Bugs fixed in S-PLUS 2000 Professional Release 3:

WINDOWS 2000 ISSUES

     The objdiff function now works correctly on Windows 2000.

     File extensions for graphsheets and scripts are now displayed and
     saved in the Save As dialog on Windows 2000.

INSTALLATION AND STARTUP

     In previous versions of the S-PLUS Uninstall program, the prompt to
     remove the local working directory was set to "Yes" by default. This 
     is now set to "No".

     The mechanisms S-PLUS uses to search for a valid S_HOME directory
     during startup are improved. 

     In previous versions, the value of S_HOME in both the system registry 
     and the environment settings had precedence over a value set from the 
     command line (i.e., the Properties dialog of the S-PLUS shortcut 
     icon). If S_HOME is defined in the registry or in the environment, a 
     command line definition is no longer ignored.

     The mechanisms S-PLUS uses to search for S_HOME and S_PROJ 
     directories during startup now keep track of valid paths they find.
     This allows S-PLUS to write paths to the registry, so that you are 
     not prompted to enter new directories the next time you open S-PLUS.

     Two new environment variables for disabling startup dialogs have been 
     added. The /NOSTARTUPDLGS parameter disables all dialogs related to 
     S_PROJ, and the /NOSTARTUPDLGSATALL parameter disables all dialogs 
     related to both S_PROJ and S_HOME. If /NOSTARTUPDLGSATALL is set and
     S_HOME cannot be validated during startup, S-PLUS shuts down silently.

     A new registry entry has been added that allows you to define
     environment variables that are needed during startup via automation.

AUTOMATION

     The OLE support functions, including create.ole.object, now account
     for the changes between daylight savings time and standard time.

     When calling S-PLUS via Visual Basic Automation, character string
     data in arrays is now sent correctly.
   
     When calling S-PLUS via Visual Basic Automation, character string
     data sent to Microsoft Excel no longer results in an Application
     Error.

     If you open a graph sheet in a minimized S-PLUS (using automation),
     the graph sheet itself is no longer automatically minimized. 
  
     Creating a condition plot via automation, then increasing the number
     of panels, no longer crashes S-PLUS.    
                                                  
GRAPHICS

     If you create a trellis plot, then change the conditioning variable
     type from numeric to factor, S-PLUS no longer runs forever. 

     If data set is embedded in a conditioning plot, column names are now
     used correctly to create panel names. 

     Graphs now update when the Type property in the Conditioning Columns
     group of the Multipanel page is changed.    

     Modifying the current page property no longer results in a problem
     with recursive updating. 
    
     Plots made with Log Axes (base 10) no longer arbitrarily plot points
     for which log is undefined. 
    
     Extracting data from an embedded graphsheet no longer loses
     additional conditioning variables. 

     Histograms conditioned on two factor variables now plot correctly.

STATISTICS

     Although 95% confidence intervals are the default for a Linear
     Regression from the GUI, two-sided 90% confidence limits were
     returned in previous S-PLUS versions. The default 95% confidence
     intervals are now computed.

     The GUI calculation of skewness now returns a signed value, instead
     of a strictly positive value.

     The predict.discrim function now works properly with the 'newdata'
     argument, when the formula used to create the discrim.object includes
     the period syntax: <response> ~ .

MISC. PROGRAMMING

     The cut.dates function now gives sensible output when the 'breaks'
     argument is set to "months" or "years", and the 'x' argument contains 
     years of the form 20xx.

     The rowsum function now calls a C routine for improved speed and
     performance.

               

Bugs fixed in S-PLUS 2000 R2 fall into the following broad categories:

        S-PLUS Database
        Dates
        Graphics
        Data Manipulation
        Import/Export
        Object Explorer/Workspace
        Statistics

S-PLUS DATABASE

The "Error: Unable to create summary in summary file" should now occur
much less frequently.Many of the conditions that could lead to that error
have now been corrected. 

Running 'For()' in S-PLUS 4.x/2000 for Windows often led to spurious
summary file warning messages: "Unable to create summary
file..._Data/_sum4.txt in: assign(".Steps", steps,....)." The cause of the
warnings has been found and corrected. 

For() did not work if customers install S-PLUS for Windows into a
directory with spaces in it(for example: 'Program Files'!) It fails with
the error message: "Errors occurred in subprocess -- results are wrong or
incomplete in:..". The underlying problem, that For() did not understand
SHOME environment variables with embedded spaces, has been corrected. 

Objects copied from one _Data directory to another using 'objcopy()'
didn't always show up in a call to 'objects()' that looked at the target
directory.  S-PLUS did not always detect that a directory database was
already attached, and thus did not always update its internal list of
objects for that database.  This has now been fixed. 

If you attached more than one database (not necessarily a
dataframe/object) in a function (or top-level expression) that resulted in
an error (in the GUI only), the following error message appeared when you
detached a dataframe: "Error in UseMethod("assign", where): No specific or
default method found(Dumped)." Once this occurred, detaching any database
caused a GPF. 

If you changed the units in a Graph Sheet to mm, then saved graph
properties as default, your next graph sheet would correctly default to
mm, but a 2D plot created in the new Graph Sheet would have overly large
tickmarks because they did not take into account the new default units. A
variety of similar problems relating to saving mm units as default have
also been corrected.

On exiting Splus, the __init and __prev directories are now removed from
the directory containing database 1 after the Commit Changes dialog has
executed.  The Commit Changes dialog was giving objects modified in
previous sessions, and it kept growing and growing, and when you attempted
to restore an object with an earlier object, you often got one weeks or
months out of date. 

DATES

Dates data can now be entered consistently using either the Data Window or
the Commands window with the dates() function. Both the default origin for
Julian dates and the default century cutoff have been set consistently for
both methods. 

When seq.dates() was used, it created a vector with right dates in engine,
but the same vector was 60 years off in GUI. For example,

  > DateList <- seq.dates("1/31/1990", "12/31/1999", by = "months")

If you opened this in the Data Window, it would begin in 1930. The
seq.dates() result now displays correctly in the Data Window. 

GRAPHICS

In S-PLUS 2000, the colors in the "Color Scale Legend" do not match with
colors in the plot. If you opened a data set in the Data Window, selected
3 columns, selected "Color" from the 2D Plots palette, then clicked on
"Color Scale Legend" button on Standard toolbar, the points were all
shades of blue, while the legend displayed other colors. The colors now
match the legend. 

The Y Axis range on a Smoothing Spline plot is now calculated correctly
and the Spline plot does plot correctly when the X Axis Scaling is set to
Log. 

The 'NLS Fit' plot type available in the Axum 6.0/Splus-2000 GUI now works
correctly. The fitted line is now drawn onto the Graph Sheet.

A problem with polar plots not displaying as circles in .eps files has
been corrected. 

Axis labels using year series are now calculated correctly when
incremented by values less than -36.  The bigger the increment value you
use, the more apparent the calculation problem appears. This problem also
occurred in the Color Scale Legend labels. 

When the X axis is Date data, setting the Tick Range for First Tick and
Last Tick to Axis now work and correct tick marks are set at axis. 

Specifying the number of intervals for date data axes now gives the
correct number of intervals; previously, you saw one interval less than
was specified. 

Using a strip label of "Column of Labels" previously worked
inconsistently. This now works consistently. 

In Trellis plots, the conditioning variables were not displaying correctly
in the panels. They now correctly display as "Names and Values". 

A GPF no longer occurs when a single row number is entered in the subset
row field for QQ plot. 
		
Pressing F9 after creating a graph was causing changes to the plot and
ticks. This has been fixed so that F9 redraws the plot correctly.

Multiple filled contour plots on the same graphsheet now print correctly
if "display lines" is off.  The printed graphsheet was omitting some of
the plots. 

Contour plots are now plotted correctly when they are projected on the XZ
or YZ planes. 

After creating a boxplot, using the Data To Plot dialog to select a second
x column to produce side by side boxplots now works correctly.

Creating a QQ Normal with Line plot, then dragging a variable to create a
Trellis plot was producing a plot with no lines in the panels. This has
been corrected.

If no printers are installed on the computer, the command graphsheet() (or
any engine command that calls graphsheet) now results in a graph with the
correct orientation as specified in the settings.

Editing the y axis label of an embedded graph in another application was
causing the x axis labels to shift to the left hand side of the graph
until the edit mode was deactivated. The x axis labels now stay in place. 

Attempting to export to GIF resulted in zero-length files. In most cases,
this was caused by the omission of two DLL's from the CMD directory
(HSIIGIF.DLL and HSIOGIF.DLL). These DLL's are included in release 2, but
zero-length files may still result from sufficiently complex graphs. If
this occurs in release 2, contact MathSoft Technical Support. 

When editing the y-axis label of an embedded graph in Mathcad or other
application, the x-axis label shifted to the left hand side of the graph
until you exited edit mode. 

In creating 2D and 3D graphs, no warning message was generated if
non-existent column was specified.

DATA MANIPULATION
 
When a logical column was inserted, only the value of T was displayed. The
correct values are now displayed. 

When Packing Columns from a data sheet, the relationship of the row names
with the correct row is now maintained. 

When binning from the Create Categories dialog, binning was being done on
the log scale.  The bins now correctly correspond to the numeric values in
the source columns. 

When Count was selected from the Create Categories dialog, the minimum
value in the source column was being represented in the target column as
NA. The correct value now appears in the target column. 

Using the Data Unstack dialog, if an object with the name specified in the
To Data Set field did not already exist, an object with a default DS name
(e.g. "DS1") was being created along with the specified data set. The data
set with the default name was the one displayed. This has been fixed so
that only the object with the specified name is created and displayed. 

In the Data Unstack dialog, the Group Column in the Row Grouping group is
now correctly updated when the From Data Set field is changed.

The following dialogs were missing <ALL> as an option in the dropdown list
of the Columns field:  Pack Columns, Remove Columns, Clear Columns, Remove
Block, Clear Block.  <ALL> is now the default for these dialogs. 

ts.union no longer fails when given an its constructed from a matrix with
a single column. 

In the Data Window, using Copy and Paste on a numeric column retained only
the display precision.  The text sent to the clipboard now uses full
precision: either "%.15g" or "%.6g". 

Exporting to ASCII did not retain the full precision. The exported text
now uses full precision: either "%.15g" or "%.6g". 

IMPORT/EXPORT

Importing Excel files with formulas that create logical and char values
caused error "Out of memory or not enough memory for what you are trying
to do...." This was corrected by setting TextAsFactors to Never for these
files. 

The extension on SAS Transport Files that the File Import/Export utility
uses by default has been changed from *.tpt to *.xpt. This is the new
format that the Food and Drug Administration officially supports. 

Importing SAS *.sd2 files that are generated from code that contains a
call to the SAS 'format' statement now works correctly. 

When using the Export Data From File dialog, and choosing columns to
export - you cannot use CTRL-click method to export multiple columns. You
have to type in the names. 

The origin was wrong in the Dates column returned by import.data.fame. It
was 1/1/2000 but has been corrected to be 1/1/1960. 

OBJECT EXPLORER/WORKSPACE

The GUI no longer unnecessarily loads the data from the engine when
columns of a DS are selected in the OE, synchronizing selection in the OE
when the window becomes active and on top, or initializing any dialog with
a wide formula control. 

Workspace files with type *.sws now launch S-PLUS from the Windows
Explorer. 

STATISTICS 

A mistake in the default weight expression returned by glm.weight resulted
in incorrect answers from many calls to glm(). 

The function menuChisq() incorrectly applied Yates' correction algorithm
in all cases when a contingency table was given as input, even when the
checkbox for Yates correction was cleared. It now respects the checkbox. 

plot.multicomp() now correctly computes the number of comparisons per page
if the number of comparisons is an exact multiple of 21.

A new function, print.survfit.computations, can be used to obtain the
matrix of results printed by print.survfit. 

A browser statement left in plot.discrim has been removed. 



BUGS FIXED FOR S-PLUS 2000 Release 1

GUI Trellis no longer drops the maximum value from each conditioning level
when double precision data columns are used for conditioning. 

Error reporting of clipped lines in traditional command line graphics
sometimes resulted in a GPF.  This has been fixed. 

Unless the "short date" was set in the control panel setting, any date not
in the range 1900-1999 was recorded incorrectly in the engine. This has
now been fixed. 

The "Table Name" field in the "Import ODBC" dialog always displayed the
wrong table name when connecting to an ODBC database. This has now been
fixed. 

A limit of 127 attaches per S-PLUS session has been removed.
 
OLE Automation Bug Fixes:
- Fix for passing date/time values in a two-dimensional array to S-PLUS.
- Corrected reference counting problems when copying incoming variants to
  temporary variants.
- Corrected reference counting problems when calling methods and setting
  properties on AutomationClient objects from S-PLUS.

The arima.forecast function now gives the correct number of forecasts. 

In importing data from Excel or Access, column names are no longer
truncated to 31 characters. 

If you select two numeric columns and a factor column, then click on the
Line plot button, a separate line plot is now created for each factor
level. 

You can now specify the default font list for command line graphics. See
the Release Notes for further details. 

Formatted date columns from Excel now import correctly into S-PLUS.

Changes to column width and description made in the Column properties
dialog are now saved correctly.

The toolbar item for inserting columns has been redesigned so that its
purpose is now clearer.  Confusion with changing column types has been
reduced. 

You can now specify a column to use as labels when identifying points
using the Annotations palette. 

The first 74 characters from the Description field of a column are now
used for axis labels if present.  (The previous limit was 31 characters.) 

Predictions, residuals, fitted values, and other results can now be
appended to the original data set when fitting tree models using the menus
and dialogs. 

The predict function now gives correct answers for lm and glm objects when
the newdata argument is given that includes factors with just a subset of
the factors used in the original fit. 

There is now an option (set by default) that allows the Commands window to
retain focus when you create a graph from the Commands window. 

When importing a SAS or SPSS file containing missing values, NAs were read
as NaNs when data type was "factor". This has been corrected. 

When importing a SAS file, the options StartRow and EndRow are now
recognized. 

The following bootstrap bugs have been fixed:

        Fixed bug when using group and any group had size 1.

        Fixed handling of seeds.

        Fixed calculation of bias in `jack.after.bootstrap'.

        Fixed bug that caused failure if `B'=1 + multiple of `block.size'

In the uniroot and optimize functions, argument handling for passing ...
arguments and the attributes from previous function calls back to the
function has been improved.