inform7/resources/Changes/Change Logs/6M62.txt

Inform build 6M62 (14 December 2015)

Contents:
	1. Introduction
	2. Minor new features in the language
	3. The Inform application
		3 (a). Inform for Mac OS X
		3 (b). Inform for Windows
		3 (c). Inform for Linux
		3 (d). Inform for Command-Line Linux
		3 (e). Inform for Android
	4. Maintenance
		4 (a). Core Inform maintenance
		4 (b). Extension maintenance
		4 (c). Examples maintenance
		4 (d). Interpreter maintenance

Note: Throughout this document, bug numbers in the seven-digit form "0000123"
refer to tickets at the Inform bug tracker: http://inform7.com/mantis/.
Suggestion numbers, also large but not starting with a 0, refer to tickets at
Inform's Uservoice forum, a public venue for suggestions to be made about
future features: http://inform7.uservoice.com/.


1. Introduction

This is for the most part a maintenance release of Inform, incorporating fixes
for nearly 300 issues reported with the previous build. All bugs reported
before 3 September 2015 (and most later ones) have been removed. Fans of bugs
need not despair, however, as the code has internally been much rewritten, so
new ones will doubtless appear.

A modest selection of minor improvements to the underlying language can be
found in §2, but little is fundamentally altered.

The Mac OS X user interface application sees two new developments in this
release. One is the new concept of an "extension project", intended for writers
of Inform extensions, which manages the extension source text better, and can
automatically test the extension's examples of usage, which should make
maintenance easier. The other development is that the previous Skein and
Transcript panels have been merged together into one, simply called Testing,
with a much clearer and simpler user interface. As in the past, some of
these developments are likely to appear on other platforms in due course,
but users should expect this to take a little time.

Many users gave up their time to report bugs in 2014/15, and we're grateful to
all of them, but a particular tip of the hat goes to Daniel Fremont, who not
only tested many obscure corners of the language but also combed through the
documentation for out-of-date usages. It is thanks to Daniel that we learned
that repeating the sentence "Foo is a room." 117030 times would crash the
compiler, and that the documentation made what he called "a 300 sextillion kg
error in the mass of Jupiter" (in our defence, that makes less difference
than you might expect); and, at Daniel's urging, we did indeed "pick a
consistent spelling for 'discombobulate'".


2. Minor new features in the language

* Usage of the "stored action" kind has been streamlined considerably. For
one thing, it can now be called simply "action". As when "indexed text" and
"text" became the same thing, so too "action" and "stored action" are now
synonymous. Constant actions no longer need to be written using the
"action of" phrase, so for example these usages are now legal:

	The best idea yet is an action that varies.

	When play begins:
		now the best idea yet is pushing the button.

Previously one had to call it a stored action that varies, and one had to
say "now the best idea yet is the action of pushing the button". The
"action of..." phrase still exists, to keep old source text working, but
it now has the following comically redundant definition in the Standard
Rules:

	To decide what action is the action of (A - action): (- {A} -).

* In past builds, switch-like ifs were valid only for simpler kinds of value
such as numbers; they can now switch on any value, and in particular on texts.
For example:

	if the new password text is:
		-- "xyzzy": say "Nothing happens.";
		-- "password": say "Would you care to accept a position at the NSA?";
		-- otherwise: say "Accepted."

* Two minor improvements to the handling of named constants, which are
relatively new in the language. 

(a) The number of blank rows in a table can now be specified using a named
constant, like so:

	The spare capacity is always 10.

	Table of Stuff
	whatever (a number)
	with spare capacity blank rows

The named constant must, of course, hold a non-negative integer. Note that
it is allowed to be 0, which adds no blank rows. Inform doesn't allow
completely empty tables, so if, as in this case, there are no explicit rows
and the name comes out to 0 as well, Inform quietly gives the table a single
row of blanks, but does compile it. (See bug report 0001724.)

(b) Map connections can now be made to or from rooms referred to by named
constants, not explicitly as usual. Thus:

	The Lounge is a room.
	The designated room is always the Lounge.
	The Exit Area is east of the designated room.

places the Exit Area east of the Lounge.

* When one action is converted to another one inside the Standard Rules, as
most often happens when "removing X from Y" is converted to "taking X",
and the action then succeeds, the after rulebook for the original action
is now followed. For example,

	Some coins and some bills are in the tip jar.

	After removing something from the tip jar:
		say "Alice is shocked at your bold impropriety."

will now work as might be expected from its wording. (This was filed as
bug report 0001476, but it was really a feature request.)

* A new Understand token "[any things]" is a sort of cross-breed between
"[any ...]" and "[things]", in that it behaves like "[things]" (accepting
a multiple object) but with no restriction on scope, that is, applying to
anything however distant or invisible, as with "any". (This was suggestion
751340.)

* A new text substitution for adaptive text:

	[regarding N]

makes it easier to write responses such as the parser error internal rule (Q):

	"[if number understood is 0]None[otherwise]Only [number understood in
	words][end if] of those [regarding the number understood][are] available."

which inflects the verb to agree with the number.

* A subtle change to the rules on kind inference: if a temporary value X
is created with "let", in the shape

	let X be V;

then Inform gives X whatever kind V has - for example, "let X be 10" will
make X a number, and so on. Traditionally, though, if V is any kind of
object, X has the kind "object", not something more specific. This is so
that sensible-looking source text like this:

	let X be Peter;
	now X is Jane;

...won't cause a problem message. If X is given the kind "man", the kind
of Peter, then the "now" can't be allowed because Jane has kind "woman".
In previous builds, then, Inform gave X the kind "object".

The change is that it now gives X the right top-level kind of object,
instead of "object" itself. In the case of "let X be Peter", X will now
be given the kind "thing", because Peter has the kind "man", but man is
a kind of person, which is a kind of thing, and thing is a kind of object.
More generally, where X would once have been an "object", it will now be
one of "thing", "room", "region" or "direction". This removes a number of
small annoyances (see for example bug report 0001501).

* Most phrase definitions are not case sensitive, but an exception is made
for text substitutions where there are different definitions depending
on whether the first word is capitalised or not. This enables "[the X]"
and "[The X]" to have different meanings. In the new build, case sensitivity
has been extended to cover all text substitutions whose definition is a
single letter. This means that "[S]" will no longer be misread as if it
were "[s]", the notation for an optional "s"; so source text such as

	let S be 10;
	say "S is [S].";

will no longer peculiarly say "S is s." or "S is .", but will say "S is 10."
(Prompted by bug report 0001492.)

* In past builds, the adjective "empty" couldn't usefully be applied to
relations created in sentences in the usual way; for example, given

	Noticing relates various people to one person.

the condition "the noticing relation is empty" and the instruction "now the
noticing relation is empty" weren't allowed. They now are. (Prompted by
bug report 0001416.)

* A new rulebook has been added:

	multiple action processing rules

This is called when a command would produce a multiple action, such as
TAKE ALL. Rules in it can then take the opportunity to look at the list of
objects to which the action would apply, and can change or rearrange this.

* "Understand" can now apply to descriptions of values. For example,

	Understand "glowing" as something lit.

now works; previously this could only be done with the circumlocution

	Understand "glowing" as a thing when the item described is lit.

(Prompted by bug report 0001796.)

* The following rules have been added to the Standard Rules:

(a) The "describe what's on mentioned supporters in room descriptions rule"
improves the look of room descriptions when a non-scenery, non-undescribed
supporter is holding up one item that has an initial appearance and another
that hasn't, and is for some reason mentioned in that first item's initial
appearance; this parallels the existing handling of scenery supporters.

(b) The "exclude indirect possessions from take all rule" means that
TAKE ALL won't try to take items which are inside open containers already
being carried, and so forth. (See bug report 0001474.)

* The following existing rules in the Standard Rules have changed:

(a) The "standard taking rule" has been extended to add:

	if the actor is the player, now the noun is handled.

This ensures that even items only fleetingly carried are "handled"; in
previous builds, a successful taking action might not result in an object
being "handled" if it were then lost again before the end of the same turn.
(This can happens as a result of implicit takes. See bug report 0001460.)

(b) The "immediately undo rule" has a new response (F) which reports that
the interpreter's UNDO capacity has been exhausted. (See bug 0001689.)

(c) The following existing responses in existing rules have been made
adaptive (see bug report 0001715). Old and new definitions are as follows -

	basic visibility rule response (A)
		"It is pitch dark, and you can't see a thing."
		"[It] [are] pitch dark, and [we] [can't see] a thing."

	block climbing rule response (A)
		"I don't think much is to be achieved by that."
		"Little [are] to be achieved by that."

	parser nothing error internal rule response (B)
		"There are none at all available!"
		"[There] [adapt the verb are from the third person plural] none
		at all available!"

(d) The "announce items from multiple object lists rule" now has a new
response (A), which is responsible for the traditional format of the text
dividing up replies relating to different objects (see bug report 0001762).
This has the value

	"[current item from the multiple object list]: [run paragraph on]"

and is responsible for e.g. "yellow marble: " and "red marble: " in -

	yellow marble: Taken.
	red marble: Taken.

Being a response, it's now customisable. For example,

	The announce items from multiple object lists rule response (A) is "[current 
action]: [run paragraph on]".

leads to

	taking the yellow marble: Taken.
	taking the red marble: Taken.

* The implementation of integer square root and cube root has been improved,
so that Glulx's floating-point operations are used if available. This is
faster and normally more accurate. (See bug report 0001547.)

* When releasing with an existing story file, this file can now be named:

	Release along with an existing story file called "Zork1_sg.z5".

(Previously it had to be "Story.z8".)

* A clarification: a list is sayable if and only if its contents are sayable.
In previous builds lists were always sayable, but would produce meaningless
output if the contents weren't sayable (for example, for a list of topics).

* A clarification: creating kinds of direction is not allowed, and it now
produces a problem message rather than failing more obscurely. (See bug
0001588.)

* A clarification: in past builds the documentation said that literal patterns
could contain up to 4 parts; internally, the limit was 16, but would not
always work correctly for high values; the limit is now set at 8.

* The "phrase" token works differently. This was undocumented and unsupported,
and was to have been removed entirely in this build; but the extension
"Hypothetical Questions by Jesse McGrew" turned out to use it regardless
of its unofficial status, and the popular "Consolidated Multiple Actions
by John Clemens" made use of this in turn. We have therefore retained it,
a little reluctantly. Even so, its use has been somewhat restricted: it
can't accept control constructs such as if, while, repeat, otherwise, now,
or say. On the other hand it expands in a more predictable fashion, without
generating braces which are problematic to deal with. It nevertheless
remains an unsupported feature of Inform, and there's no guarantee it
won't change.


3. The Inform application

3 (a). Inform for Mac OS X

Firstly, the Mac OS X application has made the transition to 64 bit only. This
change only adversely affects users who are still running OS X 10.6.8 on 32
bit hardware. Users of Inform on OS X 10.7 and above are still fully
compatible (because by definition they are already running on 64 bit
hardware). Similarly, users still running OS X 10.6.8 on 64 bit hardware
should remain compatible. This change is made to help improve stability and
smooth future development.

New to this version is the Testing panel (which replaces the old Skein and
Transcript panels of previous versions). The Testing panel is designed to help
ensure the story consistently responds as expected while the development of
the source continues. The Testing panel consists of ‘knots’ - all commands
that have on previous occasions been played in the story with their associated
response text. Knots are displayed in a branching fashion - knots lower down
the panel are later in the story than those nearer the top. Each knot’s
response text can be ticked (‘blessed’) to signify that it is correct (i.e. as
expected). Any deviation from the blessed text in future plays of the story
will be shown on the knot with a red star, and the differences highlighted in
the response text itself. Each knot has a menu of options. Knots can be
dragged and dropped onto other knots to be copied or moved, even onto other
projects. Undo is supported. A Help section of the Testing panel describes
each element. The Play All button allows all knots to be automatically played
in the story.

Also new to this version are Extension Projects. Extension Projects are
designed to help developers of Inform extensions. An Extension Project can be
created from an existing extension or from scratch. The extension’s source can
be edited, its ‘examples of use' played and tested individually, or all its
examples tested automatically. The extension can then be installed or exported
as required.

A smaller improvement is that the story is now only recompiled on ‘Go!' if
needed, for example because its source text has changed.

Bug fixed (0000089): 'More...' overlay is not properly hidden.

Bug fixed (0000378): Bug with CocoaGlk and the Automap extension.

Bug fixed (0000538): Problems with Skein and Transcript panels - now rewritten
and replaced with the new Testing panel.

Bug fixed (0000545): Problems with Skein and Transcript panels - now rewritten
and replaced with the new Testing panel.

Bug fixed (0001076): Invisible characters in source text editor.

Bug fixed (0001093): glk_buffer_to_title_case_uni crashes in Mac IDE
interpreter.

Bug fixed (0001111): CocoaGlk: The built-in interpreter becomes unresponsive
after RESTART.

Bug fixed (0001189): CocoaGlk crash when measuring an unhinted style.

Bug fixed (0001404): Crash with VoiceOver enabled.

Bug fixed (0001437): Issues with 10.6.8 resolved.

Bug fixed (0001481): Issues with 10.6.8 resolved.

Bug fixed (0001641): Double-clicking the "Go!" button causes an I6 error.

3 (b). Inform for Windows

The progress bar shown when compiling a story (or downloading extensions) is
now rather more prominent, and shown in the middle of the window.

The toolbar icons have been updated to match the OS X front end. This has also
involved removing the "Stop" button from the main toolbar. Instead this can
now be found as a button on the "Story" tab.

Two bugs have been fixed: 0001418 and 0001511.

3 (c). Inform for Linux

The GUI code is no longer hosted on SourceForge. The place to go is now GitHub
(https://github.com/ptomato/gnome-inform7). The location for official
downloads — the Inform website — has not changed.

Inform on Linux can be used in with a GUI in French and Spanish, following a
bug fix. Those wishing to help translate the GUI into other languages can now
visit Transifex (https://www.transifex.com/ptomato/gnome-inform7/).

* Packaging issues fixed; thanks to Vincent Petry and Bart Massey for
contributing these fixes.

* Trying to open a nonexistent story from the recently-opened list will now
remove that story from the list (0001732).

* Bug fixed (0001612) whereby the Public Library had stopped working.

* Bug fixed (0001446) whereby extensions marked "for Glulx only" were
installed incorrectly when downloaded from the Public Library.

* Bug fixed (0001497) whereby code pasted into the story caused extra
indentation to be inserted.

* Bug fixed (0001757) whereby some Inform 6 code was highlighted incorrectly.

* The correct version of Glulx Text Effects is included with this release
(0001447).

* A few bugs fixed that caused crashes in certain circumstances: 0001286,
0001406.

3 (d). Inform for Command-Line Linux

3 (e). Inform for Android


4. Maintenance

4 (a). Core Inform maintenance

As ever, a number of cosmetic bugs have been fixed, which aren't worth
itemising here in any detail: 0001287, 0001398, 0001400, 0001412, 0001442,
0001445, 0001453, 0001454, 0001455, 0001465, 0001466, 0001471, 0001477,
0001478, 0001479, 0001480, 0001504, 0001535, 0001543, 0001544, 0001546,
0001551, 0001554, 0001556, 0001557, 0001579, 0001581, 0001598, 0001605,
0001608, 0001615, 0001616, 0001621, 0001627, 0001628, 0001632, 0001638,
0001640, 0001647, 0001650, 0001655, 0001656, 0001659, 0001668, 0001669,
0001677, 0001678, 0001697, 0001718, 0001726, 0001727, 0001728, 0001729,
0001737, 0001738, 0001739, 0001740, 0001743, 0001758, 0001769, 0001771,
0001773, 0001774, 0001780, 0001783, 0001786, 0001790, 0001795, 0001798, 0001800.

4 (a) 1. Source text and punctuation

Bug fixed (0001520) whereby a source text opening with a colon would throw
an internal error rather than a problem message.

Problem messages added (see 0001521) to catch various punctuation accidents:
colons at ends of paragraphs or sentences, and semicolons at ends of sentences
or following colons or full stops.

Bugs fixed (0001523, 0001529) whereby instances and rules whose names included
brackets would sometimes throw internal errors rather than problem messages.

Bug fixed (0001797) whereby a plural consisting entirely of a bracketed text,
such as "The plural of wug is (nothing).", would throw an internal error rather
than a problem message.

4 (a) 2. Headings

4 (a) 3. Extensions

Bug fixed (0001537) whereby overlong extension or author names in Include
sentences would cause Inform to issue a problem message (correctly) and then
end in exit code 2 or 11 rather than 1.

4 (a) 4. Assertions and creations

Bug fixed (0001610, 0001617) whereby source texts making very repeated use of
creation sentences would crash (e.g., at least 117030 copies of the sentence
"Foo is a room.").

Bug fixed (0001588) whereby attempts to create nameless instances of subkinds
of "direction" would crash Inform by repeating the same problem message
until the stack ran out.

Bug fixed (0001600) whereby creating objects with very long names would lead
to unexpected problem messages about quotation marks.

Bug fixed (0001571) whereby constants left undefined would sometimes lead
to I6 errors rather than problem messages.

Bug fixed (0001572) whereby constants defined indirectly in terms of themselves
would throw an internal error rather than producing a clear problem message.

Bug fixed (0001539) whereby names containing long I6 inclusions (which is
a very odd thing to try to do) would sometimes lead to internal errors.

Bug fixed (0001411, 0001532, 0001686) whereby a sentence using a relative
clause to equate things would in some cases (e.g. "Zog is a person that is
false.") produce an internal error.

Bug fixed (0001643) whereby changing the meaning of the verb "to be", not
perhaps the wisest course in any event, would fail to produce the problem
message for inconsistent verb definitions.

Bug fixed (0001558) whereby "There is an object." was rejected, rather than
doing something not very useful (creating a thing called "thing" and placing
it nowhere) as it should.

Bug fixed (0001587) whereby assemblies which create qualified common nouns
would sometimes produce a bogus problem message about circular kind
definitions.

Bug fixed (0001662) whereby adjective definitions using object names in their
domains could under some circumstances produce a "cannot establish position P"
internal error.

Bug fixed (0001667) whereby a sentence creating something "(called the)" or
"(called a)" would produce an internal error not a problem message.

Bug fixed (0001672) whereby a nonsensical assertion like "Edible has bar"
could crash Inform rather than produce a problem message.

Bug fixed (0001675) whereby a redeclaration of, say, a scene as a variable
could sometimes cause an internal error not a problem message.

Bug fixed (0001658) whereby assertions about map directions, which are
vague about the origin, could throw an internal error if the subject of
discussion had changed to something which wasn't an object (e.g. by
creating a new kind).

Bug fixed (0001693) whereby assertions about map directions which give
the destination as a value other than an object would (usually) be ignored
rather than throwing a problem message.

Bug fixed (0001683) whereby the map might come out with confused connections
if directions were declared in an unusual order, mixed in with sentences
making map connections using them.

Bug fixed (0001691) whereby some sentences creating things would be allowed
even though they were said to be never or seldom true: for example, "In the
Dining Room is never a door."

Bug fixed (0001580) whereby bare existential sentences with multiple objects,
such as "There are X and Y", giving no relationships between X, Y and anything
else, would result in nothing being created.

Bug fixed (0001593) whereby implications whose premiss is a negated adjective
would sometimes have their sense reversed: e.g. "Something not lit is usually
edible" would be read as if it said "Something lit is usually edible".

Bug fixed (0001685) whereby assertions in the form "Home contains X with P.",
where P is a property, would sometimes have P ignored.

Bug fixed (0001456) whereby adjectives could be defined over topics, but with
unfortunate consequences. There's now a problem to defend against this.

4 (a) 5. Model world

Bug fixed (0001560) whereby creating 10000 or more objects in a single story
would result in a "far too many relatees" error. There is now no upper limit,
except of course the boundary between sanity and madness.

Bug fixed (0001665) whereby creating infinitely large assemblies of things
connected to other things would pause for a very long time after issuing the
problem message. Technically not a hang, but not a good use of our limited
time on this Earth either.

Bug fixed (0001670) whereby implicit contradictions in one-to-various or
various-to-one relationships caused by assembly might not result in problem
messages being issued.

Bug fixed (0001671) whereby recursive assemblies of non-object instances
would sometimes fail to recurse as they should, thus halting early. (In
practice this would mean that circular assemblies of this kind wouldn't
result in problem messages.)

Bug fixed - or really, convention changed (0001586) - so that if a room
is made using "X is D of Y", where D is a direction with no reverse, then
X is assumed likely to be a room in the absence of other evidence. Previously
this happened only if D had a reverse. (This was because the convention was
that assumptions were made only about the destination Y: when D had a
reverse direction, un-D, the above sentence implied also "Y is un-D of X",
and so X was also a destination.)

Bug fixed (0001639) whereby timed events occurring close to midnight would
sometimes never arrive if the clock had been altered to jump from before
them to after midnight between turns, e.g., by replacing the advance time
rule.

Bug fixed (0001597) whereby putting a region initially inside a room was
being allowed - it should of course produce a problem message, and now does.

A problem message has been added (see 0001687) to catch the inevitably
disastrous consequences of making "container" a kind of "supporter" or vice
versa; similarly (0001692) for making "door" a kind of "backdrop" or vice
versa.

Bug fixed (0001607) whereby a proper-named person would, if his/her printed
name were changed at run-time, sometimes be given the capitalised form of
the original name in some responses.

Bug fixed (0001603) whereby an item marked "undescribed" would nevertheless
be described if it both provided an initial appearance text and was found
on a supporter. (See the "initial appearance on supporters rule".)

Bug fixed (0001493) whereby an item hidden through the use of the "deciding
the concealed possessions of" activity would nevertheless be mentioned
in the bracketed part of a list of objects, as in, "You can see an aquarium
(in which are some pebbles and a goldfish) here.": in previous builds the
pebbles would have been included even if the activity had said otherwise.

Run-time problem added (0001463) for testing the adjacency relation with
regions rather than rooms. (Previously this simply gave unhelpful answers.)

4 (a) 6. Properties

Bug fixed (0001528) whereby "of" used with an either-or property would sometimes
be rejected with internal errors rather than problem messages.

Bug fixed (0001538) whereby adjective names containing brackets or quotes were
rejected with internal errors rather than problem messages.

Bug fixed (0001524) whereby properties with very long names might crash Inform
on declaration.

Bug fixed (0001540) whereby absurdly long property names caused internal
errors not problems.

Bug fixed (0001772) whereby attempts to assign by something like "now V is P",
where V is a variable of kind K and P is a property holding K, now produce
problem messages rather than failing to compile through I6.

Bug fixed (0001564) whereby a verb which means a property would cause
run-time problems when tested or asserted with "now", if the owner of the
property was something other than an object.

Bug fixed (0001530) whereby a nonsensical property value could be given in
a property list and it would be ignored rather than reported.

Bugs fixed (0001530, 0001657) whereby a property list giving a value which
involved "with" might lead to an internal error.

Bug fixed (0001620) whereby giving a nameless property to a single instance of
a non-object kind such as "scene" or "narrative viewpoint" would lead to an
internal error. (It's legal to do this, and now works.)

Bug fixed (0001673) whereby giving a condition to something whose name
contained the word "from" could lead to an internal error.

Bug fixed (0001652) whereby a property list attached to something which can't
have properties could in some cases lead to an internal error not a problem
message.

4 (a) 7. Relations

Bug fixed (0001540) whereby absurdly long relation names caused internal
errors not problems.

Bug fixed (0001552) whereby some forms of implicit reference to
relationships such as "equality" could cause an internal error.

Bug fixed (0001676) whereby trying to use a relation's bare name as a
description of things in a list would lead to an internal error rather
than a problem message.

Bug fixed (0001583) whereby typechecking of the universal relation would
sometimes give unexpected results when applied to other unusual relations
(such as provision).

Bug fixed (0001583) whereby the provision relation can now be tested a little
more freely with properties identified by variable.

Bug fixed (0001550) whereby some indirect uses of some built-in relations
could produce meaningless results; they now produce a run-time problem.

Bug fixed (0001553) whereby various-to-one relations cannot always relate 0
to something, or other values internally stored as 0 at run-time.

4 (a) 8. Actions

Bug fixed (0001525) whereby deficiencies in action descriptions assigned to
named behaviours would sometimes be reported at the wrong line of the source.

Bug fixed (0001651) whereby a description of actions consisting only of an
article would throw an internal error not a problem message.

Bug fixed (0001660) whereby certain very malformed descriptions of actions
would throw an internal error not a problem message.

Bug fixed (0001496) whereby actions written as constants, for example in
entries in a table, and which contained arbitrary text topics, would
sometimes have those text topics lost when tried at run-time.

Bug fixed (0001469) whereby Inform would sometimes wrongly decide which form
of asking was intended (asking for, or about?) in a phrase like "asking or
telling Bob about...".

Bug fixed (0001789) whereby Inform would guess past participles of action
names wrongly in the case of present participles like "playing", "slaying",
"preying" or "toying", where the vowel A, E or I occurs before YING. These
now correctly produce "played", "slayed", "preyed" and "toyed".

4 (a) 9. Activities

--

4 (a) 10. Rules and rulebooks

Bug fixed (0001527) whereby rulebook variables declared with nonsensical kinds
would produce an internal error not a problem message.

Bug fixed (0001629) whereby a rule with "in... while..." clauses might lead
to an internal error not a problem message if not understood.

Bug fixed (0001666) whereby a timed event rule whose name consisted only of
an article such as "the" or "a" would cause an internal error.

Bug fixed (0001591) whereby a rule with a "during..." clause describing the
scenes to which it applied would not have this scene description properly
checked, so that nonsensical clauses like "during three scenes" or "during
a scene which is in the paper bag" would be accepted but then either throw
I6 errors or fail at run-time.

Bug fixed (0001541) whereby rulebooks could be declared with indefinite
kinds such as "value based rulebook". There's now a problem message.

Bug fixed (0001642) whereby rules would sometimes be allowed to "substitute
for" other rules with incompatible kinds.

Bug fixed (0001470) whereby rules based on actions whose actor's name
included a participle (say, "tough-looking coach") would sometimes be misread.

Bug fixed (0001508) whereby the syntax for declaring first/last rules for
rulebooks with names beginning with "for" could be slightly erratic.

Bug fixed (0001748) whereby rulebooks couldn't produce something referred
to using an indefinite article, e.g., "an object-based rulebook producing
a list of texts" was being refused because of the "a".

4 (a) 11. Lists

Bug fixed (0001618, 0001619) whereby a long list of constant action names
could slow Inform's parser down to a crawl, or in extreme cases cause Inform
to halt as having allocated too much memory.

Bug fixed (0001723) whereby an attempt at run-time to resize a list to a
negative size would print a brief error into the transcript, but not throw
a formal run-time problem message.

Bug fixed (0001503) whereby, under unusual circumstances, it's possible for
removal of an item from a list to lead, some time later, to memory
allocation or other heap-related errors.

Bug fixed (0001741) whereby lists of constant stored actions would sometimes
not be allowed, with the actions being rejected as non-constant.

4 (a) 12. Tables

Bug fixed (0001595) whereby including a large number of tables, with some
of them nameless, would sometimes cause I6 errors due to I7 accidentally
using the same I6 identifier twice.

Bug fixed (0001464) whereby a table entry holding an action which used a
variable parameter somewhere would be accepted by I7 but then lead to I6
errors; it now produces a problem message.

Bug fixed (0001519) whereby trying to create a value defined by a table
outside of that table would throw a problem message if it happened before
the table in the source, but not if it happened after.

Bug fixed (0001542) whereby amending a table used to define things would
sometimes produce the amended property values, sometimes the old ones.
This is now ruled illegal, and there's a problem message accordingly.

Bug fixed (0001664) whereby a failed table amendment might produce an
internal error not a problem.

Bug fixed (0001653, 0001654) whereby a duplicated column name in a multiple
table continuation might produce an apparent hang not a problem.

Bug fixed (0001555) whereby sorting a table in order of a column holding
real numbers would sort in order of increasing positive reals, then of
decreasing negative ones, rather than in the obvious numerical order.

Bug fixed (0001706) whereby names of phrases used as constants in table
cells would sometimes result in the kind of the table column being inferred
too vaguely (as "phrase value -> value"), rather than using their actual
kinds, so that it would then be difficult to use the column for anything
without throwing kind-checking problems.

4 (a) 13. Scenes

Bug fixed (0001711) whereby a sentence like "Scene A ends when Scene B ends
interestingly" occurring before the declaration of Scene B's interesting end
would throw a problem message saying the ending in question was unknown.

4 (a) 14. Kinds and typechecking

Bug fixed (0001601) whereby a kind with a very long name would cause a
crash once either a notation or some instances were created for it.

Problem messages improved (0001575) in being clearer about names not
yet defined.

Bug fixed (0001509, 0001778) to do with matching a list of K against a list
of L when L is a kind of K.

Bug fixed (0001495, 0001425, 0001448) whereby Inform would allow a text value
to be assigned to a topic variable, but the result would be an I6 error about
a nonexistent routine called TEXT_TY_to_UNDERSTANDING_TY, or in some cases
a Glulx virtual machine error. Such assignments aren't legal, and there's now
a problem message for it.

Bug fixed (0001648) whereby long expressions of what look arithmetic but
don't in fact contain meaningful terms would cause Inform to run very slowly,
or eventually run out of memory, rather than produce a problem message
quickly.

Bug fixed (0001570) whereby adjective Definitions could define over kinds
of kinds such as "sayable value", leading to problems later since this
isn't type-safe. There's now a problem message.

4 (a) 15. Phrases and functional programming

A local variable can't have the same name as a non-object kind. This was
previously only sometimes enforced, due to a bug: it's now always enforced.

Bug fixed (0001526) whereby a phrase token containing a quantifier, such
as "N - three numbers", would produce an internal error not a problem message.

Bug fixed (0001576) whereby a phrase with more than 10 tokens in its
definition could crash the compiler.

Bug fixed (0001567) whereby the functions with mathematical names (such as
"abs function") could not be used as constants, because a mysterious problem
message would appear claiming that their definitions in the Standard Rules
were unrecognised.

Bug fixed (0001613) to do with generic programming and the interpretation
of kind variables when a phrase is used as a value which itself involves
those variables: see the bug report for more.

Bug fixed (0001533) whereby Inform would fail to recover after a problem
message involving a function kind declared with codomain equal to an
undeclared kind variable, and would issue an internal error as well.

Bug fixed (0001674) whereby an implicitly circular adjective definition
such as "Definition: a big number is big if it is 123." would crash
Inform rather than produce a problem message.

Bug fixed (0001645) whereby duplicated cases in an "if ... is: --" switch
would be allowed without warning. (They now cause a problem message,
since they can never be useful.)

Bug fixed (0001646) whereby named constants couldn't be used as cases
in an "if ... is: --" switch.

Bug fixed (0001583) whereby conditions using the present perfect or past
perfect forms of some verbs would not be understood - for example, "X had
provided Y".

Bug fixed (0001514) whereby "value", or other kinds of kinds, could be
accepted where definite kinds were expected, resulting in undefined
behaviour at run-time.

4 (a) 16. Equations, units and arithmetic

Bug fixed (0001417) whereby arithmetic producing a dimensionless result
after calculating with dimensioned operands would sometimes wrongly use
integer rather than real arithmetic, so that, for example, 1500m divided
by 10m would print out as 1.12552 × 10^9 not 150.0m.

Bug fixed (0001569) whereby Inform would sometimes crash after issuing a
problem message in the event of dimensionally impossible arithmetic in the
optional parts of an action description.

Bug fixed (0001625) whereby more part names could be given than there were
parts to name.

Bug fixed (0001625) whereby multiple part names might be slow to parse.

Bug fixed (0001625) whereby cumulatively over-long part names would be
rejected with a misleadingly worded problem message.

Bug fixed (0001625) whereby multiple part names could be identical. These
now produce problem messages.

Bug fixed (0001561) whereby attempting to scale units by non-positive
amounts produced an internal error not a problem message.

Bug fixed (0001589) whereby an equation defining three or more variables
as having the same kind (e.g. "where a, b, and c are numbers") might lead
to internal errors.

Bug fixed (0001562) whereby giving a notation to specify an arithmetic kind
of value, with example value as an integer but scale factor a real number,
would silently ignore the scale factor. There's now a problem message
saying how to fix this to get the intended effect.

Bug fixed (0001649) whereby an equation with no name and no number would
throw an internal error, not a problem message.

Bug fixed (0001644) whereby incrementing a variable containing an
unusual arithmetic kind of value could in some cases cause an internal
error.

Bug fixed (0001573, 0001574) whereby equations whose solution would require
fourth or higher-degree roots would produce a spurious problem message about
that power not being a constant number, or which would calculate out with
the wrong answer by mixing real and integer arithmetic.

Bug fixed (0001582) whereby equations working out I^R, where I is a number
and R a real number, would sometimes assign this to a number variable
without converting it back from real arithmetic.

4 (a) 17. Understanding

Bug fixed (0001414) whereby "[something related by R]" would, for a few
exotic relations R such as "provision", throw an internal error. There's
now a problem message.

Bug fixed (0001451) whereby AGAIN on the first turn would not produce a
sensible reply on Glulx, though it would on Z.

Bug fixed (0001500) whereby the low-level I6 routine UnknownVerb would be
called only if the actor in a command was the player. (This didn't affect
regular English-language stories, where I7 doesn't use UnknownVerb.)

Bug fixed (0001604) whereby malformed instructions for people other than
the player would sometimes result in no response at all.

Bug fixed (0001661) whereby an "[any any thing]" token, with unexpected
extra quantification, would give an internal error not a problem message.

Bug fixed (0001694) whereby the "after reading a command" rulebook was run
after some disambiguation questions are asked and answered, but not others.
It should now consistently be run.

Bug fixed (0001475) whereby the use of a command involving "[text]", followed
by THEN and another command in the same piece of typing, would cause that
text to become a row of spaces.

Bug fixed (0001568) whereby the use of the "if the player consents" phrase,
which asks for a yes/no answer, would invalidate the contents of the
"player's command" variable.

4 (a) 18. Glulx issues, file I/O, figures, sounds

The implementation of the Memcpy utility has been made faster on Glulx.

Bug fixed (0001429) to do with file input/output disrupting the Glulx output
channel, since writing to a file would fail properly to reset the channel
when finished. (Thanks to Dannii Willis for the patch to fix this.)

Bug fixed (0001506 but see also 0001502) to do with sound channels not being
properly identified after a restore or restart.

4 (a) 19. I6 template layer and inclusions

Bug fixed (0001578) whereby seemingly inconsistent combinations of use
options ("Use scoring and no scoring") could lead to I6 errors.

Bug fixed (0001631) whereby I6 inclusions of more than about 600000
characters at a time would produce internal errors rather than problems.

Bug fixed (0001759) whereby use in an I6 inclusion of "ÿ", the largely
mythical letter needed only by the band Queensrÿche and its devotees,
would result in a spurious I6 error ("MAX_QTEXT_SIZE exceeded").

Problem message added (0001535) to do with using I6 inline definitions for
named functions without I6 functions backing them up.

Problem message added (0001594) for overlong I6 definitions of adjectives
and the like.

Problem message added (0001596) for broken attempts at template hacking.

4 (a) 20. Text, spacing and printing

Bug fixed (0001397) whereby text substitutions for adapting modal verbs
required the modified verb to take the third person, not the infinitive:
which for regular English verbs made no difference, but meant that you
had to write "[might are]" instead of "[might be]", for example.

Bug fixed (0001577) whereby ridiculously nested multi-part text substitutions
could crash Inform rather than produce a problem message.

Bug fixed (0001599) whereby "say" statements including comma-separated lists
of more than 4097 terms to be printed would lead to an internal error. There's
now no limit on this: enjoy.

Bug fixed (0001637) whereby assigning a text to itself could corrupt its
value, e.g., writing "now T is T" where T holds text which Inform has had to
store in written-out form. (This could also happen with stored actions and
some other kinds of value, but was most likely seen with text.)

Bug fixed (0001663) whereby attempting to display the empty text as a boxed
quotation could crash Inform. (There's now a problem message instead.)

Bug fixed (0001606) whereby the you-can-also-see rule would, if running
in the second person perfect tense, produce "You has been able to see..."
instead of "You have been able to see..."

4 (a) 21. Indexing

Bug fixed (0001428) whereby the coloured folder links in the Extensions
panel wouldn't work - i.e., they could be clicked on, but wouldn't show the
relevant location in the file system.

4 (a) 22. Testing commands

Bug fixed (0001430) whereby "Test <case> with ..." would cause I6 errors if
the case name were given in quotes. There's now an explanatory problem.

Bug fixed (0001491) whereby the ACTIONS command's output would produce
gibberish when printing out some actions involving "[text]".

4 (a) 23. Releasing, bibliographic data, and cBlorb

Bug fixed (0001413, 0001499) whereby releasing along with a named file would
in some cases cause an internal error.

Bug fixed (0001713) whereby releasing with an image or audio file whose
filename ends in a space or full stop might, on Windows, cause its file
format (e.g. JPEG versus PNG, or AIFF versus OGG) to be wrongly guessed.

Bug fixed (0001494) whereby releasing with an existing story file wasn't
working.

4 (b). Extension maintenance

"Glulx Image Centering" updated to version 4, to maintain compatibility
with other extensions.

4 (c). Examples maintenance

Two new examples, "Western Art History 305" and "The Best Till Last",
demonstrate the use of the multiple action processing rulebook.

4 (d). Interpreter maintenance

Inform comes with browser interpreters built in, so that it can act on release
instructions such as "Release along with an interpreter.", which tell Inform
to make a website in which the current story is immediately playable.

Parchment has been updated to 2015-09-25, fixing a bug (0001426).

Quixe has been updated to 2.1.2, as compared with 1.3.1 in the last two
releases of Inform. This is quite a large improvement, since it takes in
four successive releases of Quixe:

- 2.1.2 (November 22, 2015)
  - Added basic WAI-ARIA support to buffer windows.
  - Fixed a bug where setting a graphics window's color and then clearing
    it (in the same turn) would fail.

- 2.1.1 (June 13, 2015)
  - Restructured generated JS code for better optimization in modern
    browsers. (No more eval() calls!) Thanks to Alex Munroe for pointing
    out the problem and offering solutions. Also everyone else who
    joined in the JS-wonkery discussion.
  - Use Math.imul instead of native multiplication, so that large integer
    multiplies (which overflow 32 bits) are computed correctly.

- 2.1.0 (April 24, 2015)
  - Include GlkOte 2.1.0 (graphics windows; image display; mouse input).
    Thanks to Alex Munroe for original implementation.
  - Remove the #layouttestpane from all the HTML templates.
  - Added the ability to send transcript data to an external server.
  - Fixed a bug where non-ASCII characters in metadata were not properly
    decoded. (This could affect the <title> of the browser window.)

- 2.0.0 (February 12, 2015)
  - Switched from Prototype over to jQuery.
  - Switched from my old ad-hoc license to the MIT license.
  - Added the ability to download a saved-game file.
  - Increased the font size in the included CSS stylesheets.
  - Fixed save bug where the IFhd chunk could appear late in the file
    (violating the Quetzal spec). Also, a bug where odd-length chunks
    were not padded (violating the IFF spec).