoVerb/xoVerb property is defined in such
    a way that it "overrides" xobjGen for the object.  Here's an
    example:

       class cursedItem: item
           dobjGen(a, v, i, p) =
       {
           "As you touch <<self.thedesc>>, a bolt of lightning
           leaps from the object and sends you reeling away!";
       }
       iobjGen(a, v, d, p) = { self.dobjGen(a, v, d, p); }
       verDoInspect(actor) = { pass verDoInspect; }  // allow 'inspect'
       ;

    The change means that the presence of verDoInspect in the *class*
    prevents the system from calling dobjGen when the verb is "inspect",
    even for subclasses.  With the old system, since the subclass objects
    themselves didn't define verDoInspect, dobjGen would be called even
    though the verDoInspect logically overrides the dobjGen.

  - The restore() intrinsic has been extended to allow your game
    program to explicitly restore the saved game specified by the
    user as a parameter to your stand-alone game program.  This is
    currently only useful on the Macintosh, but the inclusion of
    code to test for this case will make your game work better on
    the Macintosh (and possibly other platforms in the future).

    The new functionality is invoked by calling restore(nil).  If
    a saved game was specified by the user at start-up time, the
    game will be restored, and nil will be returned.  If no file
    was specified, or an error occurred restoring the game, the
    function returns true.  To use this new behavior, we recommend
    placing the following code in your init function, before your
    introductory messages and other startup code:

       // check for a file to restore specified as a startup parameter
       if (restore(nil) = nil)
       {
           "\b[Restoring saved game]\b";
           scoreStatus(global.score, global.turnsofar);
           Me.location.lookAround(true);
           return;
       }

    Note that the run-time will still automatically restore the
    game provided as a parameter (on the Macintosh) after init returns
    if you do NOT include this code in init.  The reason for including
    this code is that it provides your game with greater control over
    the sequence of events during startup.  If you allow the run-time
    to perform the restore automatically, your entire init function will
    be executed; this may be undesirable, because it forces the user to
    view the entire introductory text even though they'll immediately
    restore a game after reading it.  If you place the restore(nil) call
    before your introductory text, the user will be spared the long text
    display; however, you'll still have complete control over any text
    that you want the user to see even when restoring a game, such as
    your copyright message.

  - New built-in function:  objwords(num), which provides a list of
    the actual words the user typed to refer to an object used in
    the current command.  The argument (num) is a number specifying
    which object you're interested in:  1 for the direct object, or
    2 for the indirect object.  The return value is a list of strings;
    the strings are the words used in the command.  If a special
    word, such as "it", "them", or "all", was used to specify the
    object, the list will have a single element, which is the special
    word used.

    Examples:

       >take all
       objwords(1) -> ['all']
       objwords(2) -> []

       >put all in red box
       objwords(1) -> ['all']
       objwords(2) -> ['red' 'box']

       >put blue box in it
       objwords(1) -> ['blue' 'box']
       objwords(2) -> ['it']

       >put blue folder and green book in red box
       blue folder:
       objwords(1) -> ['blue' 'folder']
       objwords(2) -> ['red' 'box']
       green book:
       objwords(1) -> ['green' 'book']
       objwords(2) -> ['red' 'box']

    This function could potentially be useful in such cases as
    "ask actor about object", because it allows you to determine
    much more precisely what the player is asking about than would
    otherwise be possible.

  - The setit() function now takes nil as a parameter; this prevents
    using "it" in a command until another object has been referenced.
    nil can be used for "him" and "her" as well, as described below.

  - Enhancements to the setit() built-in function:  you can now directly
    set the 'him', 'her', and 'them' values using the setit() function.

       - To set 'them', simply call setit() with a list value:
         setit([redBook blueBook boots]);

       - To set 'him', add a second argument value of 1 to the call:
         setit(joe, 1);

       - To set 'her', add a second argument value of 2:
         setit(harriet, 2);

  - The restart() built-in function now takes an optional set of
    arguments:  a pointer to a function, and a parameter value for
    the function (if one is provided, the other is also required, but
    both can be omitted).  If they're provided, TADS calls the function
    with the provided parameter value *after* resetting the game, and
    before running the init() function.  This allows you make the game
    start slightly differently after a restart than on the initial startup.
    adv.t uses this feature to call the function initRestart, with the
    parameter value global.initRestartParam, upon restart.  The initRestart
    function defined in adv.t simply sets the flag global.restarting to
    true.  Your game can test global.restarting (in the init function or
    elsewhere) to determine whether the game has been restarted, or is
    being run for the first time.  You can also replace initRestart()
    with your own function if you wish to do something more complicated;
    in this case, if you wish to pass information to the function, you
    can simply store it in global.initRestartParam, and it will be passed
    to the function automatically by adv.t upon restarting.

  - New built-in function:  inputkey() reads a single keystroke from
    the keyboard.  The function takes no arguments.  When called,
    inputkey() will flush any pending output text, then pause the game
    until the player hits a key.  It then returns a string containing
    the single key hit by the player.  Note that the function does NOT
    provide a portable mechanism for reading non-standardized keys;
    special keys such as cursor arrow keys and function keys will return
    a string specific to the type of computer being used.  Your game
    will not be portable if you make use of any non-standardized key
    values returned by inputkey().  To ensure portability, use inputkey()
    strictly with standard ASCII keys (alphabetic, numeric, and punctuation
    keys).  It is also fully portable if you simply ignore the return value
    and use the function only to pause and wait for a key.

  - Several changes have been made for better national language
    support.  First, the DOS version now allows 8-bit characters
    (characters in the extended character set, with ASCII code
    from 128 to 255) in text displayed by the game, vocabulary
    words, and player commands.  Characters in the extended character
    set are always considered to be alphabetic, so these characters
    can only be used in input as parts of words (hence, symbols
    from the extended character set that appear as punctuation
    marks can't be used as punctuation in player commands).

  - The debugger now has "More" mode in the command window.  When
    a single display won't fit in the window (for example, a long
    stack traceback), the debugger will prompt with "[More]" each
    time the window fills up.  Hit the space bar to scroll by a
    whole screen, or the Return/Enter key to scroll by a single line.

  - The debugger has a new "call history" feature.  This feature
    captures information about every function and method call,
    including argument lists and return values (if any), and saves
    the information for future inspection.  Several new commands
    have been added to the debugger to support call history:

        c+    Enables call history, and clears previous history.
        c-    Disables call history capture.
        cc    Clears all current history information.
        c     Displays current history information.

    The reason that call history can be enabled and disabled is
    that enabling the feature slows down the debugger substantially,
    because it must store information every time a method is called.

    This feature could be useful if you're trying to figure out the
    sequence of method calls that occurs during the execution of
    a command.  At the debugger command line, type c+ to turn on
    call history; then, type g to resume your game.  Type the command
    that you want to debug, then type DEBUG at the game prompt to
    return to the debugger.  Now type c- to turn off call history,
    and c to display the history information from the command you
    just executed.  This will allow you to see every method and
    function that was called by TADS, as well as all the methods
    and functions called by your code.

    The call history display will have each function/method call indented
    by a number of spaces indicating the nesting depth; any method/function
    called by TADS will be at the left margin, any methods/functions called
    by the first one will be indented one space, any methods/functions
    called by those will be indented two spaces, and so on.  The return
    values will be indented by the same number of spaces as the function
    itself was.  Note that a return value may be separated from its
    entrypoint by several lines, because calls made by the function
    will appear between the function entry and the return value.

  - Another new national language feature is the addition of
    several new parser-called user functions that allow better
    user control over the generation of parser messages.  The
    new functions have been added because some people have found
    that the parseError() function is not sufficiently flexible
    for some situations, because it only allows changing the text
    of messages on a piecewise basis; when complete messages need
    to be built out of several pieces, it's necessary to be able to
    take over the entire process of building the message.  The new
    functions allow full control of the generation of certain messages.

    parseAskobj(v, ...):  This function is called when the parser
    needs to ask the player for a direct or indirect object to complete
    the command.  For example, if the player just types "take", and
    several objects are present that could be taken, the parser must
    ask the player what to take.  If a direct object is being requested,
    the function will have only one argument (the verb).  If an indirect
    object is being requested, the function will have *two* arguments;
    the second argument will be the preposition.  Note that the preposition
    can be nil, in which case you can assume that "to" is to be used.
    The implementation below emulates the default behavior.

        parseAskobj: function(v, ...)
    {
        "What do you want to <<v.sdesc>>";
        if (argcount = 2)
        {
            local p := getarg(2);
            " it << p ? p.sdesc : "to" >>";
        }
        "?";
    }

    parseDisambig(string, list):  This function is called by the
    parser when objects need to be disambiguated.  If this optional
    function is provided, it is called with the string that the player
    typed which is in need of disambiguation, and a list of the objects
    that match the string.  The implementation below emulates the
    parser's default behavior.

        parseDisambig: function(str, lst)
        {
            local i, tot, cnt;

            "Which << str >> do you mean, ";
            for (i := 1, cnt := length(lst) ; i <= cnt ; ++i)
            {
                lst[i].thedesc;
                if (i < cnt) ", ";
                if (i+1 = cnt) "or ";
            }
            "?";
        }

    parseError2(v, d, p, i):  The parser calls this function to
    generate the default error message stating that the verb attempted
    isn't accepted by the objects involved; this happens when either
    the indirect object doesn't define an appropriate verIoXxxx method,
    or the direct object doesn't define an appropriate verDoXxxx method.
    Only one of 'd' (direct object) or 'i' (indirect object) will be
    non-nil.  If 'i' is nil, so will 'p' (preposition).  The verb, 'v',
    will never be nil.  Note that 'p' can be nil even when 'i' is not,
    in which case you should assume that the preposition is "to".
    The implementation below behaves the same as the parser's default.

        parseError2: function(v, d, p, i)
        {
            "I don't know how to << v.sdesc >> ";
            if (d)
                "<< d.thedesc >>.";
            else
            "anything << p ? p.sdesc : "to" >> << i.thedesc >>.";
        }

    parseDefault(obj, prp):  This function is called when the parser
    is assuming a default object.  If a default direct object is being
    assumed, prp (the preposition) will be nil; otherwise, prp will have
    the object corresponding to the preposition preceding the indirect
    object.  The implementation below provides the default behavior.

        parseDefault: function(obj, prp)
        {
            "(";
            if (prp) "<< prp.sdesc>> ";
            obj.thedesc;
            ")";
        }

    Note that all three of these new functions are optional.  If any
    is omitted, the parser uses the default behavior, so existing games
    will run unchanged.  You can include any one without including the
    others; these new functions are all independent.  Note also that
    the default parser behavior continues to use parseError the same
    way it has since parseError was introduced; however, when these
    new functions are provided, the corresponding parseError calls will
    obviously no longer be made.

  - COMPATIBILITY NOTE:  By default, the .GAM files produced by
    the 2.1.0 compiler will NOT be compatible with previous
    versions of the runtime, due to several changes to the .GAM
    file format.  However, a new compiler option has been added
    that allows you to specify which .GAM file format to produce:

      -fv a   produces .GAM format compatible with 2.0.14 or earlier
      -fv b   produces game file format requiring 2.1.0 or later
      -fv *   (default) produces latest file format (currently b)

    If you want your game to be compatible with older versions of
    the runtime, use -fv a.  The 2.1.0 runtime is compatible with
    .GAM files produced by ANY version of the compiler; the runtime
    automatically detects which file format version it is reading.

    Note that using -fv a will prevent you from being able to call
    an external function from within the init function (see the bug
    fix described below).  In addition, even when using -fv a, since
    previous versions of the run-time did not provide the new built-in
    functions, your game will be incompatible with runtimes prior to
    2.1.0 -- regardless of whether you use -fv a or not -- if you
    use any new built-in functions.

    In the future, if there is another incompatible .GAM file format
    change, additional -fv options will be added to the new compiler.

  - One of the changes to the .GAM file format makes it much more
    compressible with archiving tools (such as ZIP).  Previous
    .GAM files typically compressed by only 10 to 20%; the files
    produced with file format B are generally compressible by 40
    to 50%.

  - runfuses and rundaemons had a bug that reset the run-time
    stack, causing problems if a nested method or function
    called these functions.  This has been corrected.

  - Subtracting one list from another didn't work as documented.
    This has been corrected.

  - In previous versions, external functions could not be called
    while the init function was running.  This was an unintentional
    side-effect of the way external functions were loaded, and the
    problem has been corrected.  External functions can now be called
    at any time.

  - A new warning has been added that can help you track down
    unterminated strings.  Whenever the compiler sees a line that
    begins with a semicolon or close brace (';' or '}') inside
    a string, it will issue a warning.  While this is just a guess
    that the string may be unterminated, it's often right, especially
    if you follow the general formatting style used by adv.t:  always
    end a function with a brace in the first column of a new line,
    and always end an object with a semicolon in the first column of
    a new line.

    Note that we strongly recommend that you follow this formatting
    style in your code, both for general readability and because it
    may enhance your code's compatibility with future High Energy
    Software products that use assumptions about formatting style
    that are similar to that used to generate the new unterminated
    string warning.

  - Several small enhancements and bug fixes have been made to
    adv.t:

       - A new property has been added to nestedroom objects:
         statusPrep, which displays "on" or "in" (or whatever),
     as appropriate, for messages such as "Spaceship, in chair".
     This defaults to "in" for nestedroom, and "on" for beditem.
     Other nestedroom objects you define may want to customize it.

       - There was a bug that allowed the player to throw a fixeditem
         that was (indirectly) being carried (for example, a fixeditem
         that is part of another object that can be carried) at something.
     This has been fixed.

       - The follower class did not 'exit' at the end of its actorAction.
         This has been fixed.

       - The follower class now makes use of iobjGen and dobjGen to respond
         with an appropriate message ("the actor is no longer here") to
     any command other than "follow".

       - The clothingItem class has been enhanced to allow "get out of"
         to be used to take off the item.

       - All of the verbs containing the word "look" now have synonyms
         with "l" as well:  l at, l on, l in, l under, l around, l thru,
     and so on.

       - A bug has been fixed that allowed the command "take all from foo"
         to remove the contents of "foo" even if it was closed.  The
     change is to the doDefault method in takeVerb.

       - The vehicle class has been adjusted so that the player can't
         take a vehicle or otherwise manipulate it while the player is
     currently in the vehicle -- this is important for things such
     as rubber rafts which can be used both as vehicles and ordinary
     carryable items.  dobjGen and iobjGen are used to accomplish
         this; the only allowed verbs on a vehicle while it's occupied
     by the player are inspectVerb, getOutVerb, outVerb, and putVerb
     with the vehicle as an indirect object (allowing objects to be
     put into the vehicle while it's occupied).  If you want to allow
     additional verbs in a particular vehicle, override dobjGen or
     iobjGen as appropriate, and simply return if the verb matches
     any of the verbs you wish to add:

         dobjGen(a, v, i, p) =
         {
             // allow "launch" and "land" while in the magic raft
             if (v <> launchVerb and v <> landVerb)
             pass dobjGen;
         }

  - The compiler now detects usage (both explicit and implicit)
    of "self" in functions.  This has always been illegal, but
    in previous versions the compiler did not notice; any uses
    of "self" in functions resulted in a run-time error (often
    a mysterious error, such as a cache manager severe error and
    abnormal termination due to a reference to a non-existent
    object).  This was especially troublesome when a property
    name was used as a local variable when the local variable
    wasn't declared; since the v2 compiler assumes "self" in
    references to properties that don't include an object
    qualification, the compiler would silently turn an undefined
    variable usage into a reference to "self".  The compiler will
    now flag a new error in these cases:  TADS-344, "self" is not
    valid in this context.  If you get this error without an
    explicit reference to "self", you probably have an implicit
    reference, which means you probably are using an undeclared
    local variable.  Adding a "local" declaration for the variable
    should clear the error.

  - "Her" was not set properly, even when the isHer property was
    set to true for an object.  This has been corrected.

  - A new compiler option has been added:  -v, for "verbosity".
    This option lets you tell the compiler how much warning
    information you'd like to see.  By default, the verbosity
    level is 0 (zero), which causes certain warnings to be
    suppressed.  You can specify -v followed by a number to
    set a higher verbosity level.  So far, only the messages
    listed below are affected by -v, but the verbosity level
    for certain warnings may be changed in the future (and new
    warnings may be added at high verbosity levels).  Currently,
    the general meaning of the verbosity levels is:  0, report
    only serious errors and warnings; 1, report suspicious
    situations that may or may not indicate errors; 2, report
    all information, including general warning information that
    usually does not indicate any actual problem.

  - A new compiler option has been added:  -e file, for "error logging".
    This option captures all messages generated by the the compiler to
    the specified file.  Messages are also displayed interactively as
    normal.  WARNING: if the file specified with -e already exists, it
    is overwritten with the error information.

  - The compiler warning messages about "object not found" for the
    optional objects (preinit, preparse, parseError, commandPrompt)
    are now suppressed if the verbosity level is less than 2.
    If you specify -v2 (or -v with a higher number than 2), these
    messages will be displayed for all optional objects not found;
    otherwise, no warnings will be generated.

  - The compiler warning messages about "#include file already included"
    are now suppressed if the verbosity level is less than 1.


2.0.14  02/10/93  bug fixes, minor enhancements

  - A new backslash code has been added to the output formatter that
    causes the formatter to pass the next two bytes unchanged.
    This has been added primarily for 16-bit character sets, to allow
    two-byte characters that contain a backslash ('\', ASCII 92) as
    one of their two bytes to be passed through the formatter without
    interpretation as part of a backslash sequence.  The new code
    is "\-"; the two bytes following the \- are not interpreted by
    the formatter.  For example:

       "\-\bTesting...\nDone!";

    displays:

       \bTesting...
       Done!

    Note that the "\b" sequence is not interpreted as a blank line,
    as it would normally be, but is simply displayed, because the \-
    suppresses any interpration of the next two bytes.  The "\n",
    however, is interpreted as a newline as normal, since it is not
    quoted by a \- sequence.

  - You can now break out of an infinite loop in your game while
    running under the debugger.  On DOS, if your game goes into a loop,
    hit the Control + Break keys - the loop should immediately be
    aborted and control returned to the debugger command line.
    The Control-Break sequence also works with the runtime; it
    causes control to be returned to the player command line.
    Note that the interrupted command is automatically undone, so
    the interrupt sequence will not leave the game in an inconsistent
    state.  Note also that only long loops can be interrupted; the
    system only checks for interruptions once every several hundred
    instructions for greater efficiency.

  - The debugger will now catch run-time errors, activating the
    debugger command line when an error occurs.  The source will
    be positioned at the location of the error, as though a breakpoint
    had been set there, and the error message will be displayed.  Local
    variables can be evaluated as normal to help determine the cause of
    the error.  When you resume execution (with Trace, Step, or Go),
    the current command will be aborted and control will return to the
    player command prompt.  Note that there's no way to "fix" the error
    once it's been caught, but you can at least see exactly where the
    error occurred and determine the conditions that caused it.  Note
    also that certain types of errors, such as memory management errors,
    will not invoke the debugger; only errors that are directly caused
    by an error in your game program will trap to the debugger.

  - The debugger incorrectly reported files as "not found" in the
    list of modules produced by the "fl" command.

  - The runtime was inconsistent in its calls to ioDefault.  Sometimes
    it called ioDefault(actor, prep), and other times it called it as
    ioDefault(actor, prep, dobj) - this made it impossible to define the
    method correctly if argument checking was enabled.  This has been
    corrected so that the dobj parameter is never included.  When attempting
    to identify a default indirect object, the parser never has a direct
    object available, since the indirect object must be resolved first;
    hence, the dobj that was occasionally being passed by the parser was
    always nil.  The unnecessary extra parameter has been removed:  the
    method is now always called as ioDefault(actor, prep).

  - The compiler generated incorrect code if the implicit "self" object
    was used to reference an object (that is, a property was used without
    an object specifier).  This resulted in "invalid opcode" errors at
    run-time.

  - The compiler sometimes complained that an included file couldn't
    be found, even when the included file was explicitly loaded as part
    of a precompiled header.  This happened any time the included file
    was not in the current directory at compilation time.

  - The compiler aborted with an "assertion failure" (which indicates
    that the compiler detected that it was in an internally inconsistent
    state, which should not be attainable under any circumstances) when
    the game program used a variable or expression on the right hand
    side of a dot operator and an object on the left hand side.

2.0.13  01/16/93  enhancements and bug fixes

  - If a vocabulary word contained a single quote character, the
    word could not be matched at run-time.

  - The run-time now allows all characters from the extended character
    set (ASCII codes above 127) to be displayed.  The run-time
    previously converted some extended characters into spaces.

  - The compiler did not allow a label to precede the first goto
    that referred to the label.

  - The debugger will now stop at a breakpoint in a method that
    is inherited by an object.  For example, if a breakpoint is
    set at room.lookAround, and startroom inherits lookAround
    from the class room, the debugger will stop at startroom.lookAround.
    It does not, however, stop on startroom.lookAround if startroom
    overrides lookAround.

  - The compiler will now flag an assignment to an undeclared
    symbol as an error.  It previously assumed that the symbol
    referred to a property, with an implicit object of "self".
    This was almost never desirable, because this type of
    assignment was most often coded in error -- the game author
    usually meant to code an assignment to a local variable, but
    either misspelled the variable name or forgot to declare it.

  - remfuse/remdaemon/unnotify no longer signal an error if the
    item being removed is not active.  Several game authors have
    indicated that this error is not helpful, since it makes it
    impossible to unconditionally remove a fuse -- you have to
    check to make sure it hasn't fired yet, which create a lot
    of unnecessary overhead.

  - NEW BUILT-IN FUNCTION:  intersect(list1, list2) returns the
    intersection of two lists; that is, it returns the list of
    all of the elements of the shorter of list1 and list2 that
    are also in the other list.  For example:

        [1 2 3 4 5] and [1 3 5 7]     ->    [1 3 5]
        ['abc' 'def'] and ['abc']     ->    ['abc']

    This new function can be used to improve performance in cases
    where (effectively) one list of items is being searched for
    the presence of another list of items.

  - NEW BUILT-IN FUNCTION:  runfuses() runs all expired fuses, if any.
    Returns true if any fuses expired, nil otherwise.  This function has
    been added to allow greater control over fuse processing.  Note that
    fuses set with both the setfuse() and notify() built-in functions
    are run.  This function takes no arguments.

  - NEW BUILT-IN FUNCTION:  rundaemons() runs all daemons.  This function
    runs daemons set with both the setdaemon() and notify() functions.
    rundaemons() takes no arguments and returns no value.

  - NEW BUILT-IN FUNCTION:  getfuse allows you to determine if a fuse
    (set with either setfuse or notify) is active.  It returns nil if
    the fuse is not active (i.e., it has been activated or removed),
    or the number of turns left.

    For setfuse() fuses:   getfuse(fuse_func, parameter)
    For notify() fuses:    getfuse(object, &message)

  - NEW BUILT-IN FUNCTION:  gettime() returns the current time.  The
    time is returned as a list of numeric values for easy processing by
    your game:

      [year month day weekday yearday hour minute second elapsed]

    The specific meanings of the values are:

          year        - calendar year (e.g., 1992).
          month       - month number (January = 1, February = 2, etc.)
          day         - number of the day within the current month
          weekday     - day of the week (1 = Sunday, 2 = Monday, etc.)
          yearday     - day of the year (1 = Jan 1)
          hour        - hour of the day on 24-hour clock (midnight = 0,
                        noon = 12, 3 PM = 15, etc.)
          minute      - minute within the hour (0 to 59)
          second      - second within the minute (0 to 59)
          elapsed     - the number of seconds since January 1, 1970,
                        00:00:00 GMT.  This last value is useful for
                        computing the difference between two points
                        in time.

  - NEW FEATURE:  The parser now calls an additional method in
    each direct and indirect object under certain circumstances.
    These new methods are called dobjGen (general use of an object
    as a direct object) and iobjGen.  These methods are called
    immediately prior to the appropriate verXo<Verb> method.  The
    sequence of calls depends on the command, as detailed below.

    The purpose of these methods is to allow you to define a catch-all
    routine that is called whenever an object is used in a command.
    It is sometimes desirable to be able to take some action whenever
    an object is mentioned, regardless of the verb involved.  For
    example, you might wish to define a cursed object that damages
    the player (such as by taking away a treasure) whenever the
    object is touched or manipulated in any way; these new methods
    make it possible to do this without having to override every
    possible verb handler.

    When a command is issued with an indirect object, the parser
    checks to see if the indirect object *directly* defines the
    io<Verb> method.  If not, the parser calls iobj.iobjGen(actor,
    verb, dobj, prep).  The parser then checks to see if the direct
    object *directly* defines the verDo<Verb> method.  If not, the
    parser calls dobj.dobjGen(actor, verb, iobj, prep).

    When a command is issued with only a direct object, the parser
    checks to see if the object *directly* defines the do<Verb> method.
    If not, the parser calls dobj.dobjGen(actor, verb, nil, nil).

    Note that an object "directly defines" a method if the method
    is defined in the object itself -- that is, the object does
    not merely inherit the method from its class.

    These new methods have no return value, and need not do anything.
    If they're undefined, the behavior is exactly the same as in
    previous versions.  So, existing games should continue to run
    unchanged.

    The reason that these methods are not called when the object
    directly defines an appropriate verb handler is that these
    methods are intended as a generic catch-all verb handler.
    When a specific handler for the current verb is already defined
    in the object, there should be no need to call the generic
    handler, since the object already defines specific behavior
    for that verb.


2.0.12  No such release (for internal release coordination)  bug fixes

  - Switch statements did not properly process all datatypes.

  - Assignment operations did not work correctly on list elements.
    For example:  l := [1 2 3]; l[2] += 5; did not properly leave
    the value of l as [1 7 3].

2.0.11  12/20/92  bug fixes

  - Goto statement labels were occasionally not released properly,
    resulting in spurious internal errors.

  - The 'continue' statement did not work as documented in 'for'
    statements.  Instead of jumping to the re-initializer expression,
    as it now does, the 'continue' incorrectly jumped to the condition.

  - The run-time is slightly smaller and faster.

  - The compiler did not process locals correctly when multiple disjoint
    blocks within a function or method had locals, and a later block had
    fewer locals than a previous block (yes, it's a somewhat obscure bug).

2.0.10  No such release (to synchronize with Mac release levels)

2.0.9  12/12/92  MS-DOS bug fixes and new features

  - The file selector dialog displayed incorrect information when a
    directory name that used all 11 characters was displayed.

  - The file selector now saves and restores all default directory and
    disk information.  The current disk, and the current directory on
    each disk, will be the same when TR is terminated as it was when
    TR was first run.  (This applies to TDB as well.  It's particularly
    important for TDB, because TDB needs to have the source files in
    the current working directory if an absolute path was not specified
    with -I.)

  - NEW FEATURE:  the new user function commandPrompt, if provided by
    the user's game program, will be called prior to each player command.
    If the commandPrompt function is provided, the default ">" prompt is
    NOT displayed; if no commandPrompt function is defined, the default ">"
    is used.  This should not affect existing games at all, unless a game
    defines its own function, method, or property called commandPrompt
    having a different purpose.  The commandPrompt function returns no
    value.  The function takes a single argument:  a number, indicating
    the type of command that the system is prompting for:

       0  - normal command
       1  - command after invalid word (allowing "oops" to be used)
       2  - disambiguation (after "which x do you mean..." question)
       3  - command after askdo (after "what do you want to x?")
       4  - command after askio

    Note that the default prompt in all cases is simply ">", and in all
    cases a new command can be entered.  However, when the type code is
    2, 3, or 4, a question has just been displayed by the run-time, so
    the commandPrompt function may want to suppress any pre-command
    information or prompt text.  Case 1 is generally identical to case 0.

    NOTE:  As with the other optional user-provided functions, the
    compiler will issue a warning message if commandPrompt is not
    defined by your game.  If your game doesn't provide a commandPrompt
    function, you can ignore this warning.  The warning is provided so
    that, if you intended to provide a commandPrompt function, you will
    be informed if the compiler didn't find it (which could mean that
    you forgot to define it, or misspelled it).

  - NEW FEATURE:  A new built-in function has been added, which allows
    the game program to suppress the display of text that would normally
    be displayed with double-quoted strings or the say() function.  The
    function is called outhide(), and it takes one argument:  a flag,
    indicating whether to suppress or re-enable the display of output.
    If the flag is true, output is suppressed; if the flag is nil, output
    is re-enabled.  Any output that occurs between outhide(true) and
    outhide(nil) is discarded.  However, outhide(nil) returns a value
    indicating whether any output did in fact occur since the call to
    outhide(true); this allows you to determine if any output would have
    occurred, even though the output is not seen by the player.  Note
    that this is effectively the same mechanism used by the player command
    parser for noun disambiguation using the verDoXxx and verIoXxx
    methods, as described in the TADS author's manual.  There is no way
    to recover the text that was suppressed by outhide(); the text is
    simply discarded, so the only information available is whether any
    text was generated.

2.0.8  12/03/92  (tc/tdb 2.0.7, tr 2.0.8) MS-DOS bug fixes and minor changes

  - The display initialization code was incorrectly attempting to clear
    a zero-line region of the display.  This resulted in extremely long
    delays on some computers (due to an incorrect BIOS call made by TADS).

  - NEW FEATURE:  When the keyword $$ABEND is typed as the entire command,
    the run-time immediately terminates and returns to DOS.  This emergency
    escape is provided so that TR can be terminated if the game somehow
    gets into a state where a more graceful exit is not possible.

  - The compiler did properly detect when an undefined object was used
    as the superclass of another object.  This generally resulted in
    unpredictable behavior during execution of preinit.

  - NEW FEATURE:  The parser now calls two new optional methods in the
    game program.  These new methods are intended to help speed up the
    identification of words when many objects have the same vocabulary.
    If the new methods are not present, behavior is the same as before,
    so existing games will run unchanged.  The new methods are validDoList
    and validIoList; they are associated with the "deepverb" object for
    the current command.  They are called with three parameters:  the actor,
    the prep, and the other object (indirect object for validDoList and
    direct object for validIoList; the value of the parameter will be nil
    if not applicable).  These methods are called prior to the disambiguation
    pass (using verDoXxx/verIoXxx), and prior to testing any objects with
    validDo/validIo.

    The return value of validDoList/validIoList is a list of all of the
    valid objects for the verb.  It is fine for these methods to return
    *too many* objects, since each object is still tested with validDo
    (or validIo) and the appropriate verDoXxx/verIoXxx methods.  Generally,
    these methods should simply return a list of all of the accessible
    objects in the actor's current location (or the actor's location's
    location), plus a list of all of the "floating" objects (which use
    methods for the location properties).

    An appropriate definition for validDoList in the deepverb object
    appears below:

       validDoList(actor, prep, iobj) =
       {
           local ret;
          local loc;

       loc := actor.location;
       while (loc.location) loc := loc.location;
       ret := visibleList(actor) + visibleList(loc)
              + global.floatingList;
       return(ret);
       }

    This same definition (with the name changed) is appropriate
    for validIoList in deepverb.  This returns a list of all of the
    objects visible in the current location, plus the global list of
    all floating objects; this should be a superset of the list of
    accessible objects in most games.  The only verbs that normally
    requires a different value of validIoList/validDoList are verbs
    such as "ask" and "tell" that allow any object (whether accessible
    or not) to be used as indirect objects; for these, simply use this
    definition:

       validIoList = nil

    This takes advantage of the reverse compatibility feature:  when the
    method returns nil, all objects with matching vocabulary are used.

    The one additional game change required to take advantage of this
    new feature is that global.floatingList must be built during
    initialization.  This can be done easily with the following loop:

       global.floatingList := [];
       for (o := firstobj(floatingItem) ; o ; o := nextobj(o, floatingItem))
          global.floatingList += o;

    This should be placed in the preinit or init function.  Note that
    all objects which have location methods should be declared to be
    of class floatingItem:

      class floatingItem: object;

    This class doesn't do anything except serve as a flag that an
    object should be placed in the floatingList.

2.0.7  12/01/92  MS-DOS bug fix release

  - The run-time occasionally missed the \ in an escape sequence in
    output strings.

  - The run-time couldn't operate in 132 column mode.

  - The run-time abnormally terminated in the file selection dialog
    when the dialog box was exactly filled.

2.0.6  11/24/92  first general MS-DOS TADS 2.0 release