Kawa/TODO

A GRAB-BAG OF IDEAS AND THINGS TO DO, SOME QUITE OLD

Consider using Unsafe.defineAnonymousClass

Optimize splicing of methods calls.
Remove createVarargsArrayIfNeeded logic from PrimProcedure.

SRFI-46

KawaScriptEngine should implement javax.script.Invocable

Use Parboiled as parser library?  https://github.com/sirthias/parboiled/wiki

Before Kawa 1.12:
- Jemacs type ctrl-C --> TooLongAction
- Evaluate reported JEmacs bugs.
- Go through remaining "news" entries and update manual.
- go through Evaluating-Scheme-expressions-from-Java.
- gnu/bytecode:
-- Document emitIf../emitThen else, at overview level.
- Verify that Path is consistent with new javax.nio (JDR 203)
  http://openjdk.java.net/projects/nio
- internals: Update about procedures, inlining, tail-calls.

HTTP/SERVLETS
* Fix formatting: Language format -> HttpPrinter
  #t should print correctly
  BRL should by default not HTML-quote strings
* Think about REST.
* Support explicit and implicit require (BRL's sitedefs.scm).
* Module should not be static - vars should be session variables.
* Dynamic Environment should be per-session.
* Get repl.xml working reliably.
* Annotations ...
- In ReplDocument don't use SwingContent.

OPTIMIZATIONS
* optimize fluid-let and fluid references:
  - in fluid-let call getLocation before calling setWithSave so we don't
  have to call it again in setRestore.
  - In local references inside a fluid-let use the location from above.
* Optimize (make T) where T is a known non-simple class.
  This should not need to use reflection.
* optimize zero? etc
* other support for other single-element struct types, including unsigned int
* warning about String conversions
(See email from helmut Eller 2009/09/26)
* warning about non-tail-calls
(See email from helmut Eller 2009/09/26)

HIGH PRIORITY:
* Allow co-variant return types.
* compile-file doesn't display warnings
* When deferring imports (require) the 'run' actions shouldn't be deferred. FIXME
* Optimize and/or in the case of ConditionalTarget or <boolean> target.
* properties of symbols
* Compilation: not implemented mustConvert restarg
* XQuery distinguish qexo: kawa: fn: default function namespaces.
* XQuery: Implement/document external functions.
* Name mangling, document library modules.
* Output formatting of empty arrays? (make-array (shape 0 0 0 3))
* Format filter chain API?

Localization support: gettext - see doc-resource.patch

Remove pretty-print implementation in slib.

Optimize NumberCompare further.

BRL/KRL:
* sitedefs.scm should be loaded
* brl function should write to output, not be identity.
The default output-pourt shold be re-directed.
It should be the same as the ctx.consumer.

* Resolve whether #!/FILENAME can be followed by embedded options (meta-arg).
* CL reader interface
* More define-class functionality?
* Finish modules: define-module?  constructor arguments?
* Handle defmacro definition and use in same source file.
* load.loadSource if (true) should be if (false).
* Implement Scheme API for serialization.
* Document in internals.html how ModuleMethod/ModuleBody works.

Optimize to avoid creating frame classes:
If a local variable is captured, and all functions intermediate between
the capture site and the definition are call-only functions, and the
local variable is never written (except before all calls to these
intermediate functions), then add the variable as an extra implicit
parameter to thse intermediate functions, instead of storing the
variable in a frame.

Should inline when accessing a final static field (using static-field),
for int (or shorter) types - but only for "compile-time constants".
(Needs checking .class file that field is given ConstantValue.)

Starting (emacs) has wrong keybindings unless --elisp is specified.

Move kawa.lang.Lambda to gnu.kawa.lispexpr.Lambda.  Merge with Elisp version.
Remove Scheme.getTypeValue.

(define plus +) ... (plus ...) should not inline call to +.

Implement IntRange class.
Use it to replace Char's hashtable.

Modules:
* Infer parameter and return types when defaulted and otherwise match
method in explicit module-extends or module-interface.

compile-file should generate .jar, not .zip.
Should perhaps take multiple file names.
Update manual to removing .zip references.

Binding for imported unknown names (idN$NAME) should not be static!

(this) needs to be more general/robust.
Start by forcing it to make the generated method non-static.
What about 'super' ?

Re-store code to check argument count when applyN is involved.
(in Compilation and maybe ApplyExp).

BindingInitializer.emit: global Environment should be cached

TO BE DONE FOR 2.0:

'object': being able to pass constructor arguments. eg., to simulate:
   new javax.swing.JDialog(myFrame) {  public void foo() {} };
Perhaps:  (object ((<javax.swing.JDialog> myFrame)) ...)

Generate better code for `(and test1 test2 test3)'.
[See maybe-report-section in testing.scm]

Module compilation:
* Allocated selector indexes "densely" within each applyX method
of a ModuleBody, so tableswitch gets used, rather than lookupswitch.

define-syntax and friends are not properly hygienic when checking
literals:  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.

Complain if syntax or macro is used in other than function-call position.

Let LineBufferedReader pos, limit, buffer be private,
but provide methods to get/set them.

exact->inexact only implemented for RealNums, not other Complex.

Fix scheme-window so #!eof closes window.

Combine primitive field with record facility and define-class.

Provide getopt-long (Guile, check scsh).

Implement full call-cc by re-writing.
This would be slower, so should be optional, probably not the default.

Implement a pretty-printing AWT container.

Complete R4RS procedures.
  trig functions for complex numbers

Complete R5RS number syntax.

Implement system open-input-pipe open-output-pipe close-pipe.

Add a way to specify a load-path (or maybe just use CLASSPATH?).
First seach for loaded files in directory where current file was found.

OTHER:

Provide a way to compile a Scheme program into an Applet class (with
init/start/stor/pain/etc methods).  Also provide a stripped down
"kawa-run.zip", without eval, load, or compiler.

Update to match kawa-tour:
  Vector -> FVector
  SyntaxRule.match should use Pattern.

Should SyntaxRule inherit from Pattern?

Move readSchemeObject readSchemeNumber etc from InPort
to new class(es) Parser:
	class Parser;
	class SchemeReader extends Parser;  // Provides read
	class SchemeParser extends SchemeReader;  // Provides read+rewrite
Move Interpreter.syntaxError to Parser.
Treat ReadError as SyntaxError.
Support warning, error, fatal.
Parser accumulates list of syntax errors.
When done, if errors or warnings, throw SyntaxErrors.
Printing SyntaxError prints error log.

Make closure representation consistent with inner classes wrt
use of Attributes (for possible improved debugging).

Optionally issue a warning if a symbol is rebound.

Inline calls to simple methods, especially builtins.
Ties in with module specifications (to determine which bindings are builtin).

Quantities:
Arithmetic involving Units should return Units.
Printing a Unit should return something readable, maybe:
  (quantity->unit 1cm) -> "1cm". while 1cm -> "1.0cm".

EVENT CALLBACKS - Proposed design

Syntax:  (listener (EXTRA-INTERFACE ...)
           ((METHOD-NAME+ [EVENT-TYPE])+ ARG ACTIONS ...) ...)

Example:  (invoke component 'addMouseListener
	    (listener ((mousePressed mouseReleases) evt
              (display (list "mouse pressed or released:" evt))))

This creates a new "listener" instance.  The instance has one
method for each METHOD-NAME; the method takes a single argument
of the specified EVENT-TYPE.  The EVENT-TYPE can be left out
in the case of standard java METHOD-NAMES;  for example
if the METHOD-NAME is mousePressed, the implied EVENT-TYPE
is <java.awt.event.MouseEvent>.  When the method is executed,
the actual Event is bound to ARG, and the ACTIONS executed.
(The type is ARG is the least common super-type of all
the explicit or implicit EVENT-TYPES for corresponding METHOD-NAMES.)
The allocated listener instance implements all the interfaces
listed in the EXTRA-INTERFACE list;  if addition, if an EVENT-TYPE
was implied, the corresponding Listener interface is also implied.
(For example a METHOD-NAME of mousePressed implies an EVENT-TYPE
of <java.awt.event.MouseEvent>, and hence an interface of
<java.awt.event.MouseListener>.)
New syntax:
(listener [(EXTRA-INTERFACE ...)]
  ((METHOD-NAME ... EVENT-ARG-NAME [:: EVENT-TYPE]) ...))

Listeners on JComponent etc:
(javax.swing.JButton text: "Click Me" item-state-changed: (lambda (e) ...))
It KEYWORD does not match a setter:
Make list of all 1-argument methods that match addXxxListener.
Look at parameter type - assume an interface type Yyy or abstract class.
See if there is a method named KEYWORD (after mangling) in type Yyy.
If so add to "handler list".
If "handler list" is non-empty, create an anonymous "handler" class.
(If any Yyy is an abstract class, create separate handler classes for each.)
Handler class implements all the Yyy interfaces.
Add methods for each argument.  Add any other void non-implemented methods.
For example:
  (object (ItemListener)
    ((itemStateChange $e::ItemEvent) ((lambda (e) ...) $e)))
Then: call addItemListener.
Issue: (set! foo:item-state-changed (lambda (e) ...new handler...))
In that case we want to remove old handler before adding new one.
Can use getItemListeners followed by removedItemListener.
Problem is that this may remove other unrelated handlers,
unless we have one handler only for each method.
Or have the handler have a set of settable slots.

SYMBOLS
If we have pre-process environments, (define exit ...) should only re-define
exit in the current environment, not all.

JEMACS

clean up function-binding hacks

vertical-motion

Make it so that JEmacs does not require Swing.  There is a start in
that gnu.jemacs.buffer mostly contains toolkit-independentcode, and
gnu.jemacs.swing contains Swing-specific implementations.  This
separation shoudl be finished, and we should add one or more
implementation on other toolskits:
- using Gnome2 (http://developer.gnome.org/doc/API/2.0/gtk/textwidget.html)
text widgets with Pango (http://developer.gnome.org/doc/API/2.0/pango/);
- one that uses the SWT toolkit (http://www.eclipse.org/articles/) 
that is part of the Eclipse project (http://www.eclipse.org);
- one that uses plain AWT;
- and one that works on plain terminals, perhaps using charva
(http://www.pitman.co.za/projects/charva/).
It is important that JEmacs be able to run well using Free Software only,
and it cannot do hat as long it depends on Swing, until there is a
free re-implementation of a suitable Swing subset.
Note that the Buffer class should be a pure "model" which is toolkit
independent:  we want to be able to display the same buffer in both
a Swing Window and a telnet window at the same time.

The JEmacs widgets should be integrated into the gnu/kawa/slib/gui.scm
design.

Should implement Common Lisp-style "type specifiers" and declarations.
This would make it practical to write-lower level code that calls Java
methods directly in ELisp (and Common Lisp), without having to use Scheme.

BENCHMARKS
http://www.ccs.neu.edu/home/will/Twobit/kvwbenchmarks.html