Changes in 2.0 beta 5
- Persistent Scriptable Classes! (Experimental Feature)
Using the system property -DsaveClasses= you can instruct
BeanShell to write out persistent class files for class definitions found
in the script. In this mode BeanShell will not execute code, but will
generate and store a class files for each class definition it finds.
These classes are the same as the ordinary runtime generated delegate
classes used internally by BeanShell but additionally have static
initializer code that allows them to bootstrap an instance of the BeanShell
interpreter to execute their scripted body if the class is used from
an external application.
Currently we are using a simple convention whereby a class "Foo.class"
expects to find its associated scripted class definition in a file
"Foo.bsh" in the same location as the class file. Each stored class
must have a corresponding script definition. The scripts itself may
contain any amount of BeanShell code beyond the class definition
and may refer to other scripted classes or saved classes. However each
saved class will currently initialize its own interpreter instance.
So, for example, to create a class for the following scripted class
definition:
class Foo {
Foo() { print("I'm being constructed!"); }
}
you could place the script into a file "Foo.bsh" in the current directory
and execute:
java -DsaveClasses=. bsh.Interpreter Foo.bsh
This will create the class Foo.class. The Foo class may be used like any
ordinary Java class and invoked from ordinary Java code. Upon being
loaded it will automatically create a BeanShell interpreter instance
to load and execute its corresponding Foo.bsh script file. The script
is expected to correspond to the scructure of the class: to initialize
static instance members and implement all static and instancee methods
required by the class. Beyond that, the script is free to change and
include additional script body outside the scripted class definition.
For example, we might later change the Foo.bsh script to the following:
welcomeMessage = "Hello World!";
class Foo {
Foo() { print( "I'm being constructed: "+ welcomeMessage ); }
}
Here's another example showing loose code in the script:
// Begin script Foo.bsh
class Foo {
public static void main( String [] args ) { }
}
message = "Hey";
message += " you!";
num = 2 * 2;
message += " The answer is: "+num;
// end script Foo.bsh
now we have a script that can be launched like any other Java class
via 'java Foo'. The loose script body is run upon class initialization.
Moving that text into the class body would cause it to be run on class
construction (instance creation... along with any constructor).
- Added JSR223 javax.script adapter engine implementation in a new
bsh.engine package.
This implementation of the spec supports pluggable ScriptContext. Please
see the TestBshScriptEngine.java file (and the spec) for a starting point
on how to use this.
BeanShell maintains all of its info on variables (type, modifiers,
etc.) and all of its internal state for a given ScriptContext via a
BeanShell NameSpace embedded in the context. This allows us to be
completely pluggable with respect to the spec, but still keep the full
interpreter state between evaluations.
- Added a new class, ExternalNameSpace to the distribution.
This is a BeanShell namespace that knows how to externalize its variables
to a standard Map. This class is, of necessity, dependent on JDK1.2+,
however there is no dependency in the core on this class, so you can
exclude it in the usual way.
- Fixed static import syntax. It's now the same as Java 5:
import static Foo.*;
(previously was "static import ..." )
- Static blocks in scripted classes are now working.
Added test case.
Changes in 2.0 beta 4
- Source compiles under JDK 1.5
- Better defined classloading precedence:
in-script evaluated class (scripted class)
in-script added / modified classpath
optionally, external classloader
optionally, thread context classloader
plain Class.forName()
source class (.java file in classpath)
- Method selection code has been rewritten to move towards Java 5 rules.
This should correct a number of problems where inaccessible methods are
selected and also correct security problems in applets and other secure
environments. Method selection should no longer call getDeclaredMethods()
unless accessibility is explicitly activated. The same is true for
constructors and fields.
- Updated applet demos (http://beanshell.org/demo). JConsole and desktop
now run in untrusted environment, with limitations. Applets cannot
create classloaders, so BeanShell cannot script classes in the unsigned
environment. These demos are very rough and need some work.
- The show() command now works in both interactive and non-interactive
modes. This means that for experimentation, instead of doing things like:
print( myObject instanceof Blah );
print( myObject.getFoo() );
print( myObject.somethingElse() );
You can simply do:
show();
myObject instanceof Blah;
myObject.getFoo();
myObject.somethingElse();
This can be combined with the trace property -Dtrace=true (currently a
static, interpreter wide flag) to display line by line execution and
results.
The bsh.show variable is gone and has been replaced with a property
on the interpreter.
- Fixed bug in super.method() resolution inside nested blocks.
- Added workaround for load() command to use the BshClassManager for
deserializing classes.
- Fixed some cases where SecurityException could not be caught by scripts
and handled.
Changes in 2.0 beta 3
String literals are now interned, as per JSL ch3 10.5.
Changes in 2.0 beta 2
Added code to address the class redefinition problem. BeanShell will now
use the bsh classloader to allow the classes to be reloaded as necessary.
Note: this only affects users who are utilizing the new BeanShell 2.0+
features to script classes and interfaces using the standard Java
class/interface syntax. There is no problem with scripting interfaces
using bsh objects or the anonymous inner class style syntax that has always
been supported.
Changes in 2.0 beta 1
- With release 2.0 BeanShell supports the full Java syntax.
You can now script Java classes that serve as real Java types.
Scripted classes may extend ordinary Java classes, abstract classes, and
implement interfaces as usual. Constructors and methods invoked on the
scripted classes are delegated to the BeanShell interpreter. This
means that within scripted classes you may use the full power of
BeanShell scripting including commands, loose types, and other BeanShell
Java extensions such as method closures if you wish.
BeanShell will now attempt to load classes from ".java" source files in
the classpath if the compiled class is not found.
Current Limitations:
- In order to run a class file's main method with arguments from the
command line the class containing your main method must be the last
item in the file. e.g. a .java file with the public class appearing
last.
- Static final fields (constants) in interfaces are currently broken.
- Non-static inner classes are currently implemented in a non-standard
way (using mix-ins) so mixing interpreted classes with their
pre-compiled inner classes will not work properly in all cases.
- Inner classes are not properly "imported" in some cases. e.g. Class A
contains one ormore inner classes. Class B extends class A and
attempt to refer to those inner classes by their unqualified names.
- Static blocks in the class body are not yet implemented.
- If you have a source class at the wrong place in the classpath you
may get a non-fatal error when bsh attempts to load the class again.
- JDK 1.5 style static class imports.
You can import the static methods and fields of a java Class into a
BeanShell namespace. e.g.
static import java.lang.Math.*;
sqrt(4.0);
- Instance object imports (mix-ins) with the importObject() command.
You can import the methods and fields of a Java object instance into
a BeanShell namespace. e.g.
Map map = new HashMap();
importObject( map );
put("foo", "bar");
print( get("foo") ); // "bar"
Changes in 1.3 beta 2
- Added reserved Java keywords. Added: "strictfp", "implements",
"extends", "package", and the new Java 1.5 keyword "enum". strictfp
is now accepted as a modifier and ignored. The rest are not supported
(yet).
- Properties style auto-allocation of variables.
// foo is initially undefined
foo.bar.gee = 42;
print( foo.bar.gee ); // 42
print( foo.bar ); 'this' reference (XThis) to Bsh object: auto: bar
print( foo ); 'this' reference (XThis) to Bsh object: auto: foo
When assigning to a compound name, undefined components being resolved
relative to a scripted object (bsh 'this' type reference) will be
auto-allocated as scripted objects.
This means that you can use BeanShell scripts just like properties
files for initialization parameters.
- BeanShell 'this' type references can now implement more than one
interface. Use getInterface( Class[] ).
- BshMethod getReturnType() now returns Void.TYPE for voids, rather than
the (incorrect) wrapper value Primitive.VOID.
Changes in 1.3 beta 1
- Class management Changes - class management is now on a per-Interpreter
(per-global namespace) basis. Multiple interpreter instances do not
any longer by default share classpath or cached class information.
(This will help in application server deployments).
Added the method: Interpreter getClassManager()
%% some minor incompatible changes in NameSpace.java api.
The NameSpace constructor now requires a BshClassManager instance.
- Fixed bug where using reloadClasses() with no arguments (to reload
the entire classpath) did not work properly when user classpath
included components with relative paths.
- Scripts executed with the run() command (as opposed to eval() or
source()) now have their own class manager and therefore classpath.
Any namespace the is disconnected from its parent ( pruned() ) now
creates a new class manager. In the case of the run() command we
inherit the old classpath explicitly. ** this behavior may evolve
to cover all cases of a namespace with no class manager or parent **
** Note: the test suite runs a bit slower now because each test is
run() and starts with a clean class cache. Note that creating
variants on the run() behavior (e.g. the old behavior) is trivial...
See run.bsh.**
- The bsh.system object which is shared across Interpreter instances
has been renamed "bsh.shared". For backwards compatibility it is
still referenced as bsh.system as well.
- Implemented some experimental caching of class method resolution for
performance. When a method (overloaded or single) is resolved
for a set of arguments, that method is cached based on the class,
argument types, and method name. In the test case of a loop simply
calling a Java method on an object or class this results in about a
40% speedup.
- Consolidated bsh scripted method dispatching code into This.java.
- All scripted objects now implement the standard object protocol of
toString(), hashcode() and equals(). Scripted object behavior is now
more consistent.
%% One minor incompatible change: the invoke() meta-method no longer
intercepts toString(), hashcode() or equals(). If you wish to
override these you must script them directly in the object.
- All normal error messages should now include line numbers and invocation
text. Removed the no context constructors for EvalError and
TargetError. Added checked parallel exceptions UtilEvalError and
UtilTargetError for use in other areas. Use UtilEvalError
toEvalError( node, callstack ) to rethrow these as necessary.
- Added script stack trace to error messages (e.g. method a() called
method b(), etc.). EvalError and TargeError now require callstack.
Added getScriptStackTrace() method to EvalError API. toString()
prints the script stack trace by default.
- Errors indicating bad return control or return type from method now
point to the exactly location of the return.
- Removed deprecated get/setVariable methods... optimized for common case
in 1.2 branch so there should be no performance penalty for using
set over setVariable.
- Remote client can now utilize bsh:// URLs to talk to embedded bsh server.
- Added "bsh" target to build.xml for running a script from the current
build. e.g. ant -Dfile=foo.bsh bsh
- Stopped squeltching NoClassDefFound errors... This was the source of
much frustration where BeanShell would indicate that a class was not found
when it was really a dependent class that was missing. We now catch
NoClassDefFoundError and add a message specifying which class was being
loaded when the missing class was encountered.
- Added catch block in bshdoc.bsh script to report the file it was
processing when an error occurs.
- Fixed bug allowing try { } without catch or finally.
- Enhanced bsh shell script launcher to add a default bsh JAR if none
is in the classpath (as determined by a simplistic name test). Works
on cygwin and unix.
- added sourceRelative() command, which sources the file relative to the
caller's directory, not the working directory. (See the bshdoc for
more explanation).
- Upgraded to JavaCC 3.0 available from:
http://www.experimentalstuff.com/Technologies/JavaCC/index.html
(Let's see, how many places have owned JavaCC now? Sun Test, then
meta-mata, then webgain, then Sun "experimental stuff"? Sun has
promised to open source this... I hope they follow through).
Modified build.xml to use the new package name (no package) and to assume
that the jar is in the classpath for now.
- Refactored and removed the ugly "LHS" part of the grammar. This removed
two ASTs which essentially duplicated the "straight" part of the grammar.
There is still a bsh.LHS class (used as a wrapper for the target of an
assignment expression) but it is now created based on context rather than
via a special part of the grammar. Perhaps no one but the author will
care about this but this always bugged me. This also fixes some corner
cases in assignments that were broken (tests fields1.bsh fields2.bsh).
- Switched to BSF 2.3 package name by default. BSF was donated from
IBM to Jakarta and changed its package name with 2.3. If you want to use
an older release just grab the bsh-bsf-1.2x.jar and drop it in your
classpath ahead of your bsh 1.3 dist.
- Enhanced array creation to support undefined array dimensions as in Java.
e.g. int [][][] foo = new int [3][][];
- Enhanced array creation to support auto-casting in array initializers,
e.g. even though a primitive number is an int by default, the following
work (as in Java):
long [] la = { 1, 2 }; float [] fa = { 1, 2 }; byte [] ba = { 1, 2 };
- Fixed import caching bugs.
- Fixed problems with the method resolution caching: added hashcode to
Primitive.java, fixed null argument issue, separated caches into object
and static, removed side-effect from resolveJavaMethod in Reflect. Did
some initial profiling. Overall I'm seeing about a 50% speedup in code
that does primarily method dispatching.
- Improved BshServlet - EvalErrors (and ParseExceptions) no longer spew the
whole script again in the error message. Inline evals now report the
"file" name as an abbreviated portion of the script. Errors in the
servlet results show the offending line highlighted (html) red with four
lines of context and line numbers. Tweaked the template html just a bit.
- Removed deprecated overloaded NameSpace.getImportedClasses()
- Moved Name.ClassIdentifier to bsh.ClassIdentifier top level class.
ClassIdentifier allows scripted methods to accept plain class names as
arguments. e.g. javap( java.lang.String );
- Enhanced error message for undefined argument to method - now prints the
argument text instead of the method parameter name and position.
- Clarified the scoping of closures and blocks and fixed related bugs.
This should now work as expected. A 'this' reference always refers to
the nearest enclosing method or the root scope. This is true even when
the expression returning or assigning the 'this' reference is inside one
or more nested Java blocks. e.g.
myMethod() {
if ( true ) { /* inside block */ } else { /* */ }
{ /* gratuitous block */ }
}
Returning or assigning 'this' from inside myMethod() or inside any of
the blocks all use the same 'this' reference (myMethod). You may
refer to local, typed variables declared in the block this.varname as
expected, but they are not visible outside of the block (as in Java).
See tests/this.bsh.
Methods defined inside a block are now set in the parent block, as
is consistent now with everything else blocks do.
- Enhanced error message on missing or incorrect constructor.
- Switch statements should now work with all types. Previously numeric
switch statements only worked on the type int. Now any numeric type or
wrapper will serve as the switch value and auto-boxing/unboxing occur as
elsewhere in BeanShell. Switch statements also work with object types -
e.g. you can switch on Strings, Dates, or any objects that test properly
with equals(). The only comparsion that is not allowed is primitive to
arbitrary (non wrapper) object type. e.g. compare 2 to a "two". This
will cause a primtiive / object mismatch error.
- Did some minor refactoring and cleanup in Reflect.java.
- Added enhanced for loop as per JDK 1.5. (Belated thanks to Daniel Leuck
for submitting his "each" statement long, long ago.)
So you can now do things like:
List foo = getSomeList();
for ( untypedElement : foo )
print( untypedElement );
for ( Object typedElement: foo )
print( typedElement );
int [] array = new int [] { 1, 2, 3 };
for( i : array )
print(i);
for( char c : "a string" )
print( c );
Supported iterable types include all the obvious things. See the new
BshIterator API addition below for specifics.
- Added BshIterator interface and bsh CollectionManager. BeanShell now
supports a unified API for iteration over composite types including:
JDK 1.1+ (no collections): Enumeration, arrays, Vector, String,
StringBuffer
JDK 1.2+ (w/collections): Collections, Iterator
This was added to support the enhanced for loop above, but can also be
used in BeanShell commands that wish to loop over composite objects to
display them. Usage e.g.:
cm = CollectionManager.getCollectionManager();
if ( cm.isBshIterable( myObject ) )
{
BshIterator iterator = cm.getBshIterator( myObject );
while ( iterator.hasNext() )
i = iterator.next();
}
If this proves to be useful we could add additional syntactic magic
later such as making all composite types implement the magic interface
BshIterable, etc.
- Maps can now be used with the hastable style accessor syntax. e.g.
Map map = new HashMap();
map{"foo"} = "bar":
This is another feature of the new CollectionManager (1.1 compatibility
is preserved).
- Fixed void initializer bug whereby a void value in an initializer was
allowed.
- Refactored variable management a bit - all variables (both typed and
untyped) are now wrapped in the wrapper NameSpace.Variable. We could put
a listener API on there if we wanted to "watch" variables.
- Implemented BeanShell 1.3 scoping changes and preserved backwards
compatibility with the -Doldscoping=true switch.
Variable assignments now search up the parent chain for the variable
definition rather than defaulting to local scope. Variable assignments
will therefore find the nearest enclosing variable definition - as one
would expect in Java. Prior to BeanShell 1.3 this was only true when
"strict Java" mode was enabled.
So, for example, the following idiomatic things now work as expected:
incrementX() { x=x+1; }
x=1;
incrementX();
assert( x == 2 ); // true!
setFlag() { flag=true; }
setFlag();
assert( flag == true ); // true!
A typed variable declaration will define the variable as usual in Java.
An untyped variable assignment - for which there is no definition in any
enclosing scope - will end up as a global variable assignment. e.g.
int x = 1; // parent scope
foo()
{
// foo scope
x = 2; // x assigned in parent scope
int y = 1;
y = 2; // y assigned in foo scope
z = 99; // z has no enclosing scope, assigned global
}
An untyped variable can be made local in scope by assigning it with
the qualifier 'this' (as usual) or by using the new keyword 'var' to
explicitly declare it as an untyped variable.
foo()
{
// secret1 and secret2 are local to foo()
this.secret1 = 42;
var secret2 = 42;
}
The qualifier 'super' may be used, as in Java, to begin the search for a
variable definition in the parent scope.
int y = 1; // parent scope
foo()
{
int y = 2;
super.y = 3; // y assigned in parent scope
}
Object closures (scripted objects created by returning 'this' from a
method) may be assigned variables as usual:
foo() { return this; }
foo=foo();
foo.a=1;
print( foo.a ); // 1
print( a ); // ERROR 'a' undefined here
The old scoping rules have been preserved and can be switched on by
setting the system property "oldscoping" to "true".
See newscoping.bsh in the test suite for more examples.
Note: There is exactly *one* line of code in NameSpace.java that looks
at the Interpreter.LOCALSCOPING flag to determine what to do.
Everything else happens naturally. So there should be no fear that we
are accumulating legacy baggage. On the contrary, BeanShell refactoring
is making it more powerful and simpler.
- Fixed bug in evaluation of operator-assignments with postfix in the
RHS. (Java behavior seems strange to me.) e.g.
i=1; i+=i++; // should be 2 apparently
i=1; i+=i++ + i++; // should be 4 apparently
- BeanShell now supports modifiers (e.g. public, private, static, etc.)
on methods and typed variable declarations. Basic sanity checks are
done at parse time (e.g. you can't apply volatile to a method, mix
public/private, etc.) "final" is implemented for typed variables but
other modifiers are currently ignored.
- The "synchronized" modifier is now implemented for methods and
synchronized blocks. Synchronized blocks work as in Java.
Synchronized methods lock their parent namespace's This reference, so
for purposes of serialization (executing exclusively one one at a time)
they work as in Java.
// The following synchronize on the same lock
synchronized ( this ) { } // block
synchronized int foo () { } // method foo
synchronized int bar () { } // method bar
int gee() {
synchronized( super ) { } // inside gee()
}
- The "throws" clause on methods is now supported. Throws clause names
are validated to insure that they are known class types, however they
are not enforced beyond that.
- Field style object property access now supports isFoo() style getters
for boolean properties. e.g.
Float flot = new Float(0f);
print( flot.infinite ); // float.isInifinite()
- Improved error message on attempting to access private/protected
members without setAccessibility(true). Error now shows non-public
method and suggests turning on accessibility.
- User defined command prompt in interactive mode. The user may set the
value for the prompt string using the variable bsh.prompt or by
defining the scripted method (or command) getBshPrompt().
If the command or method getBshPrompt() is defined it will be called to
get a string to display as the user prompt. For example, one could
define the following method to place the current working directory into
their command prompt:
getBshPrompt() { return bsh.cwd + " % "; }
The default getBshPrompt() command returns the value of the variable
bsh.prompt if it is defined or the string "bsh % " if not. If the
getBshPrompt() method does not exist, throws an exception, or does not
return a String, a default prompt of "bsh % " will be used.
- Fixed JConsole backspace through prompt bug.
- User Command Path - You may now import BeanShell scripted or compiled
commands from any package path using the importCommands() method.
You may use either "/" path or "." package notation. e.g.
// equivalent
importCommands("/bsh/commands")
importCommands("bsh.commands")
importCommands("/mypackage/commands")
When searching for a command each path will be checked for first, a file
named 'command'.bsh and second a class file named 'command'.class.
You may add to the BeanShell classpath using the addClassPath() or
setClassPath() commands and then import them as usual.
addClassPath("mycommands.jar");
importCommands("/mypackage/commands");
If a relative path style specifier is used then it is made into an
absolute path by prepending "/". Later imports take precedence over
earlier ones.
Imported commands are scoped just like imported clases.
See news about the invoke() meta-method for information about how you
could implement your own, arbitrary command loading.
- The getResource() command now utilizes the BeanShell classpath. i.e.
you can get resources from paths added with addClassPath(), etc.
- The BshClassManager now contains the methods getResource() and
getResourceAsStream().
- The invoke() meta method is now allowed directly in scope, e.g.
invoke( String methodName, Object [] arguments ) { ... }
// invoke() will be called to handle noSuchMethod()
noSuchMethod("foo");
Previously invoke() was only allowed when called through a 'this' type
reference, e.g. a scripted interface.
One could use this capability to implement arbitrary command loading
capabilites. You would simply have to handle wrapping/ unwrapping
of primitive value argument and return types with bsh.Primitive.
- Fixed bug in bsh.util.Httpd causing class format errors
- Changed BshServlet redirect to a relative path rather than absolute URL
Changes in 1.2b8
- Some Strict Java mode bugs fixed. %(Incompatible API Change)
Strict Java mode now functions on a per-interpreter basis. The
static strictJava variable is gone (it was always labeled as
experimental) and is replaced by the instance method Interpreter
setStrictJava(). Interpreter set() methods now work (untyped
assignments using Interpreter set() is allowed and existing untyped
variables can be assigned in scripts in either mode).
+Added a test case.
Most bsh commands still fail in strict Java mode for simple reasons.
These will be cleaned up in a future release.
- Added check for return of value from void method. Fixed a couple of
places where we made this mistake.
- Added dirname() command
- Added "test" target to build file (runs tests/RunAllTests.bsh). Modified
RunAllTests.bsh to cd() to the directory containing the RunAllTests.bsh
script before executing the tests.
- Certain return type errors now point to specific locations, rather
than the method declaration.
- Fixed Interpreter setErr() bug
- Exceptions thrown from proxy interfaces are now unwrapped.
If your script throws an exception (or causes one to be thrown) that is
declared in the throws clause of your interface method then it will
be thrown from the method as one would expect.
Changes in 1.2b7
- Fixed bug where Interepreter was not closing Reader streams after
sourcing files.
- Updated JConsole to work with Java 1.4.1 (changed getText.length() calls
to getDocument().getLength()).
Changes in 1.2b6
- Made bsh.Parser public and added a main() method allowing users to
call the parser on files for simple validity checking.
- Made a small addition to grammar to provide an option to retain
formal (javadoc style) comments in the parse tree.
- Fixed accessibility bug in finding fields
- Fixed minor bugs in bsh.servlet.BshServlet
- Fixed scoping on catch blocks such that untyped variables
in the catch parameter do not leak out of the block. They now act as they
would with a declared type in Java (local).
- Fixed some thread safety bugs with try/catch blocks.
- Fixed Interpreter serialization issue - reset streams
- Fixed bug in accessibility affecting access to package
hidden superclasses
- Added bshdoc.bsh script and bshdoc.xsl stylesheet to the scripts dir.
These can be used to support javadoc style documentation of bsh files
and commands.
- Exposed bsh.BshMethod and added a public invoke() method.
- Added getMethod() method to namespace to enumerate methods
Changes in 1.2b5
- Fixed bug in using loosely typed vars as indexes to arrays.
Changes in 1.2b43
- Modified Interpreter so that it is serializable. It is now possible
to serialize the entire bsh instance. The load() and save() commands
however recognize when you are trying to save a bsh object (bsh.This)
type and detach it from the parent namespace, so that saving a bsh object
in this way doesn't drag along the whole interpreter See This.bind()
and This.unbind().
Changes in 1.2b42
- Fixed bug affecting scoping in blocks. This was breaking the test
harness in a way that caused it not to be caught before!
- Added a special test for the test harness to catch this sort of thing
in the future.
- Fixed problem with classloading causing bsh extensions not to be found
in some environments.
Changes in 1.2b3 and 1.2b4
- The class manager now uses the Thread "context classloader" to factory
classes when available. This allows bsh to be instaled in the
jre/lib/ext directory (previously user classes wouldn't be seen). Yeah!
This may also allow bsh to see user classpath in application servers
like Weblogic, which include it in the core classpath.
- added the clear() command to clear variables and methods from a namespace.
- added setOut() and setErr() methods to Interpreter.
- added new bsh.servlet package to the main distribution
- added bsh.Remote launcher
Changes in 1.2b2 and 1.2b3
- Fixed nasty bug in grammar, added test cases to expression.bsh
Note: this bug was preventing the latest version of bsh from being used
with Weblogic 6.1 (which uses bsh internally and was failing).
- Handled the "single line comment with no terminating linefeed" issue.
- Added getInterface() method to Interpreter.
- Fixed issue with identity semantics for this in cast case, e.g.:
((ActionListener)this == (ActionListener)this) is now true.
- cat() command will now print files, urls, streams, and readers
- Exceptions in native (compiled) user classes will now print with
both the BeanShell script trace and native Java stack trace.
Exceptions generated directly from scripts continue to show only the
script trace (as that is all that is meaningful).
- ParseException is now a subtype of EvalError... previously these were
caught and the contents were moved to an EvalError. This means that
now the ParseEx implements the contract as far as getErrorLineNumber(),
getErrorText(), and getErrorSourceFile()
- In interactive mode, the last uncaught exception is now available in
the variable $_e (like the $_ last result).
- Improved the way that imported class names are cached in namespaces,
reducing the number of caches created and improving performance.
- Improved javap() command slightly
- Modified This references so that you can invoke the methods: getClass()
and invokeMethod() on them.
Changes in 1.2b1
- Added support for accessibility - Using a dynamically loaded extension
(ReflectManager) bsh will use the accessibility API to allow access to
private/protected fields, methods and non-public classes.
- added the setAccessible() command to turn this behavior on/off
(it is *off* by default in this release... we may change that later.)
- added the bsh.reflect package and ReflectManager to support the above.
- Added support for IBM's Bean Scripting Framework (BSF).
- Added a new jar with the BSF adapter class bsh.util.BeanShellBSFEngine.
This is also included in the full dist (it's small).
Changes in 1.1alpha19
- Added a somewhat experimental "strict java" mode. When activated with
setStrictJava(true) BeanShell will:
1) require typed variable declarations, method arguments and
return types
2) modify the scoping of variables to look for the variable declaration
first in the parent namespace, as in a java method inside a java
class. e.g. if you can write a method called incrementFoo() that
will do the correct thing without referring to super.foo.
-Added another form of the run() command that accepts an argument to be
passed to the child context.
Changes in 1.1alpha18
- Fixed a bug affecting static field access.
- Added a static setClassLoader() method to Interpreter and BshClassManager.
This allows an external classloader to be specified.
Changes in 1.1alpha17
- Added simple mechanism to allow scripted commands/methods to accept
class identifiers as arguments. e.g.:
foo( java.lang.String );
see Name.identifierToClass();
- added which() command - shows class source
- modified class browser to show class source when class is selected
- Changed the type of the (very recently added) this.callstack reference
from an array to the actual internal CallStack type. You can use
toArray() to get it as an array or get(depth).
- added unset() method to interpreter. Fixed bug in set(name, null);
Changes in 1.1alpha16
Improved error reporting on errors during method invocation.
Changes in 1.1alpha15
- Fixed server mode and re-added demo applets
Changes in 1.1alpha14
- Important addition - nodes now remember the source file from which
they were parsed... This means that debug messages are much more useful
now, especially with errors thrown from multiple files.
- Re-worked classpath mapping user feedback mechanism (will support a GUI
now).
- Fixed general block scoping bugs. Things should act like Java now.
typed vars declared within blocks does not leak out. e.g. { int i=2; }
- Added getSourceFileInfo() command to return the name of the file or
source from which the current interpreter is reading.
(uses this.interpreter.getSourceFileInfo())
- Added a parent ref to the interpreter for future use by the child
eval interpreters.
- Added new form of source() command that takes a URL.
- Tenatively added new feature to get the text of a method invocation
from within the method body... may be useful along with this.caller.
See namespace.getInovcationText() namespace.getInvocationLine().
Changes in 1.1alpha13
- Removed core bsh package dependencies on 1.2. bsh will now compile and
run under 1.2 with some features limited.
- Moved some console related functionality out of the core and into
bsh.util. There is a new bsh.util.GUIConsoleInterface. NameCompletion
is now in bsh.util and has been replaced with a more primitive
supporting interface bsh.NameSource in the core.
Changes in 1.1alpha13
- Moved all Name object construction into NameSpace. This will support
Name resolvers caching information in the future. e.g. bsh could
know that a variable is final and optimize the resolution. Or a long
chain of names (java.lang.Integer.MAX_INTEGER) could be optimized.
- .class on array types now working. e.g. "int [][].class"
- Fixed bugs in grammar which dissallowed certain simple expressions
unless they were parenthesized.
- Fixed related case relating to expressions with assignment to a
throw-away method invocation.
Changes in 1.1alpha12
- Added "trace" option.
java -Dtrace=true bsh.Interpreter or trace() command.
turns on printing of each line before it is executed. Note that
this currently prints only top level lines as they are parsed and
executed by the interpreter. trace skips over method exectutions
(including bsh commands) etc. We may enhance this in the future.
- Fixed bugs in primitive handling:
primitives smaller than int are now properly assigned in declarations:
e.g.: byte b = 5;
primitive are internally promoted to the correct type on assignment
to a typed variable,
e.g.: long l = (byte)5; // l is a long value internally
- Cleanup in cast operation.. re-factored the cast code into two types
of cast operations.
- Tightened up re-declaration of typed variables... You can only redeclare
a typed var with the same type, else you can an error. Previously
there was odd behavior where you could redeclare with an assignable
type.
- Fixed bugs in casting
- Fixed the remainng known bugs in array initializers. They should now
work precisely as they do in Java. See the test case for examples.
Changes in 1.1alpha11
- Improved error reporting. There should no longer be any errors that
report "unknown location". At minimum, any error should report the
source file and line number. Normally errors should report the text
of the offending construct.
- Cleanup in array handling
- fixed bugs with null initializers, loose types, promotion
Changes in 1.1alpha10
- Major internal change: added callstack.
two new magic references:
bsh.This this.caller
NameSpace [] this.callstack
- Some commands now behave correctly, some slight changes- see notes.
- made change to insure that pathToFile always returns canonical path
this fixed bug in addClassPath
Changes in 1.1alpha9
- Added a very nice ANT build file, yeah!
- added a bit more synchronization in console
- removed unnecessary file/cwd related code
- removed unnecessary code and lib file supporting old java packages
optimization
- Redefined scoping behavior in for-init to act like java. See notes
elsewhere.
- Added switch statement.
- made sure Ctrl-d in interctive mode shuts down interpreter.
- Fixed cleanup of weak references in bsh class manager
Changes in 1.1alpha8
- import changes -
- Defined order of precedence - in bsh ambiguous imports are compile
time error. In bsh (where imports are allowed at any time) the later
import takes precedence.
- fixed bug causing stack overflow
- Internal restructuring - made Interpreter its own class rather than being
auto-generated by javacc. This allows us to generate decent javadoc
without hundreds of parser methods. Improved javadoc.
- Fixed the precedence of name resolution to be consisten with Java.
variables now always trump class name resolution. This simplifies
name resolution a bit and also makes things a bit faster. However
sometimes you have to use import now to get around a variable name
hiding a qualified class name.
- Performance improvement: empty loop test case four times faster
overall appears as much as 30% faster
- Fixed bugs in certain runtime exception handling
- Fixed bug in static vs. dynamic overloaded method resolution
Changes in 1.1alpha6
- Fixed grammar relating to Expression/BlockStatement parsing... This was
causing the bug with declaring a typed var with a fully qualified class
name. e.g. java.sql.Date date;
- Fixed 'return' control problem... return now works properly for
interactive and non-interactive scripts.
F- ixed CTRL-D bug - now exits properly in interactive mode.
Changes in 1.1alpha51
- Fixed bug causing debug code to call toString() for every method call on
an object, whether debug was on or not.
Changes in 1.1alpha5
- License change only... Added SPL Sun Public License agreement with
dual licensing under the LGPL
Changes in bsh1.1alpha3
- Added .field style properties access in addition to the {} style
- Added anonymous inner class style allocation for interfaces
- Class browser now reflects changes in class path
Changes in bsh1.1alpha2
- Added namespace based name completion for console and classbrowser
completes variables, classes, (coming... bsh methods)
- Added import * capability to read full classpath.
- Added overloading of bsh scripted methods
- Added classpath extension and class reloading
reloadClasses() reloadClasses(item)
- improved the class browser - now has tree for packages and reflects
the extended classpath , other improvements
- Loose arguments in catch clauses now work.
- deprecated getVariable/setVariable api. Moving to get()/set()/eval();
- internally removed BshVariable - it wasn't doing what it was historically
intended for. Replaced with a new bsh.system object which is shared
across all interpreter instances
- Added bsh.system.shutdownOnExit variable... set to false to prevent
exit() command and desktop close from exiting the VM.
- Loosened grammar to allow import statments in any block scope.
e.g. you can now import within a bsh method and the import will be local
to that method
-Added javap() command... accepts object, class, or string name and
prints public fields, methods
-Added -Doutfile=foo capture output to file (for Windows nightmare)
- Cleaned up a lot of internal structure in NameSpace, etc.
-Interpreter now implements ConsoleInterface
Changes in bsh-1.01
- Modified parse exceptions to print more tersely...
Removed the "Was expecting one of..." text *except* when debug is on.
- Fixed the for statement forinit scope bug.
- Got rid of remaining deprecation (using Readers vs. InputStreams)
- Fixed top level expression bug #111342
- Fixed problems with compiling bsh under limited JDK verions (1.1, 1.2).
We now have factories insulating the core from all extended version
capabilities (e.g. proxy mechanism stuff, accessibility TBD)
- Modified browseClass() command to accept string class name,
class instance, or arbitrary object
- Modified search for constructors to accommodate package and private
access.
- Added close button to class browser.
Changes in bsh-1.0
Not necessarily in order of importance