These changes are in more-or-less reverse chronological order, with the most recent changes first.
See also the list of Qexo (XQuery)-specific changes.
Many enhancements to arrays and vectors:
Shape specifiers (used when creating an array) can now be one of a rank-2 array of low/high-bounds, as in SRFI-25; a vector of upper bounds; or a vector of ranges.
Generalized array indexing:
If A is an array (or a vector), then the expression:
(A I J K ...)
in general evaluates to an array B such that:
(B i1 i2 ... j1 j2 ... k1 k2 ... ...) is
(A (I i1 i2 ..) (J j1 j2 ...) (K k1 k2 ...) ...)
If an index I is an integer, it is treated as a zero-index array - a scalar.
For example: if (define B (A 2 [4 <: 10]))
then (B i) is (A 2 (+ i 4)).
The procedure array-index-ref is does the above
indexing exlicitly: (array-index-ref A I J K ...) is
(A I J K ...).
The result is a read-only snapshot.
The procedure array-index-share is like
array-index-ref but creates a modifiable view into argument array.
(build-array shape procedure)
is a general constructor for lazy arrays:
If A is the result, then (A i j k ...)
is (procedure [I J K ...]).
array-transform creates a view, with a mapping of the indexes.
Other new procedures (like those in trhe Racket math package):
array-size, array-fill!,
array-copy!,
array-transform, array-reshape,
array-flatten, array->vector,
index-array, build-array.
Add Common Lisp array reader syntax (#rankA)
with Guile extensions, including reader sypport for multi-dimensional
uniform (primitive) arrays.
This is also used when printing arrays.
New format-array procedure print an array a tabular
2-dimensional (APL-like) format.
This format is used by default in the top-level of the REPL.
Various REPL enhancements when using DomTerm.
The new --with-jline2 configure option
builds support for the JLine 2
library for handling console input, similar to GNU readline.
Context-dependent command-completion (tab-completion) works when using JLine2.
Lots of little changes, and some big changes to sequences and strings.
Enhancements to the Kawa tutorial.
Added parameter as a new typename, for Scheme parameter
objects. It can be parameterized (for example parameter[string])
for better type inference when "calling" (reading) the parameter.
We now define interactive mode
as a REPL or a source module
that uses the default global top-level environment or
or a source module import/required by a interactive module.
Interactive mode attempts to support dynamic re-definition and
re-loading of function and other definitions.
This is a work-in-progres; interactive mode currently uses extra
indirection to support re-definitions (at a slight performance cost).
Various changes and fixes in Path/URI handling. Most significantly,
the resolve argorithm used by resolve-uri was re-written
to use the algorithm from RFC-3986, rather than the obsolete RFC-2396
algorithm used by java.net.URI.resolve.
Change to mangle class and package name in Symbolic Freedom style. This means that class names and class filenames usually match the source file, even if special charaters are used, except for a small number of disallowed characters. Note this is currently only used for class and package names.
Allow 'synchronized and 'strictfp
as access flags for methods.
You can now have a type-specier for define-variable.
Better support for forward references between macros.
Added unsigned primitive integer types ubyte,
ushort, uint, and ulong.
These are represented at run-time by the corresponding signed types,
but Kawa generates code to do unsigned arithmethic and comparisons.
Corresponding boxed classes are gnu.math.UByte,
gnu.math.UShort, gnu.math.UInt, and
gnu.math.ULong.
Improvements and unification of sequences and strings:
The new sequence type generalizes lists,
vectors, arrays, strings, and more. It is implemented as the
java.util.List interface, but strings
(java.lang.CharSequence) and Java arrays are compatible
with sequence and converted as needed.
The length function is generalized to
arbitrary sequences. (For strings it uses the CharSequence.length
method, which returns the number of (16-bit) code units. This is different from
the string-length function, which returns the number
of Unicode code points.)
A new pseudo-character value #\ignorable-char
is introduced. It is ignored in string-construction contexts.
The function-call syntax for indexing works for all sequences.
If the sequence is a string, the result is the Unicode (20-bit) scalar value
at the specified index. If index references the trailing surrogate
of a surrogate pair the result is #\ignorable-char.
This allows efficient indexing of strings: Handing of surrogate pairs
are handled automatically as long as #\ignorable-char is skipped.
Indexing of uniform vector types (such as
s64vector or f64vector or u16vector) now
return the standard
primitive type (such as long or double) or the new unsigned primitive (such as ushort).
This improves performance (since we can generally use primitive types),
and improves compatibility with Java arrays. Specifically,
s64vector now implements Sequence<Long>,
and thus java.util.List<Long>
Note that indexing a f64vector returns
a double which as an object is a java.lang.Double,
not the Kawa floating-point type gnu.math.DFloNum.
The result is usually the same, but eqv? might
return a different result than previously.
The arguments to map,
for-each, and vector-for-each can now be
any sequence (including strings and native arrays).
The arguments to vector-for-each can now be
arbitrary java.util.List values.
All of these are inlined. If the sequence type is known,
more efficient custom code is generated.
A range represents an enumerable sequence,
normally integers, but it is represented compactly using
the start value, the step (usually 1), and size.
There is a new convenient syntax for writing a range: if i
and j are integers then [i <=: j]
is the sequence of integers starting at i and
ending at j (inclusive).
You can also write [i <=: j] (excludes the upper bound),
[i >: j] (counts down to j, exclusive),
and [i >=: j] (counts down to j, inclusive).
You can use a sequences of integers to index a sequence.
The result is the sequence of the selected elements.
In general (seq [i0 ... in]) is
[(seq i0) ... (seq in)].
This work well with ranges: (seq [i <: j]) is the subsequence
of seq from i to j (exclusive).
If the seq is a string (a CharSequence)
then the result is also a string. In this case the indexing behavior is
slightly different in that indexing selects (16-bit) code units,
which are combined to a string.
A new dynamic type is like Object.
However, it forces runtime lookup and type-checking, and supresses
compile-time type check and errors.
(This is similar to C#. It is useful as an escape hatch if we ever
implement traditional strict static type-checking.)
Specifying the parameter type or return type of a function or method
without a '::' is deprecated and results in a warning.
In --r7rs mode:
The 'l' exponent suffix of a number literal creates a
floating-point double, rather than a BigInteger.
Added the hyperbolic functions: sinh, cosh, tanh, asinh, acosh, atanh.
The equal? function can now handle cyclic lists and vectors.
So can equal-hash.
The command-line option --with-arg-count=N
allows finer control of command-line-processing. It is used before an
action
, and specifies the N arguments
following the action are set as the command-line-arguments.
After the action, command-line-processing continues following
those N arguments.
Added the R6RS module (rnrs arithmetic bitwise).
The kawa.repl argument processor now
handles -D options.
The new class sub-form of import
allows you to import classes, and give them abbreviated names, like
the Java import statement.
The new form is more compact and convenient than define-alias.
You can also use a classname directly, as a symbol, instead of writing it in the form of a list:
(import (only java.lang.Math PI))
In the only clause of the import syntax you
can now directly rename, without having to write a rename clause.
Changes in the calling-convention for --full-tailcalls
yields a substantial speed-up in some situations.
The type of boolean literals #f and #t
is now primitive boolean rather than java.lang.Boolean.
General multi-dimensional arrays can be indexed
with function call notation.
E.g. (arr i j k) is equivalent to (array-ref a i j k).
You can also use set! with either array-ref
or function call notation.
The #!null value (Java null)
is now considered false, not true. Likewise for non-canonical
false Boolean objects (i.e. all instances of java.lang.Boolean
for which booleanValue returns false,
not just Boolean.FALSE).
New standard libraries (kawa base)
and (kawa reflect).
You can now use patterns in the let
form and related forms.
Implemented the lambda lifting optimzation.
An expression that has type T is now considered compatible with a context requiring an interface type I only if T implements I (or T is Object). (Before they were considered possibly-compatible if T was non-final because the run-time class might be a subclass of T that implements I.)
New --console flag forces input
to be treated as an interactive console, with prompting.
This is needed on Windows under Emacs, where System.console()
gives the wrong result.
You can now in a sub-class reference fields from not-yet-compiled super-classes. (This doesn't work for methods yet.)
The (? name::type value) operator
supports conditional binding.
The (! name::type value) operator supports unconditional
binding; it is similar to define-constant, but supports patterns.
More efficient implementation of call-with-values:
If either argument is a fixed-arity lambda expression it is inlined.
Better type-checking of both call-with-values and
values.
Jamison Hope enhanced the support for quaternions,
primarily the new (kawa rotations) library.
New define-alias can define aliases for
static class members.
The treatment of keywords is changing to not be self-evaluating (in Scheme). If you want a literal keyword, you should quote it. Unquoted keywords should only be used for keyword arguments. (This will be enforced in a future release.) The compiler now warns about badly formed keyword arguments, for example if a value is missing following a keyword.
The default is now Java 7, rather than Java 6. This means the checked-in source code is pre-processed for Java 7, and future binary releases will require Java 7.
The behavior of parameters and fluid variables has changed. Setting a parameter no longer changes its value in already-running sub-threads. The implementation is simpler and should be more efficient.
The form define-early-constant is similar to
define-constant, but it is evaluated in a module's
class initializer (or constructor in the case of a non-static definition).
Almost all of R7RS is now working:
Importing a SRFI library can now use the syntax
(import (srfi N [name]))
The various standard libraries such as (scheme base) are implemented.
The functions eval and load
can now take an environment-specifier.
Implemented the environment function.
Extended numerator,
denominator, gcd, and lcm to inexacts.
The full R7RS library functionality is working,
including define-library
The keyword export is now a synonym for module-export,
and both support the rename keyword.
The prefix option of import now works.
The cond-expand form now supports the
library clause.
Implemented make-promise and delay-force
(equivalent to the older name lazy).
Changed include so that by default it first seaches the
directory containing the included file, so by default it has the
same effect as include-relative. However, you can
override the search path with the -Dkawa.include.path property.
Also implemented include-ci.
Implemented define-values.
Fixed string->number to correctly handle a
radix specifier in the string.
The read procedure now returns mutable pairs.
If you need to use ... in a syntax-rules
template you can use (... template), which disables
the special meaning of ... in template.
(This is an extension of the older (... ...).)
Alternatively, you can can write
(syntax-rules dots (literals) rules). The symbol dots replaces the functionality of
... in the rules.
An underscore _ in a syntax-rules pattern
matches anything, and is ignored.
The syntax-error syntax
(renamed from %syntax-error) allows error reporting in
syntax-rules macros.
(The older Kawa-specific syntax-error procedure was renamed
to report-syntax-error.)
Implemented and documented R7RS exception handling:
The syntax guard and the procedures with-exception-handler,
raise, and raise-continuable all work.
The error procedure is R7RS-compatible, and the
procedures error-object?, error-object-message,
error-object-irritants, file-error?,
and read-error? were implemented.
Implemented emergency-exit, and modified exit
so finally-blocks are executed.
Implemented exact-integer?, floor/,
floor-quotient, floor-remainder,
truncate/, truncate-quotient,
and truncate-remainder.
The letrec* syntax is now supported.
(It works the same as letrec, which is an
allowed extension of letrec.)
The functions utf8->string
and string->utf8 are now documented in the manual.
The changes to characters and strings are worth covering separately:
The character type is now a new primitive type
(implemented as int). This can avoid boxing (object allocation)
There is also a new Functions like Implemented a Optimized Implemented New function New function The SRFI-13 function character-or-eof.
(A union of character and the EOF value, except the
latter is encoded as -1, thus avoiding object allocation.)
The functions peek-char now
return a character-or-eof value.
string-ref that take a character
index would not take into account non-BMP characters (those whose value
is greater than #xffff, thus requiring two surrogate characters).
This was contrary to R6RS/R7RS. This has been fixed, though at some
performance cost . (For example string-ref and
string-length are no longer constant-time.)
string-cursor API (based on Chibi Scheme).
Thes allow efficient indexing, based on opaque cursors (actually
counts of 16-bits chars).
string-for-each, which is now the
preferred way to iterate through a string.
string-map.
string-append! for in-place
appending to a mutable string.
string-replace! for replacing a
substring of a string with some other string.
string-append/shared
is no longer automatically visible; you have to
(import (srfi :13 strings)) or similar.
The module-name form allows the name to be a list,
as in a R6RS/R7RS-style library name.
The syntax @expression is a splicing form.
The expression must evaluate to a sequence
(vector, list, array, etc). The function application or
constructor form is equivalent to all the elements of the sequence.
The parameter object current-path returns (or sets)
the default directory of the current thread.
Add convenience procedures and syntax for
working with processes:
run-process, process-exit-wait,
process-exit-ok?, &cmd, &`,
&sh.
The functions path-bytes, and path-data can
read or write the entire contents of a file.
Alternatively, you can use the short-hand syntax:
&<{pname} &>{pname} &>>{pname}.
These work with "blobs" which may be text or binary depending on context.
The initial values of (current-output-port)
and (current-error-port) are now hybrid textual/binary ports.
This means you can call write-bytevector
and write-u8 on them, making it possible for an application
to write binary data to standard output.
Similarly, initial value of (current-input-port)
is a hybrid textual/binary port, but only if there is no console
(standard input is not a tty).
Jamison Hope contributed support for quaternions, a generalization of complex numbers containing 4 real components.
Andrea Bernardini contributed an optimized implementation
of case expressions. He was sponsored by Google Summer of Code.
The kawa.sh shell script (which is installed
as kawa when not configuring with
--enable-kawa-frontend) now handles -D
and -J options.
The kawa.sh script is now also built when usint Ant.
The cond-expand features java-6
though java-9 are now set based on the
System property "java.version"
(rather than how Kawa was configured).
An Emacs-style coding declaration allows you
to specify the encoding of a Scheme source file.
The command-line option --debug-syntax-pattern-match
prints logging importation to standard error when a syntax-rules
or syntax-case pattern matches.
SRFI-60 (Integers as Bits) is now fully implemented.
Ported SRFI-101. These are immutable (read-only) lists with fast (logarithmic) indexing and functional
update (i.e. return a modified list).
These are implemented by a RAPair class
which extends the generic pair type, which means that most
code that expects a standard list will work on these lists as well.
The class kawa.lib.kawa.expressions
contains an experimental Scheme API for manipulating and validating expressions.
Internal: Changed representation used for multiple values to an abstract class with multiple implementations.
Internal: Started converting to more standard Java code formatting and indentation conventions, rather than GNU conventions. Some files converted; this is ongoing work.
Internal: Various I/O-related classes moved to new
package gnu.kawa.io.
Various changes to the configure+make build framework:
A C compiler is now only needed if you configure with
--enable-kawa-frontend.
Improved support for building under Windows (using MinGW/MSYS).
Support for building with GCJ was removed.
You can pass flags from the kawa front-end to the
java launcher using -J and -D flags.
The kawa front-end now passes the
kawa.command.line property to Java; this is
used by the (command-line) procedure.
Various improvements to the shell-script handling, including re-written documentation.
Some initial support for Java 8.
More of R7RS is now working:
make-list, list-copy,
list-set! all the R7RS list procedures are implemented.
square,
boolean=?, string-copy!,
digit-value,
get-environment-variable, get-environment-variables,
current-second, current-jiffy,
jiffies-per-second, and features.
finite?, infinite?, and
nan? are generalized to complex numbers.
write, write-simple,
and write-shared are now consistent with R7RS.
string-copy, string->list,
and string-fill! now take optional (start,end)-bounds.
All of the R7RS string functions are now implemented.
=> syntax in case form.
'|Hello\nworld|.
Added define-private-alias keyword.
Extended string quasi-literals
(templates) as specified by SRFI-109. For example,
if name has the value "John", then:
&{Hello &[name]!}
evaluates to: "Hello John!".
Named quasi-literal constructors as specified by SRFI-108.
A symbol having the form ->type is a type
conversion function that converts a value to type.
New and improved check for void-valued expressions in a
context requiring a value.
This is controlled by the new option --warn-void-used,
which defaults to true.
The datum->syntax procedure takes an
optional third parameter to specify the source location.
See testsuite/srfi-108-test.scm for an example.
Instead of specifying --main the command line,
you can now specify (module-compile-options: main: #t) in
the Scheme file. This makes it easier to compile one or more
application (main) modules along with other modules.
A change to the data structure used to detect never-returning
procedure uses a lot less memory. (Kawa 1.13 implemented a
conservative detection of when a procedure cannot return.
This analysis would sometimes cause the Kawa compiler to run out of memory.
The improved analysis uses the same basic algorithm, but
with a more space-efficient inverted
data structure.)
Multiple fixes to get Emacs Lisp (JEmacs) working (somewhat) again.
We now do a simple (conservative) analysis of when a procedure cannot return. This is combined with earlier and more precise analysis of reachable code. Not only does this catch programmer errors better, but it also avoids some internal compiler errors, because Kawa could get confused by unreachable code.
Implement 2-argument version of log function,
as specified by R6RS and R7RS (and, prematurely, the Kawa documentation).
Implement the R7RS bytevector functions.
The bytevector type is a synonym for older
u8vector type.
Implement R7RS vector procedures.
Various procedures now take (start,end)-bounds.
Implement most of the R7RS input/output proecdures. Most significant enhancement is support for R7RS-conforming binary ports.
Various enhancements to the manual, including merging in lots of text from R7RS.
Improved Android support, including a more convenient Ant script contributed by Julien Rousseau. Also, documentation merged into manual.
Implement a compile-time data-flow framework, similar to Single Static Assignment. This enables better type inference, improves some warnings/errors, and enables some optimizations.
Jamison Hope added support for co-variant return types and bridge methods for generics.
datum->syntax and syntax->datum
are preferred names for datum->syntax-object and syntax-object->datum.
Implemented bound-identifier=?
and re-wrote implementation of free-identifier=?.
unsyntax and unsyntax-splicing,
along with the reader prefixes #, and #,@.
New and improved lazy evaluation functionality:
Lazy values (resulting from delay or future)
are implicitly forced as needed.
This makes lazy programming
more convenient.
New type promise.
The semantics of promises (delay etc) is now compatible with
SRFI 45.
Blank promises
are useful for passing data between processes,
logic programmming, and more.
New functions promise-set-value!, promise-set-alias!,
promise-set-exception!, and promise-set-thunk!.
The stream functions of SRFI-41 were re-implemented to use the new promise functionality.
Different functions in the same module can be compiled with or
without full tailcall support. You can control this by using
full-tailcalls in with-compile-options.
You can also control full-tailcalls using
module-compile-options.
Charles Turner (sponsored by Google's Summer of Code) enhanced the printer with support for SRFI-38: External Representation for Data With Shared Structure.
Optimize tail-recursion in module-level procedures. (We used to only do this for internal functions, for reasons that are no longer relevant.)
Add support for building Kawa on Windows using configure+make (autotools) and Cygwin.
Some support for parameterized (generic) types:
Type[Arg1 Arg2 ... ArgN]is more-or-less equivalent to Java's:
Type<Arg1, Arg2, ..., ArgN>
New language options --r5rs, --r6rs,
and --r7rs provide better compatibility with those
Scheme standards. (This is a work-in-progress.)
For example --r6rs aims to disable Kawa extensions
that conflict with R6RS. It does not aim to disable all extensions,
only incompatible extensions.
So far these extensions disable the colon operator and keyword literals.
Selecting --r5rs makes symbols by default
case-insensitive.
The special tokens #!fold-case and
#!no-fold-case act like comments except they
enable or disable case-folding of symbols.
The old symbol-read-case global is now only checked
when a LispReader is created, not each time a symbol is read.
You can now use square brackets to construct immutable sequences (vectors).
A record type defined using define-record-type
is now compiled to a class that is a member of the module class.
Annotations are now supported. This example shows how to use JAXB annotations to automatically convert between between Java objects and XML files.
Prevent mutation of vector literals.
More R6RS procedures: vector-map,
vector-for-each, string-for-each, real-valued?, rational-valued?, integer-valued?, finite?, infinite?, nan?, exact-integer-sqrt.
SRFI-14 ("character sets") and SRFI-41 ("streams") are now supported, thanks to porting done by Jamison Hope.
Kawa now runs under JDK 1.7. This mostly involved fixing some
errors in StackMapTable generation.
You can now have a class created by define-simple-class
with the same name as the module class. For example (define-simple-class foo ...) in a file foo.scm. The defined class will
serve dual-purpose as the module class.
Improvements in separating compile-time from run-time code, reducing the size of the runtime jar used for compiled code.
In the cond-expand conditional form you can
now use class-exists:ClassName as a
feature name
to tests that ClassName exists.
A new Kawa logo, contributed by Jakub Jankiewicz.
A new --warn-unknown-member option, which generalizes
--warn-invoke-unknown-method to fields as well as methods.
A new kawac task, useful for Ant build.xml files, contributed by Jamison Hope.
New define-enum macro contributed by Jamison Hope.
Access specifiers 'final and 'enum are now allowed in define-class and related forms.
Optimized odd? and even?.
If you specify the type of a #!rest parameter
as an array type, that will now be used for the "varargs" method parameter.
(Before only object arrays did this.)
When constructing an object and there is no matching
constructor method, look for "add" methods in addition
to "set" methods. Also, allow passing constructor args
as well as keyword setters.
See here for the gory details.
New expand function (contributed by Helmut Eller,
and enabled by (require 'syntax-utils)) for
converting Scheme expressions to macro-expanded forms.
SAM-conversion: In a context that expects a Single Abstract Method (SAM) type
(for example java.lang.Runnable), if you pass a lambda
you will get an object where the lambda implements
the abstract method.
In interactive mode allow dynamic rebinding of procedures. I.e. if you re-define a procedure, the old procedure objects gets modified in-place and re-used, rather than creating a new procedure object. Thus calls in existing procedures will call the new version.
Fix various threading issues related to compilation and eval.
When format returns a string, return a
java.lang.String rather than a gnu.lists.FString.
Also, add some minor optimization.
Inheritance of environments and fluid variables now work properly
for all child threads, not just ones created using future.
Now defaults to using Java 6, when compiling from source.
The pre-built jar works with Java 5, but makes
use of some Java 6 features (javax.script, built-in HTTP server) if available.
You can write XML literals in Scheme code prefixed by a #, for example:
#<p>The result is &{result}.</p>
New functions element-name
and attribute-name.
Various Web server improvements.
You have the option of using JDK 6's builtin
web-server
for auto-configued web pages.
Automatic import of web server functions, so
you should not need to (import 'http) any more.
Kawa hashtables now extend java.util.Map.
If a source file is specified on the kawa command line
without any options, it is read and compiled as a whole module
before it is run. In contrast, if you want to read and evaluate a
source file line-by-line you must use the -f flag.
You can specify a class name on the kawa command line:
$ kawa fully.qualified.nameThis is like the
java command. but you don't need to
specify the path to the Kawa runtime library, and you don't need a
main method (as long as the class is Runnable).
The usual bug-fixes, including better handling of the ~F
format directive; and fix in handling of macro hygiene of the lambda (bug #27042).
Spaces are now optional before and after the '::' in type specifiers. The preferred syntax leave no space after the '::', as in:
(define xx ::int 1)
define-for-syntax and begin-for-syntax work.
You can now use car, cdr etc to work with
syntax objects that wrap lists, as in SRFI-72.
You can now define a package alias:
(define-alias jutil java.util) (define mylist :: jutil:List (jutil:ArrayList))
--module-static is now the default.
A new --module-nonstatic (or --no-module-static)
option can be used to get the old behavior.
You can use access: to specify that a field
is 'volatile or 'transient.
You can now have type-specifiers for multiple variables
in a do.
Imported variables are read-only.
Exported variables are only made into Locations when needed.
The letter used for the exponent in a floating-point literal
determines its type: 12s2 is a java.lang.Float, 12d2 is a java.lang.Double,
12l2 is a java.math.BigInteger, 12e2 is a gnu.math.DFloat.
Internal: Asking for a .class file using
getResourceAsStream on an ArrayClassLoader
will now open a ByteArrayInputStream on the class bytes.
A new disassemble function.
If exp1 has type int,
the type of (+ exp1 1) is now (32-bit) int,
rather than (unlimited-precision) integer.
Similar for long expressions, other arithmetic operations
(as appropriate), and other untyped integer literals
(as long as they fit in 32/64 bits respectively).
Many more oprimization/specializations of arithmetic, especially when argument types are known.
Top-level bindings in a module compiled with --main
are now implicitly module-private, unless there is an explicit module-export.
SRFI-2 (and-let*: an and with local bindings, a guarded * special form) is now supported.
The reader now supports shared sub-objects,
as in SRFI-38
and Common Lisp: (#2=(3 4) 9 #2# #2#).
(Writing shared sub-objects is not yet implemented.)
A module compiled with --main by default exports no
bindings (unless overriden by an explicit module-export).
Factor out compile-time only code from run-time code.
The new kawart-version.jar
is smaller because it has less compile-time only code. (Work in progress.)
More changes for R6RS compatibility:
The reader now recognizes +nan.0,
+inf.0 and variations.
The div, mod,
div0, mod0, div-and-mod,
div0-and-mod0, inexact
and exact functions were implemented.
command-line and exit.
Support for javax.script.
Support for regular expressions.
Performance improvements:
Emit iinc instruction (to increment a local
int by a constant).
Inline the not function if the argument is constant.
If call-with-current-continuation is only used to exit
a block in the current method, optimize to a goto.
Generate StackMapTable attributes when targeting Java 6.
Kawa can now inline a function with multiple calls (without
code duplication) if all call sites have the same return location
(continuation). For example: (if p (f a) (f b)).
Also mutually tail-recursive functions are inlined, so you get constant
stack space even without --full-tailcalls.
(Thanks for Helmut Eller for a prototype.)
A number of changes for R6RS compatibility:
The char-titlecase, char-foldcase,
char-title-case? library functions are implemented.
Imported variables are read-only.
Support the R6RS import keyword, including
support for renaming.
Support the R6RS export keyword (though
without support for renaming).
Implemented the (rnrs hashtables) library.
Implemented the (rnrs sorting) library.
CommonLisp-style keyword syntax is no longer supported (for Scheme): A colon followed by an identifier is no longer a keyword (though an identifier followed by a colon is still a keyword). (One reason for this change is to support SRFI-97.)
The character names #\delete, #\alarm,
#\vtab are now supported.
The old names #\del, #\rubout, and
#\bel are deprecated.
Hex escapes in character literals are supported. These are now printed where we before printed octal escapes.
A hex escape in a string literal should be terminated by a semi-colon, but for compatibily any other non-hex-digit will also terminate the escape. (A terminating semi-colon will be skipped, though a different terminator will be included in the string.)
A backslash-whitespace escape in a string literal will not only ignore the whitespace through the end of the line, but also any initial whitespace at the start of the following line.
The comment prefix #; skips the following
S-expression, as specified by
SRFI-62.
All the R6RS exact bitwise arithmetic functions are
now implemented and documented in the manual.
The new standard functions (for example bitwise-and)
are now preferred over the old functions (for example logand).
If delete-file fails, throws an
exception instead of returning #f.
The code-base now by default assumes Java 5 (JDK 1.5 or newer),
and pre-built jar files will require Java 5.
Also, the Kawa source code now uses generics, so you need to
use a generics-aware javac,
passing it the appropriate --target flag.
New SRFIs supported:
In BRL text outside square brackets (or nested like ]this[)
now evaluates to UnescapedData, which a Scheme quoted string
evaluates to String, rather than an FString.
(All of the mentioned types implement java.lang.CharSequence.)
You can now run Kawa Scheme programs on Android, Google's mobile-phone operating system.
The macro resource-url is useful for accessing resources.
A new command-line option --target (or -target)
similar to javac's -target option.
If there is no console, by default create a window
as if -w was specificed.
If a class method (defined in define-class,
define-simple-class or object) does not have
its parameter or return type specified, search the super-classes/interfaces
for matching methods (same name and number of parameters), and if these
are consistent, use that type.
Trying to modify the car or cdr of a
literal list now throws an exception.
The .zip archive created by compile-file
is now compressed.
Java5-style varargs-methods are recognized as such.
When evaluating or loading a source file, we now always
compile to bytecode, rather than interpreting simple
expressions. This makes semantics and performance more consistent,
and gives us better exception stack traces.
The Scheme type specifier <integer>
now handles automatic conversion from java.math.BigInteger
and the java.lang classes Long,
Integer, Short, and Byte.
The various standard functions that work on <integer>
(for example gcd and arithmetic-shift)
can be passed (say) a java.lang.Integer.
The generic functions such as + and the real
function modulo should also work.
(The result is still a gnu.math.IntNum.)
If a name such as (java.util) is lexically
unbound, and there is a known package with that name, return the
java.lang.Package instance. Also, the colon operator
is extended so that package:name
evaluates to the Class for
package.name.
`prefix:,expression works
- it finds a symbol in prefix's package (aka namespace),
whose local-name is the value of expression.
A quantity 3.0cm is now syntactic sugar for
(* 3.0 unit:cm).
Similarly:
(define-unit name value)
is equivalent to:
(define-constant unit:name value)
This means that unit names follow normal name-lookup rules
(except being in the unit package
), so for example you can
have local unit definitions.
You can specify whether a class has public or package access, and whether it is translated to an interface or class.
You can declare an abstract method
by writing #!abstract as its body.
If a name of the form type? is undefined,
but type is defined, then
treat the former as (lambda (x) (instance? x type)).
A major incompatible (but long-sought) change:
Java strings (i.e. java.lang.String values) are
now Scheme strings, rather than Scheme symbols.
Since Scheme strings are mutable, while Java Strings are
not, we use a different type for mutable strings:
gnu.lists.FString (this is not a change).
Scheme string literals are java.lang.String values.
The common type for Scheme string is java.lang.CharSequence
(which was introducted in JDK 1.4).
Scheme symbols are now instances of gnu.mapping.Symbol, specifically the SimpleSymbol class.
A fully-qualified class name such as java.lang.Integer
now evaluates to the corresponding java.lang.Class object.
I.e. it is equivalent to the Java term java.lang.Integer.class.
This assumes that the name does not have a lexical binding,
and that it exists in the class-path at compile time.
Array class names (such as java.lang.Integer[])
and primitive types (such as int) also work.
The older angle-bracket syntax <java.lang.Integer>
also works and has the same meaning.
It also evaluates to a Class.
It used to evaluate to a Type, so this is a change.
The name bound by a define-simple-class now evaluates to a
Class, rather than a ClassType. A define-simple-class is not allowed
to reference non-static module-level bindings; for that
use define-class.
New convenience macro define-syntax-case.
Fix some problems building Kawa from
source using configure+make.
New types and functions for working with paths and URIs.
Reader macros URI, namespace, duration.
Simplified build using gcj, and added configure flag --with-gcj-dbtool.
If two word
values are written, a space is written between them.
A word is most Scheme values, including numbers and lists.
A Scheme string is treated as a word by write but by not
display.
A new --pedantic command-line flag.
It currently only affects the XQuery parser.
The load-compile procedure was removed.
The string printed by the --version switch
now includes the Subversion revision and date
(but only if Kawa was built using make rather than ant from a checked-out Subversion tree).
Kawa development now uses the Subversion (svn) version control system instead of CVS.
Show file/line/column on unbound symbols (both when interpreted and when compiled).
Cycles are now allowed between require'd modules.
Also, compiling at set of modules that depend on each other can now
specified on the compilation command line in any order, as long as
needed require forms are given.
The colon notation
has been generalized..
The syntax object:name
generally means to extract a component with a given name
from object, which may be an object, a class,
or a namespace.
New command-line options --debug-error-prints-stack-trace
and --debug-warning-prints-stack-trace provide stack trace
on static error messages.
The license for the Kawa software has been changed to the X11/MIT license.
A much more convenient syntax for working with Java arrays.
The same function-call syntax also works for Scheme vectors, uniform
vectors, strings, lists - and anything else that implements java.util.List.
The fields and methods of a class and its bases classes are in scope within methods of the class.
Unnamed procedures (such as lambda expressions) are printed with the source filename and line.
The numeric compare functions (=, <=, etc)
and number->string now work when passed standard Java
Number objects (such as java.lang.Long
or java.math.BigDecimal).
SRFI-10
is now implemented, providing the #,(name args ...) form.
Predefined constructor names so far are URI
and namespace.
The define-reader-ctor function is available
if you (require 'srfi-10).
A new --script option makes it easier to write
Unix shell scripts.
Allow general URLs for loading (including the -f flag),
compilation and open-input-file, if the
file name
starts with a URL scheme
like http:.
Classes defined (e.g. with define-simple-class) in
a module can now mutually reference each other.
On the other hand, you can no longer define-class if the
class extends a class rather than an interface; you must use
define-simple-class.
KawaPageServlet now automatically selects language.
provide macro.
quasisyntax and the convenience syntax #`,
from SRFI-72.
define-for-syntax, syntax-source,
syntax-line, and syntax-column,
for better compatibility with mzscheme.
SRFI-34
(Exception Handling for Programs), which implements
with-exception-handler, guard, and
raise, is now available,
if you (require 'srfi-34).
Also, SRFI-35
(Conditions) is available, if you (require 'srfi-35).
The case-lambda form from SRFI-16 is now implemented more efficiently.
SRFI-69 Basic hash tables
is now available, if you (require 'hash-table) or (require 'srfi-69).
This is an optimized and Java-compatible port whose default
hash function calls the standard hashCode method.
A define-simple-class can now have one (or more)
explicit constructor methods. These have the spcial name *init*.
You can call superclass constructors or sibling constructors
(this constructor calls) using the (admittedly verbose but
powerful) invoke-special form.
The runnable function creates
a Runnable from a Procedure.
It is implemented using the new class RunnableClosure,
which is now also used to implement future.
The kawa command can now be run in-place
from
the build directory: $build_dir/bin/kawa.
The special field name class in
(static-name type 'class) or
(prefix:.class) returns the java.lang.Class
object corresponding to the type or
prefix. This is similar to the Java syntax.
Contructing an instance (perhaps using make) of a class
defined using define-simple-class in the current module
is much more efficient, since it no longer uses reflection.
(Optimizing classes defined using define-class is
more difficult.)
The constructor function defined by the
define-record-type macro is also optimized.
You can now access instance methods using this short-hand:
(*:methodname instance arg ...)
This is equivalent to:
(invoke instance 'methodname arg ...)
You can now also access a fields using the same colon-notation as used for
accessing methods, except you write a dot before the field name:
(type:.fieldname)
;; is like: (static-field type 'fieldname).
(*:.fieldname instance)
;; is like: (field 'fieldname instance)
(type:.fieldname instance)
;; is like: (*:.fieldname (as instance type))
These all work with set! - for example:
(set! (*:.fieldname instance) value).
In the above uses of colon-notation, a type can
be any one of:
- a namespace prefix bound using define-namespace to a namespace
uri of the form "class:classname";
- a namespace prefix using define-namespace bound to a <classname> name, which can be
a fully-qualified class name or a locally-declared class, or an alias
(which might be an imported class);
- a fully qualified name of a class (that exists at compile-time),
as in (java.lang.Integer:toHexString 123); or
- a <classname> variable, for example: (<list>:list3 11 12 13).
New fluid variables *print-base*, *print-radix*,
*print-right-margin*, and *print-miser-width*
can control output formatting. (These are based on Common Lisp.)
You can new emit elipsis (...) in the output of a
syntax template using the syntax (... ...),
as in other syntax-case implementations.
The args-fold program-argument processor from
SRFI-37
is available after you (require 'args-fold) or (require 'srfi-37).
The fluid-let form now works with lexical bindings, and should
be more compatible with other Scheme implementations.
(module-export namespace:prefix) can be used to export a namespace prefix.
Static modules are now implemented more similarly to non-static modules.
Specifically, the module body is not automatically run by the class
initializer.
To get the old behavior, use the new --module-static-run flag.
Alternatively, instead of (module-static #t) use
(module-static 'init-run).
Implement SRFI-39
"Parameter-objects". These are like anonymous
fluid values and use the same implementation.
current-input-port, current-output-port,
and current-error-port
are now parameters.
Infer types of variables declared with a let.
Character comparisons (such as char-=?, char-ci<?)
implemented
much more efficiently — and (if using Java5) work for characters not
in the Basic Multilingual Plane.
Major re-write of symbol and namespace handling.
A Symbol is now
immutable,
consisting of a "print-name" and a pointer to
a Namespace (package).
An Environment
is a mapping from Symbol
to Location.
Rename Interpreter
to Language and
LispInterpreter
to LispLanguage.
Constant-time property list operations.
Namespace-prefixes are now always resolved at compile-time, never at run-time.
(define-namespace PREFIX <CLASS>) is loosely the same as
(define-namespace PREFIX "class:CLASS") but does the right thing
for classes defined in this module, including nested or non-simple classes.
Macros capture proper scope automatically, not just when using require. This allows some internal macros to become private.
Major re-write of the macro-handling and hygiene framework.
Usable support for syntax-case; in fact some of the primitives
(such as
if) are now implemented using syntax-case.
(syntax form) (or the
short-cut #!form) evaluates to a syntax object.
(define-syntax (mac x) tr)
same as (define-syntax mac (lambda (x) tr)).
The following non-hygienic forms are equivalent:
(define-macro (macro-name (param ...) transformer) (define-macro macro-name (lambda (param ...) transformer)) (defmacro macro-name (PARAM ...) transformer)Allow vectors and more general ellipsis-forms in patterns and templates.
A new configure switch --with-java-source=version
allows you to tweak
the Kawa sources to match Java compiler and libraries you're using.
The default (and how the sources are distributed)
is 2 (for "Java 2" – jdk 1.2 or better),
but you can also select "1" (for jdk 1.1.x), and
"5" for Java 5 (jdk 1.5). You can also specify a jdk version number:
"1.4.1" is equivalent to "2" (for now). Note the default source-base
is incompatible with Java 5 (or more generally JAXB 1.3 or DOM 3),
unless you also --disable-xml.
Configure argument --with-servlet[=servlet-api.jar] replaces --enable-servlet.
Function argument in error message are now numbered starting at one. Type errors now give better error messages.
A new function calling convention, used for --full-tailcalls.
A function call is split up in two parts: A match0/.../matchN
method checks that the actual arguments match the expected
formal arguments, and leaves them in the per-thread
CallContext.
Then after the calling function returns, a zero-argument apply()
methods evaluates the function body. This new convention has
long-term advantages (performance, full continuations), but the
most immediate benefit is better handling of generic (otherloaded)
functions. There are also improved error messages.
Real numbers, characters, Lisp/Scheme strings
(FString) and symbols
all now implement the Comparable interface.
In define-class/define-simple-class:
[Most of this work was funded by Merced Systems.]
access: ['private|'protected|'public|'package] to set
the Java access permissions of fields and methods.
access: 'static specifier.
invoke , field ,
static-field , slot-ref , slot-set!
can now access non-public methods/fields when appropriate.
expr in
init-form: expr is now evaluated in the outer scope.
init: expr evalues expr in the inner scope.
allocation: can now be a string literal or a
quoted symbol. The latter is preferred: allocation: 'class.
'static as a synonym for 'class following allocation:.
allocation: 'class init: expr) now
works, and is performed at class initialization time.
dummy fieldsto add initialization-time actions not tied to a field:
(define-simple-class Foo ()
(:init (perform-some-action)))
Various fixes and better error messages in number parsing. Some optimizations for the divide function.
New framework for controlling compiler warnings and other features,
supporting command-line flags, and the Scheme forms with-compile-options
and module-compile-options. The flag --warn-undefined-variable is
useful for catching typos. Implementation funded by Merced Systems.
New invoke-special syntax form (implemented by Chris Dean).
New define-variable form (similar to Common Lisp's defvar).
KawaPageServlet allows automatic loading and on-the-fly compilation in a
servlet engine. See http://www.gnu.org/software/qexo/simple-xquery-webapp.html.
The default source-base requires various Java 2 features, such as collection.
However, make-select1 will comment out Java2 dependencies, allowing you
to build Kawa with an older Java implementation.
The -f flag and the load function can take an absolute URL.
New Scheme functions load-relative and base-uri.
Imported implementation of cut and cute from SRFI-26 (Notation for Specializing Parameters without Currying).
The way top-level definitions (including Scheme procedures) are mapped into Java fields is changed to use a mostly reversible mapping. (The mapping to method names remains more natural but non-reversible.)
define-alias of types can now be exported from a module.
New --no-inline and --inline=none options.
You can use define-namespace to define namespace aliases
.
This is used for the new short-hard syntax for method invocation:
(define-namespace Int32 "class:java.lang.Integer")
(Int32:toHexString 255) => "ff"
(Int32:toString (Int32:new "00255")) => "255"
Alternatively, you can write:
(java.lang.Integer:toHexString 255) => "ff"
SRFI-9 (define-record-type) has been implemented, and compiled to a
define-class, with efficient code.
The configure option --with-collections is now the default.
Unknowns are no longer automatically static.
If type not specified in a declaration, don't infer it from it initial value.
If no return type is specified for a function, default to Object,
rather than the return type of the body. (The latter leads to
undesirable different behaviour if definitions are re-arranged.)
You can now define and use classes defined using object,
define-class, and define-simple-class
from the interpreter
, as well as the compiler.
Also, a bug where inherited fields did not get initialized has been fixed.
There are several new procedures useful for servlets.
Numerical comparisions (<, <=, etc) now generates optimized bytecode
if the types of the operands have certain known types. including
efficient code for <int>, <long>, <double>, and <integer>.
Much more code can now (with type declaration) be written just as
efficiently in Scheme as in Java.
There have been some internal re-arranging of how Expressions are processed. The Scheme-specific Translator type now inherits from Compilation, which replaces the old Parser class. A Complation is now allocated much earlier, as part of parsing, and includes a SourceMessages object. SourcesMessages now includes (default) line number, which is used by Compilation for the "current" line numbers. The ExpWalker class includes a SourceMessages instance (which it gets from the Compilation). CanInline.inline method now takes ExpWalker parameter. Checking of the number or parameters, and mapping known procedures to Java methods are now both done during the inlining pass.
The user-visible effect is that Kawa can now emit error mesages more cleanly more places; the inlining pass can be more agressive, and can emit better error messages, which yields better type information. This gives us better code with fewer warnings about unknown methods.
A new language front-end handles a tiny subset of XSLT. An example is the check-format-users test in gnu/xquery/testsuite/Makefile.
There are now converters between SAX2 and Consumer events, and a basic implementation of XMLReader based on XMLParser.
The function as-xml prints a value in XML format.
Srfi-0 (cond-expand), srfi-8 (receive), and srfi-25 (multi-dimensional arrays) are now implemented. So is srfi-1 (list library), though that requires doing (require 'list-lib).
The JEmacs code is being re-organized, splitting out the Swing-dependent code into a separate gnu.jemacs.swing package. This should make it easier to add JEmacs implementation without Swing.
The class gnu.expr.Interpreter has various new 'eval' methods that are useful for evaluating Scheme/BRL/XQuery/... expressions from Java.
Kawa now uses current versions of autoconf, autoamke, and libtool, allowing the use of automake file inclusion.
The comparisons <<, <=, -, >, and => now compile to optimized Java
arithmetic if both operands are <int> or a literal that fits in <int>.
Generated HTML and Postscrpt documents are no longer included in the
source distribution. Get kawa-doc-version.tar.gz instead.
(format #t ...) and (format PORT ...) now returns #!void instead of #t.
Support fluid bindings (fluid-let) for any thread, not just Future and main.
A Unix script header #!/PROGRAM is ignored.
You can now take the same Kawa "web" program (written in Scheme, KRL/BRL, or XQuery) and run it as either a servlet or a CGI script.
There are a number of new functions for accessing HTTP requests and generating HTTP responses.
Kawa now supports a new experimental programming KRL (the "Kawa Report Language"). You select this language using --krl on the Kawa command link. It allows Scheme code to be inside template files, like HTML pages, using a syntax based on BRL (brl.sourceforge.net). However, KRL has soem experimental changes to both BRL and standard Scheme. There is also a BRL-compatibile mode, selected using --brl, though that currently only supports a subset of BRL functions.
If language is not explicitly specified and you're running a source file (e.g. "java kawa.repl myscript.xql"), Kawa tried to derive the language from the the filename extension (e.g. "xql"). It still defaults to Scheme if there is no extension or the extension is unrecognized.
New command-line option --output-format alias --format can be used to over-ride the format used to write out top-level (repl, load) values.
XMLPrinter can now print in (non-well-formed-XML) HTML.
Changed lots of error messages to use pairs of single quotes rather than starting with a backquote (accent grave): 'name' instead of `name'. Many newer fonts make the latter look bad, so it is now discouraged.
The types <String> and <java.lang.String> new behave differently. The
type <java.lang.String> now works just like (say) <java.util.Hashtable>.
Converting an object to a <java.lang.String> is done by a simple
coercion, so the incoming value must be a java.lang.String reference
or null. The special type <String> converts any object to a
java.string.String by calling toString; it also handles null by
specially testing for it.
For convenience (and backwards compatibility) Kawa uses the type
<String> (rather than <java.lang.String>) when it sees the Java
type java.lang.String, for example in the argument to an invoke.
The default behaviour of '[' and '] was changed back to be token (word) constituents, matching R5RS and Common Lisp. However, you can easily change this behaviour using the new setBrackMode method or the defaultBracketMode static field in ReadTable.
You can now build Kawa from source using the Ant build system (from Apache's Jakarta project), as an alternative to using the traditional configure+make system. An advantage of Ant is that it works on most Java systems, without requiring a Unix shell and commands. Specifically, this makes it easy to build Kawa under MS-Windows. Thanks to James White for contributing this support.
Added (current-error-port) which does the obvious.
The new let-values and let-values* macros from srfi-11 provide a more convenient way to use multiple values.
All the abstract apply* and eval* methods now specify 'throws Throwable'. A bunch of code was changed to match. The main visible advantage is that the throw and primitive-throw procedures work for any Throwable without requiring it to be (confusingly) wrapped.
A new compilation flag --servlet generates a Servlet which can be deployed in a servlet engin like Tomcat. This is experimental, but it seesm to work for both Scheme source and XQuery source.
The interface gnu.lists.CharSequence was renamed to avoid conflitcs with the (similar) interface java.lang.CharSequence in JDK 1.4beta.
New --help option (contributed by Andreas Schlapbach).
Changed the code generation used when --full-tailcalls. It now is closer to that used by default, in that we don't generate a class for each non-inlined procedure. In both cases calling an unknown procedure involves executing a switch statement to select a method. In addition to generating fewer classes and simplifying one of the more fragile parts of Kawa, it is also a step towards how full continuations will be implemented.
Changed the convention for name "mangling" - i.e. how Scheme names are
mapped into Java names. Now, if a Scheme name is a valid Java name it
is used as is; otherwise a reversible mangling using "$" characters is
used. Thus the Scheme names '< and '$Leq are
both mapped into the same Java name "$Leq".
However, other names not containing "$" should
no longer clash, including pairs like "char-letter?" and
"charLetter?"
and "isCharLetter" which used to be all mapped to
"isCharLetter". Now
only names containing "$" can be ambiguous.
If the compiler can determine that all the operands of (+ ...) or (- ...) are floating-point, then it will generate optimized code using Java primitive arithmetic.
Guile-style keyword syntax '#:KEYWORD' is recognized. (Note this conflicts with Common Lisp syntax for uninterned symbols.)
New syntax forms define-class and define-simple-class allow you to define classes more easily. define-class supports true multiple inheritance and first class class values, where each Scheme class is compiled to a pair of an inteface and a class. define-simple-class generates more efficient and Java-compatible classes.
A new language "xquery" implements a (so far small subset of) XQuery, the draft XML Query languaage.
Various internal (Java API) changes: Changes to gnu.expr.Interpreter to make it easier to add non-Lisp-like languages; gnu.lists.Consumer now has an endAttribute method that need to be called after each attribute, rather than endAttributes that was called after all of them.
If configured with --with-gcj, Kawa builds and intalls a 'gckawa' script to simlify linking with needed libraries.
The setter function is now inlined, and
(set! (field X 'N) V) and
(set! (static-field <T> "N) V) are now inlined.
If configured --with-gcj, then a gckawa helper
script is installed, to make it easier to link Kawa+gcj-compiled applications.
The JEmacs code now depends on CommonLisp, rather than vice versa, which means Commonlisp no longer depends on Swing, and can be built with GCJ. CommonLisp and JEmacs symbols are now implemented using Binding, not String.
Kawa now installs as a .jar file (kawa.jar symlinked to kawa-VERSION.jar), rather than a collection of .class files.
The Kawa manual includes instructions for how to build Kawa using GCJ, and how to compile Scheme code to a native executable using GCJ.
Kawa now has builtin pretty-printer support, using an algorithm from Steel Bank Common Lisp converted from Lisp to Java. The high-level Common Lisp pretty-printing features are mostly not yet implemented, but the low-level support is there. The standard output and error ports default to pretty-printing.
A new formatting framework uses the Consumer interface from gnu.lists. You can associate a format with an output port. Common Lisp and JEmacs finally print using their respective syntaxes.
All output ports (OutPort instances) are now automatically flushed on program exit, using a new WriterManager helper class.
The new commmand-line option --debug-print-expr causes the Expression for each expression to be printed. The option --debug-print-final-expr is similar, but prints Expressions after optimization and just before compilation. They are printed using the new pretty-printer.
Changed calling convention for --full-tailcalls to write results to a Consumer, usually a TreeList or something to be printed. A top-level ModuleBody now uses the same CpsProcedure convention. This is useful for generating xml or html.
New libtool support allows kawa to be built as a shared library.
The new configure flag --with-gcj uses gcj to compile Kawa to both .class files and native code. This is experimental.
The reader (for Scheme and Lisp) has been re-written to be table-driven, based on the design of Common Lisp readtables.
The new gnu.lists package has new implementations of sequence-related classes. It replaces most of gnu.kawa.util. See the package.html file.
If the expected type of a non-unary + or - is <int> or <long> and
the operands are integeral types, then the operands will converted to
the primitive integer type and the addition or subtraction done
using primitive arithmetic. Similarly if the expected type is <float>
or <long> and the operands have appropriate type. This optimization
an make a big performance difference. (We still need to also optimize
compare operations like (< x y) to really benefit from <int> declarations
of loop variables.)
The implementation of procedure closures has been changed to basically be the same as top-level procedures (except when --full-tailcalls is specified): Each procedure is now an instance of a ModuleMethod, which each "frame" is an instance of ModuleBody, just like for top-level functions. This sometimes reduces the number of classes generated, but more importantly it simplifies the implementation.
A new gnu.xml package contains XML-related code, currently an XML
parser and printer, plus some XPath support. The class
gnu.lists.TreeList
(alias <document>) is useful for compactly
representing nested structures, including XML documents.
If you (require 'xml) you will get Scheme interfaces
(print-as-xml and parse-xml-from-url) to
these classes.
New package gnu.kawa.functions, for primitive functions (written in Java).
The map and for-each procedure is now inlined. This is most especially beneficial when it allows the mapped-over procedure to also be inlined, such as when that procedure is a lambda expression.
Added documentation on compiling with Jikes. Renamed some classes to avoid warning when compiling with Jikes.
The reverse! procedure was added.
Internal changes: * If a variable reference is unknown, create a Declaration instance with the IS_UNKNOWN flag to represent an imported binding. * The ExpWalker framework for "tree walking" Expressions had a bit of reorganization. * New package gnu.kawa.functions, for primitive functions (written in Java).
Added a hook for constant-folding and other optimization/inlining at traversal (ExpWalker) time. Optimization of + and - procedures to use primitive Java operations when the operands are primitive types.
Implementation of SRFI-17. Change the definitions of (set! (f x ...) val) to ((setter f) x ... val), rather then the old ((setter f) val x ...). You can now associate a setter with a procedure, either using make-procedure or set-procedure-property!. Also, (setter f) is now inlined, when possible.
Internally, Syntax (and hence Macro) no longer extend Declaration.
Various Java-level changes, which may be reflected in Scheme later: * gnu.kawa.util.Consumer interface is similar to ObjectOutput and SAX's ContentHandler interfaces. * A gnu.expr.ConsumerTarget is used when evaluating to an implicit Consumer. * These interfaces will make it easy to write functional-style but efficient code for transforming data streams, including XML. * gnu.kawa.util.FString is now variable-size.
The bare beginnings of Common Lisp support, enabled by the --commonlisp (or --clisp) command line option. This is so far little more than a hack of the EmacsLisp support, but with lexical scoping and CL-style format.
JEmacs news:
You can now specify in define and define-private the
type of a variable. If the variable is module-level,
(define name :: <type> value) creates a field named
having the specified type and initial value. (If type is not
specified, the default is not nameObject,
but rather a Binding
that contains the variable's value.)
You can now define the type of a module-level variable: In (define[-private] :: type expression) New (define-constant name [:: type] expression) definition form.
A procedure can now have arbitrary properties associated with it. Use procedure-property and set-procedure-property! to get and set them.
The new procedure make-procedure creates a generic procedure that may contain one or more methods, as well as specified properties.
New declaration form define-base-unit. Both it and define-unit have been re-implemented to be module-safe. Basically '(define-unit ft 12in)' is sugar for '(define-constant ft$unit (... (* 12 in$unit)))', where ft$unit and in$unit are standard identifiers managed by the module system. Also, the output syntax for units and quantities is cleaner.
The new declaration (module-export name ...) allows control over the names exported from a module. The new declaration (module-static ...) allows control over which definitions are static and which are non-static. This makes it easier to use a module as a Java class.
Procedures names that accidentally clash with inherited method names (such as "run") are now re-named.
Simple aliases (define-aliases defining an alias for a variable name) are implemented more efficiently.
The package hierarchy is getter cleaner, with fewer cyclic dependencies: The gnu.math package no longer has any dependencies on kawa.* or gnu.*. Two classes were moved from gnu.text to other classes, avoiding another cyclic package dependency between gnu.text and gnu.mapping. The new gnu.kawa.lispexpr is for compile-time handling of Lisp-like languages.
Compliation of literals has been re-done. A class that can be used in a literal no longer needs to be declared as Compilable. Instead, you declare it as implementaing java.io.Externalizable, and make sure it has appropriate methods.
All the standard "data" types (i.e. not procedures or ports) now implement java.io.Externalizable, and can thus be serialized. If they appear in literals, they can also be compiled.
Created a new class gnu.kawa.util.AbstractString, with the Scheme
alias <abstract-string>. The old gnu.kawa.util.FString now
extends AbstractString. A new class CharBuffer provides an
growable buffer, with markers (automatically-adjusted positions).
Many of the Scheme <string> procedures now work
on <abstract-string>.
The JEmacs BufferContnat class (contains the characters of a buffer)
now extends CharBuffer.
Some JEmacs changes to support a "mode" concept, as well as preliminary support for inferior-process and telnet modes.
New section in manual / web page for projects using Kawa.
The record feasture (make-record-type etc) how handles "funny" type and fields names that need to be "mangled" to Java names.
Re-did implementation of define-alias. For example, you can
define type-aliases:
(define-alias <marker> <gnu.jemacs.buffer.Marker>)
and then use <marker> instead of <gnu.jemacs.buffer.Marker>.
(field array 'length) now works.
Added documentation to the manual for Homogeneous numeric vector datatypes (SRFI-4).
You can now specify characters using their Unicode value: #\u05d0 is alef.
Kawa now uses a more mnemonic name mangling Scheme. For example,
a Scheme function named <= would get compiled to
method $Ls$Eq.
There is now working and useful module support, thought not all
features are implemented. The basic idea is that a module can be any
class that has a default constructor (or all of whose fields and
methods are static); the public fields and methods of such a
class are its exported definitions. Compiling a Scheme file
produces such a module. Doing:
(require <classname>)
will create an anonymous instance of <classname> (if needed), and add
all its exported definitions to the current environment. Note that if
you import a class in a module you are compiling, then an instance of
the module will be created at compile-time, and imported definitions
are not re-imported.
(For now you must compile a module, you cannot just load it.)
The define-private keyword creates a module-local definition.
New syntax to override some properties of the current module:
(module-name <name>) overrides the default name for a module.
(module-extends <class>) specifies the super-class.
(module-implements <interface> ...) specfies the implemented interfaces.
The syntax: (require 'keyword) is syntactic sugar for (require <classname>) where the classname is find is a "module catalog" (currently hard-wired). This provides compatibility with Slib. The Slib "features" gen-write, pretty-print, pprint-file, and printf are now available in Kawa; more will be added, depending on time and demand. See the package directory gnu/kawa/slib for what is available.
A lot of improvements to JEmacs (see JEmacs.SourceForge.net).
kawa-compiled-VERSION.zip is replaced by kawa-compiled-VERSION.jar.
You can now use Kawa to generate applets, using the new --applet switch, Check the "Applet compilation" section in the manual. Generating an application using the --main flag should work again. Neither --applet nor --main has Scheme hard-wired any more.
A new macro `(this)' evaluates to the "this object" - the current instance of the current class. The current implementation is incomplete, and buggy, but it will have to do for now.
The command-line argument -f FILENAME will load the same files types as load.
When a source file is compiled, the top-level definitions (procedures, variables, and macros) are compiled to final fields on the resulting class. This are not automatically entered into the current environment; instead that is the responsibility of whoever loads the compiled class. This is a major step towards a module system for Kawa.
There is a new form define-private which is like define, except that the defined name is not exported from the current module.
A procedure that has optional arguments is now typically compiled into multiple methods. If it's a top-level procedure, these will be methods in the modules "ModuleBody" class, with the same (mangled) name. The compiler can in many cases call the appropriate method directly. Usually, each method takes a fixed number of arguments, which means we save the overhead of creating an array for the arguments.
A top-level procedure declared using the form (define (NAME ARS ...) BODY ..) is assumed to be "constant" if it isn't assigned to in the current compilation unit. A call in the same compilation unit will now be implemented as a direct method call. This is not done if the prcedure is declared with the form: (define NAME (lambda (ARGS ,,,) BODY ...)
gnu.expr.Declaration no longer inherits from gnu.bytecode.Variable.
A gnu.mapping.Environment now resolves hash index collisions using "double hashing" and "open addressing" instead of "chaining" through Binding. This allows a Binding to appear in multiple Environments.
The classes Sequence, Pair, PairWithPosition, FString, and Char were moved from kawa.lang to the new package gnu.kawa.util. It seems that these classes (except perhaps Char) belong together. The classes List and Vector were also moved, and at the same time renamed to LList and FVector, respectively, to avoid clashed with classes in java.util.
New data types and procedures for "uniform vectors" of primitive types were implemented. These follow the SRFI-4 specification, which you can find at http://srfi.schemers.org/srfi-4/srfi-4.html .
You can now use the syntax name :: type to specify the type of a parameter.
For example:
(define (vector-length x :: <vector>) (invoke x 'length))
The following also works:
(define (vector-length (x :: <vector>)) ...).
(define-member-alias name object [fname]) is new syntactic sugar
for (define-alias name (field object fname)), where the default for
fname is the mangling of name.
The new function `invoke' allows you to call a Java method. All of `invoke', `invoke-static' and `make' now select the bets method. They are also inlined at compile time in many cases. Specifically, if there is a method known to be definitely applicable, based on compile-time types of the argument expressions, the compiler will choose the most specific such method.
The functions slot-ref, slot-set!, field, and static-field are now inlined by the compiler when it can.
Added open-input-string, open-output-string, get-output-string from SRFI-6. See http://srfi.schemers.org/srfi-6/srfi-6.html.
The manual has a new section "Mapping Scheme names to Java names", and a new chapter "Types". The chapters "Extensions", "Objects and Classes", and "Low-level functions" have been extensivley re-organized.
The Kawa license has been simplified. There used to be two licenses: One for the packages gnu.*, and one for the packages kawa.*. There latter has been replaced by the former. The "License" section of the manual was also improved.
There is a new package gnu.kawa.reflect. Some classes that used to be in kawa.lang or kawa.standard are now there.
The procedures slot-ref and slot-set! are now available. They are equivalent to the existing `field', but reading a field `x' will look for `getX' method if there is no public `x' field; writing to a field will look for `setX'.
The procedure `make' makes it convenient to create new objects.
There is now a teaser screen snapshot of "JEmacs" at http://www.bothner.com/~per/papers/jemacs.png.
The html version of the manual now has a primitive index. The manual has been slightly re-organized, with a new "Classes and Objects" chapter.
The new functions invoke-static and class-methods allow you to call an arbitary Java method. They both take a class specification and a method name. The result of class-methods is a generic procedure consisting of those methods whose names match. (Instance methods are also matched; they are treated the asme as class methods with an extra initial argument.) The invoke-static function also takes extra arguments, and actually calls the "best"-matching method. An example:
(invoke-static <java.lang.Thread> 'sleep 100)
Many fewer classes are now generated when compiling a Scheme file. It used to be that each top-level procedure got compiled to its own class; that is no longer the case. The change should lead to faster startup and less resource use, but procedure application will probably be noticably slower (though not so much slower as when reflection is used). The reason for the slowdown is that we in the general case now do an extra method call, plus a not-yet-optimized switch statement. This change is part of the new Kawa module system. That will allow the compiler to substitute direct methods calls in more cases, which I hope will more than make up for the slowdown.
A Scheme procedure is now in general compiled to a Java method whose name is a "mangling" of the Scheme procedure's name. If the procedure takes a variable number of parameters, then "$V" is added to the name; this indicates that the last argument is a Java array containing the rest of the arguments. Conversely, calling a Java method whose name ends in "$V" passes any excess arguments in the last argument, which must be an array type.
Many changes to the "Emacs-emulation" library in gnu.jemacs.buffer: * Implemented commands to read and save files. * We ask for file and buffer names using a dialog pop-up window. * Split windows correctly, so that the windows that are not split keep their sizes, the windows being split gets split as specified, and the frame does not change size. Now also handles horizonal splits. * Fairly good support for buffer-local keymaps and Emacs-style keymap search order. A new class BufferKeymap manages the active keymaps of a buffer. Multi-key key-sequences are handled. Pending prefix keys are remembered on a per-buffer basis (whereas Emacs does it globally).
There is now some low-level support for generic procedures.
The R5RS primitives let-syntax and letrec-syntax for defining local syntax extensions (macros) should now work. Also define-syntax works as an internal definition. All of these should now be properly "hygienic". (There is one known exception: symbols listed among the literals lists are matched as raw symbols, rather that checking that the symbol has the same binding, if any, as at the defining site.) The plan is to support general functions as hygienic rewriters, as in the Chez Scheme "syntax-case" system; as one part of that plan, the syntax-case primitive is available, but so far without any of the supporting machinary to support hygiene.
The read-line procedure was added. This allows you to efficiently read a line from an input port. The interface is the same as scsh and Guile.
define-alias now works both top-level and inside a function.
Optimized eqv? so if one of the arguments is constant and not Char or Numeric, inline it the same way eq? is. (This helps case when the labels are symbols, which help the "lattice" benchmark.) ???
The Emacs-related packages are now grouped under a new gnu.jemacs package.
Improved framework for catching errors. This means improved error messages when passing a parameter of the wrong type. Many standard procedures have been improved.
Simplified, documented, and tested (!) procedure for building Kawa from source under Windows (95/98/NT).
New macros trace and untrace for tracing procedures. After executing (trace PROCEDURE), debugging output will be written (to the standard error port) every time PROCEDURE is called, with the parameters and return value. Use (untrace PROCEDURE) to turn tracing off.
New utility functions (system-tmpdir) and (make-temporary-file [format]).
A new (unfinished) framework supports multiple languages. The command-line option --elisp selects Emacs Lisp, while --scheme (the default) selects Scheme. (The only difference so far is the reader syntax; that will change.)
The `format' function now provides fairly complete functionality
for CommonLisp-style formatting. (See the Comon Lisp hyperspec at
http://www.harlequin.com/education/books/HyperSpec/Body/sec_22-3.html.)
The floating point formatters (~F, ~E, ~G, ~$) now pass the formatst.scm
test (from Slib, but with some "fixes"; in the testsuite directory). Also,
output ports now track column numbers, so ~T and ~& also work correctly.
A new package gnu.emacs provides various building blocks for building an Emacs-like text editor. These classes are only compiled when Kawa is configured with the new --with-swing configuration option. This is a large initial step towards "JEmacs" - an Emacs re-implemented to use Kawa, Java, and Swing, but with full support (using gnu.elisp) for traditional Emacs Lisp. For more imformation see gnu/emacs/overview.html.
A new configuration option --with-swing can be used if Swing is available. It is currently only used in gnu.emacs, but that may change.
Kawa is now "properly tail-recursive" if you invoke it with the --full-tail-calls flag. (Exception: the eval procedure does not perform proper tail calls, in violation of R5RS. This will be fixed in a future release.) Code compiled when --full-tail-calls is in effect is also properly tail-recursive. Procedures compiled with --full-tail-calls can call procedures compiled without it, and vice versa (but of course without doing proper tail calls). The default is still --no-full-tail-calls, partly because of performance concerns, partly because that provides better compatibility with Java conventions and tools.
The keywords let (including named let), let*, and letrec support type specifiers for the declared variables For example:
(let ((lst :: <list> (foo x))) (reverse lst))
Square brackets [ ... ] are allowed as a synonym of parentheses ( ... ).
A new command-line flag --server PORT specifies that Kawa should run as a telnet server on the specified PORT, creating a new read-eval-print loop for each connection. This allows you to connect using any telnet client program to a remote "Kawa server".
A new front-end program, written in C, that provides editing of input lines, using the GNU readline library. This is a friendlier interface than the plain "java kawa.repl". However, because kawa.c needs readline and suitable networking library support, it is not built by default, but only when you configure Kawa with the --enable-kawa-frontend flag.
The way Scheme names are mapped ("mangled") into Java identifiers is now more natural. E.g. "foo-bar?" now is mapped to "isFooBar".
New syntax (object (SUPERS ...) FIELD-AND-METHODS ...) for creating a new object instance of an anonymous class. Now fairly powerful.
New procedures field and static-field for more convenient field access.
Syntactic sugar: (lambda args <type> body) -> (lambda args (as <type> body)).
This is especially useful for declaring methods in classes.
A new synchonized form allows you to synchronize on an arbitrary Java object, and execute some forms while having an exclusive lock on the object. (The syntax matches that used by Skij.)
New --debug-dump-zip option writes out a .zip file for compilation. (Useful for debugging Kawa.)
You can now declare parameter types.
Lot of work on more efficient procedure representation and calling convention: Inlining, directly callable statics method, plus some procedures no longer generate a separate Class.
Local functions that are only called from one locations, except for tail-recursion, are now inlined. This inlines do loops, and most "named let" loops.
New representation of closures (closures with captured local variables). We no longer use an array for the closure. Instead we store the captured variables in the Procedure itself. This should be faster (since we can use field accesses rather than array indexing, which requires bounds checking), and avoids a separate environment object.
If the compiler sees a function call whose (non-lexically-bound) name matches an existing (globally-defined) procedure, and that procedure instance has a static method named either "apply" or the mangled procedure name, them the compiler emits a direct call to that method. This can make a very noticable speed difference, though it may violate strict Scheme sementics, and some code may break.
Partial support for first-class "location" variables.
Created new packages gnu.mapping and gnu.expr. Many classes were moved from kawa.lang to the new packages. (This is part of the long-term process of splitting Kawa into more manageable chunks, separating the Scheme-specific code from the language-independent code, and moving classes under the gnu hierarchy.)
You can now write keywords with the colon first (e.g. :KEYWORD), which has exactly the same effect and meaning as putting the colon last (e.g. KEYWORD:). The latter is preferred is being more consistent with normal English use of punctuation, but the former is allowed for compatibility with soem other Scheme implementations and Common Lisp.
The new package gnu.text contains facilities for reading, formatting, and manipulating text. Some classes in kawa.lang where moved to there.
Added string-upcase!, string-downcase!, string-capitalize!, string-upcase, string-downcase, and string-capitalize; compatible with Slib.
Character constants can now use octal notation (as in Guile). Writing a character uses octal format when that seems best.
A format function, similar to that in Common Lisp (and Slib) has been added.
The default parameter of a #!optional or #!key parameter can now be #!null.
The "record" feature has been changed to that a "record-type descriptor"
is now a gnu.bytecode.ClassType (a <record-type>), rather than a
java.lang.Class. Thus make-record-type now returns a <record-typee>,
not a Class, and record-type-descriptor takes a <record-typee>,
not a Class.
More robust Eval interfaces.
New Lexer abstract class. New ScmRead class (which extends Lexer) now contains the Scheme reader (moved from Inport). Now read errors are kept in queue, and can be recovered from.
Comparing an exact rational and an inexact real (double) is now done as if by first converting the double to exact, to satisfy R5RS.
The compile virtual method in Expression now takes a Target object, representing the "destination". The special ConditionalTarget is used to evaluate the test of an 'if expression. This allows us to generate much better code for and, or, eq?, not and nested if inside an if.
Added port-line, port-column, and set-port-line! to match Guile.
The Makefiles have been written so all out-of-date .java (or .scm). files in a directory are compiled using a single invocation of javac (or kawa). Building Kawa should now be much faster. (But note that this depends on unreleased recent autoamke changes.)
How the Kawa version number is compiled into Kawa was changed to make it easier for people who want to build from source on non-Unix-like systems.
A new gnu.ecmascript package contains an extremely incomplete implementation of ECMSScript, the ECMA standardized version of JavaScript. It includes an ECMAScript lexer (basically complete), parser (the framework is there but most of the language is missing), incomplete expression evaluation, and a read-eval-print-loop (for testing only).
Improved Kawa home page with extra links, pointer to Java-generated api docs, and homepages for gnu.math and gnu.bytecode.
Implemented system, make-process, and some related procedures.
Added macros for primitive access to object fields, static fields, and Java arrays. Added constant-fold syntax, and used it for the other macros.
The --main flag compiles Scheme code to an application (containing a main method), which can be be invoked directly by a Java interpreter.
Implemented --version (following GNU standards) as kawa.repl command-line flag.
Adding make procedure to create new objects/records.
Extended (set! (f . args) value) to be equivalent to ((setter f) value . args). Implemented setter, as well as (setter car) and (setter cdr).
Can now get and set a record field value using an application: (rec 'fname) gets the value of the field named fname in record rec. (set! (rec 'fname) value) sets the value of the field named fname in rec.
A partial re-write of the implementation of input ports and the Scheme reader, to fix some problems, add some features, and improve performance.
Compiled .class files are now installed in $(datadir)/java, rather than $(prefix)/java. By default, that means they are installed in /usr/local/shared/java, rather than /usr/local/java.
There is now internal infrastructure to support inlining of procedures, and general procedure-specific optimized code generation.
There is better testing that the right number of arguments are passed to a procedure, and better error messages when you don't. If the procedure is inlined, you get a compile-time error message.
The functions created by primitive-constructor, primitive-virtual-method, primitive-static-method, and primitive-interface-method are now first-class procedure values. They use the Java reflection facily, except when the compiler can directly inline them (in which case it generates the same efficient bytecodes as before).
New functions instance? (tests type membership) and as (converts).
The kawa.html is now split into several files, one per chapter. The table of contents is now kawa_toc.html.
The syntactic form try-catch provides low-level exception handler support. It is basically the same as Java's try/catch form, but in Scheme syntax. The new procedure primitive-throw throws an exception object.
The higher-level catch and throw procedures implement exception handling where the handler is specified with a "key" (a symbol). These functions were taken from Guile.
The error function has been generalized to take multiple arguments (as in Guile). It is now a wrapper around (throw 'misc-error ...).
There is a new "friendly" GUI access to the Kawa command-line. If you invoke kawa.repl with the -w flag, a new interaction window is created. This is uses the AWT TextArea class. You can create multiple "consoles". They can either share top-level enevironments, or have separate environments. This window interface has some nice features, including editing. Added a scheme-window procedure, which is another way to create a window.
The default prompt now shows continuations lines differently.
The copy-file function was added.
The variable port-char-encoding controls how external files are converted to/from internal Unicode characters. It also controls whether CR and CR-LF are converted to LF.
The reader by default no longer down-cases letters in symbols. A new variable symbol-read-case control how case is handled: 'P (the default) preserves case; 'U upper-cases letters; 'D or -" down-cases letters; and 'I inverts case.
The gnu.bytecode package now supports exception handlers. The new syntactic form try-finally supports a cleanup hook that is run after some other code finishes (normally or abnormally). Try-finally is used to implement dynamic-wind and fluid-let.
The environment handling has been improved to support thread-specific environments, a thread-safe fluid-let, and multiple top-levels. (The latter still needs a bit of work.)
The gnu.bytecode package has been extensively changed. There are new classes representing the various standard Attributes, and data associated with an attribute is now stored there.
Added new procedures environment-bound? and scheme-implementation-version.
Scheme symbols are represented as java.lang.String objects. Interned symbols are interned Strings; uninterned symbols are uninterned Strings. Note that Java strings literals are automatically interned in JDK 1.1. This change makes symbols slightly more efficient, and moves Kawa closer to Java.
Ports now use the JDK 1.1 character-based Reader and Writer classes, rather than the byte-oriented InputStream and OutputStream classes. This supports different reading and writing different character encodings [in theory - there is no support yet for other than Ascii or binary files].
An interactive input port now has a prompt function associated with it. It is settable with set-input-port-prompter!. The prompt function takes one argument (the input port), and returns a prompt string. There are also user functions for inquiring about the current line and column number of an input port.
The R4RS procedures transcript-on and transcript-off are implemented.
Standard types can be referred to using syntax similar to RScheme.
For example Scheme strings now have the type <string> which is
preferred to "kawa.lang.FString" (which in addition to being longer,
is also more suspectible to changes in internal implementation).
Though these types are first-class values, this is so far mainly
useful for invoking primitive methods.
Execute a ~/.kawarc.scm file on startup, if it exists.
Add a number of functions for testing, renaming, and deleting files. These are meant to be compatible with scsh, Guile, and MIT Scheme: file-exists?, file-directory?, file-readable?, file-writable?, delete-file, rename-file, create-diretory, and the variable home-directory.
Fixed some small bugs, mainly in gnu.math and in load.
Generalize apply to accept an arbitrary Sequence, or a primitive Java array.
The codegen package has been renamed gnu.bytecode. The kawa.math package has been moved to gnu.math. Both packages have new license: No restrictions if you use an unmodified release, but GNU General Public License. Let me know if that causes problems. The rest of Kawa still has the old license.
Implement defmacro and gentemp.
Implement make-record-type and related functions to create and use new record types. A record type is implemented as a java.lang.Class object, and this feature depends on the new reflection features of JDK 1.1.
Implement keywords, and extend lambda parameter lists to support #!optional #!rest and #!keyword parameters (following DSSSL).
Added more primitives to call arbitrary interface and constructor methods.
Added primitives to make it easy to call arbitrary Java methods from Scheme.
Exact rational arithetic is now fully implemented. All integer functions now believed to correctly handle bignums. Logical operations on exact integers have been implemented. These include all the logical functions from Guile.
Complex numbers are implemented (except {,a}{sin,cos,tan}). Quantities (with units) are implemented (as in DSSSL).
Eval is available, as specified for R5RS. Also implemented are scheme-report-environment, null-environment, and interaction-environment.
Internal define is implemented.
Rough support for multiple threads is implemented.
Moved kawa class to kawa/repl. Merged in kawac (compiler) functionality. A 'kawa' shell-script is now created. This is now the preferred interface to both the interactive evaluator and the compiler (on Unix-like systems).
Now builds "without a snag" using Cafe 1.51 under Win95.
(Symantec JIT (ver 2.00b19) requires disabling JIT -
JAVA_COMPCMD=disable.)
Compiles under JDK 1.1 beta (with some warnings).
A testsuite (and testing framework) was added.
Documentation moved to doc directory. There is now an internals overview, in doc/kawa-tour.ps.
The numeric classes have been re-written. There is partial support for bignums (infinite-precision integers), but divide (for example) has not been implemented yet. The representation of bignums uses 2's complement, where the "big digits" are laid out so as to be compatible with the mpn functions of the GNU Multi-Precision library (gmp). (The intent is that a future version of Kawa will support an option to use gmp native functions for speed.)
The kawa application takes a number of useful command-line switches.
Basically all of R4RS has been implemented. All the essential forms and functions are implemented. Almost all of the optional forms are implemented. The exceptions are transcript-on, transcript-off, and the functions for complex numbers, and fractions (exact non-integer rationals).
Loading a source file with load now wraps the entire file in a lambda (named "atFileLevel"). This is for better error reporting, and consistency with compile-file.
The hygienic macros described in the appendix to R4RS are now impemented (but only the define-syntax form). They are used to implement the standard "do" form.
The R5RS multiple value functions values and call-with-values are implemented.
Macros (and primitive syntax) can now be autoloaded as well as procedures.
New kawac application compiles to one or more .class files.
Compile time errors include line numbers. Uncaught exceptions cause a stack trace that includes .scm line numbers. This makes it more practical to debug Kawa with a Java debugger.
Quasiquotation is implemented.
Various minor bug fixes and optimizations.
The biggest single change is that Scheme procedures are now compiled to Java bytecodes. This is mainly for efficiency, but it also allows us to do tail-recursion-elimination in some cases.
The "codegen" library is included. This is a toolkit that handles most of the details needed to generate Java bytecode (.class) files.
The internal structure of Kawa has been extensively re-written, especially how syntax transforms, eval, and apply are done, largely due to the needs for compilation.
Almost all the R4RS procedures are now implemented, except that there are still large gaps in Section 6.5 "Numbers".