CHANGES.txt

[Since 1.06]
1 June 2012

1. David Turner made these following changes, and is solely
responsible for any errors.

2. Support for unicode (glulx 3.0.0).  Astral plane support
is a bit spotty still.

3. Support for heap memory allocation and block memory operations
(glulx 3.1.0)

4. Support for accelfunc and accelparam (glulx 3.1.1)

5. Misc bug fixes in search, bit shifting, undoing past terp restarts,
filtering while streaming numbers, copyb/copys.

6. Floating point support (glulx 3.1.2)

glulxercise.ulx now passes.

[Since 1.05]
18 March 2005

1. AIFF sound no longer uses the Java Sound Clip interface, since
we've run into AIFFs that are too long for the Clip's buffer.  We're
now playing AIFFs as streams, which we should have been doing from the
start.

2. Changes to Micromod's JavaSoundOutput class, so that upon close, we
flush the buffer first.  This eliminates some delay on the Macintosh.

3. Similar change to (2) but for AIFFs in SoundChannel.java.


[Since 1.04]
20 March 2004

1. Made the memory protection range work properly...I think.

2. TextGridWindow.clear() now resets x,y = 0, as per spec.  This fixes
a problem Zag had with the status window in "Pytho's Mask."


7 June 2004

1. Fixed the problem where a graphic artifact appears on the screen when 
you first begin to type at a cursor.

2. Optimized the drawing when you simply move the cursor along a line of 
text (by, e.g., using the arrow keys).  Rather than redrawing the screen,
we now only redraw the text.


8 June 2004

1. Changed how Style.java checks to see if a font is monospaced.  It
wasn't working on Linux beforehand.

2. Slight drawing optimizations in TextGridWindow, GraphicsWindow, and
TextBufferWindow.

3. Fixed the drawing of the cursor in TextGridWindows (when line input is
active), which I broke in an earlier release.

4. Gave the TextGridWindow better margins.  Unfortunately, this causes
Lock & Key's grid window to be too small to contain its text, using the 
default 14pt font (at least, on my Mac).  (Oh well:  In l-k-notes.txt, 
Adam suggests changing to a small font and using the REFRESH verb,
which works fine.)


9 June 2004

1. Made Zag take Winglulxe .cfg files more seriously.  Previously,
.cfg settings would override Zag's defaults, but they would not
override saved user preferences.  In other words, any user who had
changed the default styles at all would never see the font faces or
sizes specified by the .cfg file.  Now, these .cfg settings override
the user's settings.  [This is a questionable practice...]

2. Turned on underlining for hyperlinks.  Using
UNDERLINE_LOW_ONE_PIXEL, which looks infinitely better than
UNDERLINE_ON.

3. Made hyperlinks keep their distinctive appearance even when the
user edits style preferences.

4. Fixed two problems with hyperlinks in TextBufferWindows: (a) the
hit testing algorithm was not taking into account the left margin, and
(b) we were listening for mouse traffic on the scroll pane, not on the
view -- so the y-coordinates were all wrong.


[Since 1.03]
4-5 February 2004

1. Now using the FIRResampler in MicroMod, since it doesn't produce
awful static in Zetaspace Demo, unlike the LinearResampler.

2. Got rid of gross Thread.sleep(20) in SoundChannel, replacing it
with a Thread.yield().  Also got rid of a superfluous synchronization.

3. Using graphics device compatible images in GraphicsWindow.

4. Previously, Zing would fire a sound completion event for a MOD,
even if it had explicitly been stopped.  This behavior violates the
Glk spec and is now fixed.

5. Zing windows now have their preferredSizes set by their pair
windows (or, in the case of the root window, by static code).  For
some reason, this solves an annoying sizing bug, which manifests
itself in Java 1.5 on Linux.  (In Photopia 2001, there would be a
white stripe new the bottom of the frame, when the background was
supposed to be all black.)

6. Line input is now printed to echo streams.

7. Streams are now closed (and flushed) when the glulx program
controlling them exits.

8. If a PairWindow is closed, its children's streams are now closed.

9. If a PairWindow is closed, its children and their streams are
destroyed.


[Since 1.02]
9 January 2004

1. The change I requested to Adam Cadre's "Narcolepsy" revealed a
minor (but ugly) problem with Zing's GraphicsWindows.  They were not
painting their backgrounds.  Now they do.

2. Bugfix to Glk.select(): Eliminiated a race condition, which could
cause a crash when timer events are turned on.

3. Obtained a Thawte certificate, which allows Java WebStart to stop
saying that it strongly recommends not running Zag.


11 January 2004

1. TextBufferWindow.scrollDown() is now always called in the Swing
event thread, which eliminiates a long-standing race condition in the
code.

2. When I fixed the layout bug in version 1.02, I introduced a new
bug, where margin images would no longer be displayed.  This has been
fixed.

3. Made TextBufferWindow's getSplit() method take into consideration
the 5 pixel top margin.

4. Gave the TextGridWindow a 5 pixel top margin, so the top line is no
longer smushed against the top of the window.


12 January 2004

1. Became trusted in the Thawte Web of Trust, which means that the
digital certificate I use to sign Zag now bears my name, and my
identity is vouched for by Thawte.

2. A change to #1 from yesterday: often resizeView() was already
called from the Swing event handling thread.  resizeView(), in turn,
calls scrollDown().  Since I (yesterday) wrapped the latter call in a
SwingUtilities.invokeLater(), it was often being deferred until *too
late*.  So, I undid the change to scrollDown() and instead ensured
that calls to TextBufferWindow.doLayout() (called from Glk.java) occur
in the event thread.  This fixes a couple of bugs I introduced
yesterday.

3. Restored the downloading progress bar to undecorated windows.


16 January 2004

1. Regression: realized that Zetaspace Demo had a problem on OS X.
The space-flight navigation animation wasn't working properly (though
it was working on Linux and Windows).  Made a change to fix the
problem on the Mac.  Still works on other platforms, but the animation
is slower now.  (Really, this is Apple's bug, not mine, but what can I
do?)


2 February 2004

More excellent bug reports from Joonas.

1. Fixed a few issues with string handling.  Type 3 Huffman nodes
(C-string nodes) were not working properly, for two reasons.  First, I
was decoding those nodes incorrectly.  Second, I was missing an small
but important bit of the insanely complex compressed-string printing
algorithm.

2. Zag's ability to handle changes to Huffman tables stored in RAM was
faulty.  Specifically, Zag assumed that the address of the tree root
would remain constan.  Now, it doesn't.


[Since 1.01]
30 September 2003

Another round of excellent bug reports by M. Joonas Pihlaja.

1. The BufferedImage in TextGridWindow must have length and width > 0.
(tetris.ulx was crashing).

2. In Style.java:  changed default styles a bit.  This will not affect
users who have used previous versions of Zag and who have saved style
preferences.

3. The menu option "End session" no longer sends "quit\n" to the
program.  After all, this will only work on English games that happen
to be requesting line input.  Rather, it now just forcibly quits the
running glulx program.

4. Attempt to optimize TextBufferWindow.doLayout()


6-7 January 2004

First time I've worked on Zag in awhile...

1. Fixed an important layout bug in TextBufferWindow, discovered some
time ago by M. Joonas Pihlaja.  If line input were requested on a
TextBufferWindow and if the cursor were positioned at the very start
of the line (i.e., sans prompt), the cursor would display at the end
of the previous line.  This is now fixed.

2. Fixed Glk.setWindow(), after a spec. clarification from Andrew Plotkin.

3. Invalid GLK object references no longer (necessarily) crash the
program.  By default, warnings are issued instead.


[Since 1.0]
9 September 2003

1. Fixed a bug in the handling of the protect opcode.  I was treating
the second operand as a "store" operand, rather than a "load."

2. Blank windows can now be split.  (Between the previous fix and this
one, "Words of Power" will run now, though Zag will not play its
music--which is in XM format, so this is not a bug in Zag; it's a bug
in "Words.")

3. (Thanks to Zak.) Fixed background color preferences.  I was saving
the preference with the name "back-color" and attempting to read it
back as "background-color."

4. (Thanks again to Zak.) Fixed TextBufferWindow.java so that the
background color of the "normal" style is used as the window
background color.


[Since the fourth beta release]
14 August 2003

1. Fixed multi-line input for buffer windows.  (Actually input worked
just fine, but display of that input did not.)

2. Again, fixed slow buffer window line input in Photopia 200x (which,
apparently, I broke when I fixed multi-line input)...  Somehow this
branch got out of sync with a (most likely) abortive branch I started.

3. Made a few more Glk functions take a fast path through the glk
opcode function.  I don't notice any difference in speed.

4. jar file is now zag-1.0.  No "b" for beta anymore, since the next
release will be 1.0.

5. README.txt updated to reflect current state of Zag.


18 August 2003

1. Fixed bug in meta-backspace keybinding, which must have existed
from the start...


[Since the third beta release]
7 August 2003

A slew of *great* bug reports from M. Joonas Pihlaja (who clearly wins
the beta testing competition, whatever Jessica Knoch says).  He
actually has a suite of programs for testing Glulx implementations.
Brilliant.  All of these fixes are prompted by his testing.

1. Fixed an automatic type promotion bug in IO.java, when checking the
type of a string to send to a GLK function.  An int comparison was
taking place, so the type byte (0xe0) was being promoted to an int,
with sign-extension, causing it to be negative.  So the comparison
failed, when it should have succeeded.

2. Fixed line input in grid windows.  This worked a long time ago, and
then I broke it...  Now it works again, although output can be
annoyingly slow.  I'll have to look into speeding that up.

3. Fixed Glk.selectPoll().  It had two obvious (and obviously stupid)
control flow problems.

4. Fixed Zag.storeShortOperand() and Zag.storeByteOperand() (perhaps
not so aptly named).  They were sign extending data when pushing it
onto the stack, which is against spec.  This caused tetris.ulx to crash.

[Oh, actually, these two bugs I discovered on my own.]

5. Fixed a "bug" in Zag.enterFunction().  Not really a bug; I just
applied a strength-reduction optimization.

6. Fixed a bug in copyb, where I was doing something obviously stupid.

[Now enhancements, suggested by M. Joonas Pihlaja]

7. Screen is not cleared anymore when a glulx program exits (but Zag
keeps going).

8. The file chooser starts in the current working directory, instead
of the user's home directory.  It also filters, so only directories,
.ulx, and .blb files are displayed by default.


11 August 2003

1. When glulx progam exits, it says so on the status bar: 
[*** End of Session ***], like Infocom used to say...

2. Style hints are cleared before a new glulx program starts.

3. Line input in grid windows has been optimized.


12 August 2003

1. Infinite UNDO added.


13 August 2003

1. SONG support added.

2. Fixed a problem with grid window display, instroduced by the
optimizations I added recently.

3. Infinite UNDO uses a single file now, rather than a whole mess of
them.

4. Grid windows explicitly buffer their drawing now (like graphics
windows).  Had to do this to resolve a serious drawing problem on
Linux. 

5. Closing the Zag window now actually exits the program.  (It used to
leave you in limbo.)


[Since the second beta release]
1 August 2003

1. Implemented WinGlulxe's configuration file support.  Can specify
whether the frame should be decorated, whether windows should have
borders, the initial size of the frame, and so forth.  Did not
implement "WindowMask", since Java frames cannot be transparent.
Supposedly, this will change in 1.5.

2. Undecorated frames pose a problem:  where do we put the "[More]"
prompt, for paging?  Some interpreters put it directly in the text.  I
put it on a status bar at the bottom of the frame.  An undecorated
frame, of course, has no status bar.  I probably should just have
omitted it altogether.  In fact, I just noticed that Andrew Plotkin's
MacGlulxe (at least, the Carbonized version) doesn't show a "[More]"
prompt, ever.  And I didn't miss it.  However, I decided to do
something tricky and possibly bad.  I display a somewhat largish
prompt in the bottom left hand corner of the frame.  But since the
prompt may be obscuring something important (I have no idea what I'm
painting over), I fade the prompt after a few seconds.  It disappears
altogether about seven or eight seconds after appearing.  Probably
unnecessary...

3. Got rid of the "[Press any key to exit]" stage when quitting.
Didn't see the point, but I can be persuaded otherwise.

4. If a program is launched with a configuration file, exiting the
program quits Zag altogether.  After all, the configuartion is
tailored specifically for the one program.

5. Some painting optimization in TextBufferWindow.  Line input isn't
nearly so sluggish now on Mac OS X (it never was on Linux,
interestingly).

6. Fixed the meta-backspace so that it works on linux.  Don't really
know why it didn't work before-- seems like Java doesn't understand
the X keyboard map, or something.  So, now mhy code allows Alt, in
addition to Meta.

7. Changed non character keys to use keyPressed() event handler,
rather than keyReleased().  This allows repeating by holding down the
key.  (E.g., arrow keys.)

8. Few changes to Zag.parseOperands().  Streamlined it a bit and fixed
a bit-masking bug which hadn't yet manifested itself anywhere.

9. Made glulx memory and stack into indirect ByteBuffers.  They were
direct, which seems...silly.  In fact, it seems silly that they are
ByteBuffers instead of byte arrays.  The latter surely have better
access times; the former are more convenient for many tasks, however.


2 August 2003

1. Added a menu option (under Edit) to use or to ignore style hints.

2. Hide the progress bar when not in use.  This saves gobs of CPU time
on OS X.

3. Eliminated a loop in Zag.parseOperands().  Should be marginally faster.

4. [see # 9, 1 August 2003] Replaced heavy use of ByteBuffers with my
own class, FastByteBuffer, which is basically a thin wrapper around a
byte[], with useful accessors.  It's much faster.

5. Frame should now remember its location and size from the previous
time Zag was run.

6. Fixed a bug on Linux, where the scrollbar could gain focus and
never give it up.


[Since the first beta release]

29 July 2003

Fixed display of accented characters (bugs casting between bytes,
ints, and chars).  [Thanks to Eric Forgeot.]


30 July 2003

1. Fixed style hinting for justification; was using wrong constants.
[Thanks to Eric Forgeot.]

2. Fixed character input to use Glk keycodes wherever possible.

3. Fixed TextBufferWindow.resizeView() to ensure that the height of
the view is always at least the height of the viewport.

4. Ripped out ana-mp and replaced it with MicroMod, by Martin Cameron
[http://website.lineone.net/~martin.cameron/].  I think it is rather
better than ana-mp.  I needed to make two minor modifications to
Martin's code, one to allow setting the volume, and another to allow
an odd number of channels.  Both were trivial changes.

5. Now using com.sixlegs.image.png code to decode PNG images, since
Java's built-in support for PNGs is incomplete.  If and when Sun
corrects this, I will remove this code.

6. Documentation changes to README.txt.  Added license directory and
put licenses there, along with README-license.txt.

7. Implemented some Emacs key bindings for line input in
TextBufferWindow;  now you can ctrl-arrow to go forward or backward
one word, and you can meta-backspace to erase the previous word.
Also, ctrl-a will position the cursor at the start of input, and
ctrl-e will position it at the end.

8. Buffered Stream.FileStream.  Now it's pretty hairy code,
unfortunately, because it has to be able to read and write.  It worked
on my limited tests, but I'm more concerned about the correctness of
this code than any other.

9. Buffered reads from the game file during saves and restores.
Between this and the last change, saving and restoring is much faster
now.


31 July 2003

1. Converted ArrayLists in TextBufferWindow to LinkedLists...which
just makes far more sense and conserves memory.

2. Changed the image cache in Glk.java to be a LIFO data structure
with a fixed size.  No more running out of memory from caching images
(in, for instance, Zeta Space).  I'm convinced that the new PNG
decoder uses far more memory than the native Java one.  Too bad the
latter doesn't work properly.