ACT-R/PM Release Notes

Tracking the version history of ACT-R/PM

Version 2.2

Released 2004.03.12
When only one or two things on the display have changed and the display is very complex, calling PM-PROC-DISPLAY could be slow. New functions PM-ADD-SCREEN-OBJECT and PM-DELETE-SCREEN-OBJECT should help.
Numerous small changes to the documentation: some small errors corrected, some stuff added.
Noise in the cursor movement output position is now a switchable feature.
Time randomization is more flexible. The :RANOMIZE-TIME parameter can now take a number as an argument to set the width of the uniform random distribution; see the documentation.
Added a VIEW-KEY-EVENT-HANDLER method for editable text dialog items, which used to break. (MCL only.)
Code updated with LGPL stuff.
Changed name of BUILD-SEVT-DMO to AUDIO-ENCODING-COMPLETE to improve trace readability.
Added hash table for word articulation durations so they can be looked up rather than computed on the fly. Set with (PM-REGISTER-ARTICULATION-TIME string time). Time is in seconds.
Added a new parameter, :REAL-TIME-SLACK-HOOK-FCT. This is a function which will be called when ACT is running in "real time" mode whenever ACT is waiting for real time to catch up to simulated time. The function is called with one argument, which is the disparity (in seconds) between ACT-R's clock and the wall clock.
Fixed bug in +aural-location> handling of the "kind" and "locations" slots.
Homing the hand to the mouse is now faster because it uses a more appropriate width.
An under-the-hood change to some of the MCL interface stuff (new RPM-OVERLAY view class).
Changed the way the cursor location is represented internally, which should solve some odd cursor movement problems.
Made some minor changes in EMMA to better support the eye spot on non-MCL Lisps. DEVICE-UPDATE-EYE-LOC will now be called on the device whenever the eye position changes, and device writers can feel free to define a method for that, or not. A default method for MCL windows is supplied.

Version 2.1.2

Released 2003.06.20.
There's a change for 2.1.1 that I forgot to document here. When doing a +visual-location>, you could include a "nearest" value of CURRENT to get the closest visual location. This uses Euclidean distance. If you care about nearness in only one dimension, you can now use CURRENT-X or CURRENT-Y.
A call to MOVE-ATTENTION will now set the PROCESSOR state to busy.
Fixed bug in prepare-only motor movements not leaving the PROCESSOR state free.
Under MCL, event-dispatch is now called much less often; at most, every 100 real ms. (You can change this in the code if you want it even less often, it's in the :after method for DEVICE-UPDATE in "mcl-device.lisp".) This should speed simulations. If you have trouble with things not updating on the screen as a result, please let me know.
Another MCL thing: static-text-dialog-items have different alignment options, which should now be handled properly in terms of the locations of the generated text.

Version 2.1.1

Released Feburary 7, 2003.
Added ASSIGN-FINST action. This allows a finst to be assigned without the actual location being attended, which is something that several people have asked for. Works like this on the RHS:
```
+visual>
  isa		assign-finst
  location	<loc>
```
where <loc> is a chunk of type visual-location. You can also do it by object, like this:
```
+visual>
  isa		assign-finst
  object	<obj>
```
where <obj> is visual-object chunk (or a subtype). [Someone is going to ask what the evidence for this ability is. I'm going with Pylyshyn's ideas here.]
Made some small corrective changes in the logic of updating attended locations.
Dan fixed various symbol and other issues with ACL.
Each module now has its own version number.
Fixed how RPM accesses ACT-R chunks to make it more robust. Also, updates the module state chunks much less often.
It now moves the cursor around properly under OS X. This involves the "cfmbundle.lisp" file.

Version 2.1b8/version 2.1

This was a stealth release.
The Vision Module now prints more trace information.
Updated MOVE-CURSOR so now the action to be programmed while the hand is still moving to the mouse from the previous action.
Fix by Dan of some lingering issues with deferred PROCESS-DISPLAY stuff.

Version 2.1b7

Released July 23, 2002. Assuming this goes smoothly at the Summer School/Workshop, this will be the final 2.1 beta.
Bug in non-MCL Lisps of not copying the DISTANCE slot in visual feature specs fixed.
Set TRACE-MODULES initform to T so that it will provide the expected trace on the initial run.
Added LOCATION slot to sound events, and SUBVOCALIZE now sets this slot to INTERNAL to represent internal speech.
SPEECH-AVAILABLE-P should now be defined even if speech-utils is not loaded.
A bunch of under-the-hood changes by Dan to the UWI. This affected a couple method/function names for MCL, too, and caused some renaming of files. Don't panic.
Now discriminates between Digitool MCL and OpenMCL. For now.
PROCESS-DISPLAY now correctly propagates CLEAR parameter when deferred.
CHECK-SPECS now prints an informative error message when there's an error, and more intelligently checks for one.
Fixed teensy-weensy glitch with sound event initializer and EVENT->DMO now includes location.

Version 2.1b6

Released June 5, 2002.
Many users have complained about the fact that RPM takes over the mouse even when their model doesn't use the mouse. Good point. There is now a paramter called :NEEDS-MOUSE which can be set to NIL to disable RPM taking over the mouse.
Outputting speech now outputs sound to the audicon. Also added a SUBVOCALIZE command which outputs the sound only to the audicon, but not to the environment.
Added an internal hook function to support the environment's stepper.
Lingering bug with display processing happening mid-production resolved by Dan adding a priority system to the queue.
Removed unusued OBJECTS slot from audio-event chunks and added EVENT slot to sound chunks to parallel vision.
Added a couple items to the stepper, including printing the icon and suppressing the menu.
Object tracking now updates the visual location buffer to the location of the tracked object, and it tracks the visual feature rather than the actual screen object.
Added text color processing stuff (used to be an extra) to MCL interface.
Added BUILD-FEATURES-FOR methods for radio buttons and check boxes (MCL-only).
Made PM-PRINT-ICON output more tabular--this should be easier to read.
PROCESSOR wasn't being set to BUSY correctly for some motor events. Fixed.
Backquoted some macros to make ACL happier.
Fixed one more (and hopefully the last one) vector bug glitch.

Version 2.1b5

Released 2002.04.11
Probably the biggest change is with display process (PM-PROC-DISPLAY). If this is called while a production is in progress, it is now deferred until the end of the present cycle. This should help with some of the strange "blink" behavior that some people were seeing.
The second biggest change is that the visual system is now 3D, in that each location has a distance as well as an X and a Y. I've only tested this in a cursory way, so please beat on this if you plan to use it.
FOCUS-ON function renamed to ENCODING-COMPLETE to make it a little clearer in the trace what's going on.
Height and width information is now correctly propagated to visual objects.
Size now propagated to visual objects, and rounded to two decimal places.
Also, the VALUE slot for visual locations was being propagated, which is bad for text (meant text could be read without an attention shift). Bad, and fixed.
Default for :VISUAL-MOVE-TOLERANCE is now 0.5 degrees (changed from zero), because it seems like this should be on.
If you weren't seeing the savings you expected for feature preparation of aimed movements, that's because there was a bug in that computation. Fixed.
Weird glitch in approach width computation fixed.
MOVE-CURSOR now checks to make sure the hand is on the mouse.
Some glitchiness with running with variable thresholds fixed by Dan.

Version 2.1b4

Released 02.01.21.
Fixed bug with GREATER-THAN-CURRENT and LESS-THAN-CURRENT in FIND-LOCATION calls.
Fixed bug with NEAREST if it used something other than CURRENT.
Motor Module now ouputs INITIATION-COMPLETE messages.
Another reduction in amount of GENTEMPing.
A bunch of declaration changes were made to be more ACL-friendly (Dan).
Some more clever WITHIN-* methods to handle variables passed in. (Dan).
Cognition module retrieval bug fixed. (Dan).
Numerous other small fixes and changes. (Mostly Dan.)

Version 2.1b3

Released 01.11.12, first public release of any version of 2.1
Returned :FULL-TIME to PM-RUN.
Added support for tracing paths of attention, the mouse, and the eyes (with EMMA). Added the "scanpath.lisp" extra so people could take advantage of this new functionality. Thanks to David Peebles for the idea.
Fixed a number of bugs in EMMA having to do with the "processor free" test.

Version 2.1b2

Internal release only
Cleaned up the icon-building process to generate fewer gentemps, which probably also resulted in a speed boost over long runs.
New RPM stepper implemented. Try (PM-STEP) to give it a whirl.
Re-wrote scheduler a little to fix some lingering problems with how long it would run.
Added a PRINT-OBJECT function for queue entries so that they're more human-readable.
Fixed a long-standing (but generally avoided) infinite recursion bug in computing approach width.
Changed attention shift latency to 0.085 sec, which is what it should always have been in 5.0.

Version 2.1b1

Internal release only
First integration of Dan and Christian's code with my most current RPM code. Lots of things new for ACT-R 5.0 so I won't list anything here.