You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@groovy.apache.org by Thierry Hanser <Th...@lhasalimited.org> on 2016/04/05 18:23:30 UTC
RE: Failing to change the value of a Java variable from wihtin a
Groovy script (potential issue?)
Thank you for your explanations Jochen.
Your assumption is correct except that I think you meant to say that *The subsequent "println '2 = ' + x" will then get the value property instead of asking for the binding.* since the value ('one') is the one from the unchanged variable/property 'x'.
I am confused by the lack of symmetry in the handling of the binding/properties value.
"print x" in a script will call getX() (if available) whereas "x = 'two'" will call setProperty() which will add 'x' (if not present) to the binding but will not subsequently call setX(). As a result when you use the value of a property 'x' in a script you are guaranteed to call getX() but if you assign a value to 'x' you are not guaranteed that setX() is called !
This dissymmetry seems odd to me (and obviously leads to unexpected behaviour).
Note that without static compilation the same dissymmetry is observed, however for some reason the property 'x' is then properly updated but still without calling setX()
Really confusing to me !!!
I understand there are issues with the CompileStatic option. Do we know what level of prioritization are these issues in the bigger Groovy roadmap?
Thanks a lot,
Thierry
--
Groovy script:
"println '1 = ' + x; x = 'two'; println binding.variables; println '2 = ' + x\n";
With static compilation:
init x: one
Getting x : one
1 = one
Setting property : x = two
[x:two]
Getting x : one
2 = one
Without static compilation:
init x: one
Getting x : one
1 = one
Setting property : x = two
[x:two]
2 = two
Java:
public abstract class JavaScriptDemo extends Script
{
public String x;
public JavaScriptDemo()
{
x="one";
System.out.println("init x: " + x);
}
public Object process(Closure code)
{
System.out.println("Processing with x : " + x);
return code.call();
}
public void setX(String text)
{
System.out.println("Setting x : " + x);
x = text;
}
public String getX()
{
System.out.println("Getting x : " + x);
return x;
}
@Override
public void setProperty(String property, Object value)
{
System.out.println("Setting property : " + property + " = " + value);
super.setProperty(property, value);
}
/**
* Demo main
* @param args
* @throws Throwable
*/
public static void main(String...args) throws Throwable
{
////
// Compilation configuration
CompilerConfiguration configuration = new CompilerConfiguration();
//configuration.addCompilationCustomizers(new ASTTransformationCustomizer(CompileStatic.class));
configuration.setScriptBaseClass(JavaScriptDemo.class.getName());
GroovyShell shell = new GroovyShell(JavaScriptDemo.class.getClassLoader(), configuration);
// source code
String scriptSource= "println '1 = ' + x; x = 'two'; println binding.variables; println '2 = ' + x\n";
// compile the source code and run the compiled script
JavaScriptDemo compiledScript = (JavaScriptDemo)shell.parse(scriptSource);
compiledScript.run();
}
}
-----Original Message-----
From: Jochen Theodorou [mailto:blackdrag@gmx.org]
Sent: 09 March 2016 13:37
To: users@groovy.apache.org
Subject: Re: Failing to change the value of a Java variable from wihtin a Groovy script (potential issue?)
On 08.03.2016 17:31, Thierry Hanser wrote:
[...]
> *In Groovy*
>
> println '1 = ' + x
> x = 'two'
> println '2 = ' + x
>
> *Output*:
>
> init x: one <- initial value assignement OK
>
> 1 = one <- successfully accessing 'x' from within the compiled script
> OK
>
> the Groovy script has picked up the value of the Java
> variable;
>
> the implicit getX() has been called
>
> 2 = one<- should be 'two' as per Groovy code (second line)
>
> but is unchanged ???
I think I know the problem...
When using a normal script your code above would have failed, because the binding does not contain x. But what exactly happens if the binding does not contain x? Groovy will ask the script class for a property of that name. Just to point it out very much: The property lookup is done only after binding is asked!
What happens in the set case then? The set case will always set a variable in the binding!
This explaines why for you "println '1 = ' + x " gives "one", but the important part is that then "x = 'two'" will leave that property untouched and goes directly to the binding. The subsequent "println '2 = ' + x" will then get the value from the binding instead of asking for the property.
The solution would be to overwrite setProperty to first try setting the property on the script class, basically: a try with this.getMetaClass().setProperty(this, property, newValue); and a super.setProperty in the exception case.
And it seems the static compiler has big issues with handling a script base class
bye Jochen
________________________________
Switchboard: +44 (0)113 394 6020
Technical Support: +44 (0)113 394 6030
________________________________
Lhasa Limited, a not-for-profit organisation, promotes scientific knowledge & understanding through the development of computer-aided reasoning & information systems in chemistry & the life sciences. Registered Charity Number 290866. Registered Office: Granary Wharf House, 2 Canal Wharf, Leeds, LS11 5PS. Company Registration Number 01765239. Registered in England and Wales.
This communication, including any associated attachments, is intended for the use of the addressee only and may contain confidential, privileged or copyright material. If you are not the intended recipient, you must not copy this message or attachment or disclose the contents to any other person. If you have received this transmission in error, please notify the sender immediately and delete the message and any attachment from your system. Except where specifically stated, any views expressed in this message are those of the individual sender and do not represent the views of Lhasa Limited. Lhasa Limited cannot accept liability for any statements made which are the sender's own. Lhasa Limited does not guarantee that electronic communications, including any attachments, are free of viruses. Virus scanning is recommended and is the responsibility of the recipient.
Re: Failing to change the value of a Java variable from wihtin a
Groovy script (potential issue?) [POTENTIAL FIX]
Posted by Jochen Theodorou <bl...@gmx.org>.
On 05.04.2016 20:00, Thierry Hanser wrote:
[...]
> The solution is indeed close to what Jochen suggested. I just think that for the sake of symmetry all we need to do is to check whether the variable exists in the binding and behave accordingly. If it exists change its value in the binding, if it does not exist delegate to the meta class (or call super.super). This would ensure a consistent (symmetrical) behaviour between getProperty() and setProperty().
>
> @Override
> public void setProperty(String property, Object newValue)
> {
> if ("binding".equals(property)) setBinding((Binding) newValue);
> else if("metaClass".equals(property)) setMetaClass((MetaClass)newValue);
> else if(getBinding().hasVariable(property)) getBinding().setVariable(property, newValue); // <== if variable exist => set binding value
> else getMetaClass().setProperty(this, property, newValue); // <== variable does not exist => delegate to meta class
> }
>
> Note that the origin of the dissymmetry is actually a consequence of the dissymmetry between the getVariable() and the setVariable() methods in the Binding class. getVariable() throws an exception when the variable does not exist where as setVariable() does not (it adds the variable to the linked map). The former exception is used by getProperty() in Script to trigger the delegation to the meta class but this can be used by setProperty() since no exception is thrown by setVariable() in Binding.
well.. this means a script "x=1", without any further doing will fail.
At the beginning the binding is empty, so
getBinding().hasVariable(property) will return false, and a setX() does
not exist.
what I suggested was actually to first ask the meta class and then the
binding.
[...]
> @Override
> public void setProperty(String property, Object newValue)
> {
> if ("binding".equals(property)) setBinding((Binding) newValue);
> else if("metaClass".equals(property)) setMetaClass((MetaClass)newValue);
> else
> {
> getBinding().setVariable(property, newValue); // <== update binding
> getMetaClass().setProperty(this, property, newValue); // <== notify meta class
> }
> }
hmm... what about exceptions when setting the property? The obvious one
to handle is of course if the property does not exist. But there is also
the type conversion problem. For example if your property is an int and
your value is a String. Will this cause an exception being thrown to the
user level code? If yes, you will not have the old behaviour. Will the
binding still have the value? And besides actually being able the call
the setter, let's not forget the method code itself can still throw
exception... for example NullPointerException. Different level, but same
question as before.
bye Jochen
RE: Failing to change the value of a Java variable from wihtin a
Groovy script (potential issue?) [POTENTIAL FIX]
Posted by Thierry Hanser <Th...@lhasalimited.org>.
Hi everyone,
After further investigating the unexpected behaviour due to the getProperty()/setProperty() behaviour dissymmetry, I understood the underlying ground which is indeed compatible with what Jochen described.
In the Script class, getProperty() implicitly checks if the variable exists in the binding using the exception try/catch. If it exists it gets the value from the binding and if it does not exists if delegates to the super class (GroovyObjectSupport) which in turn delegates to the meta class which eventually calls getX().
On the other hand setProperty() simply sets the variable in the binding regardless of its existence prior to the assignment. At this stage the meta class gets no chance to act when the variable does not exists which introduces a non symmetrical behaviour between getProperty() and setProperty().
In the Script class:
public Object getProperty(String property) {
try {
return binding.getVariable(property); // <== property exists -> value from binding
} catch (MissingPropertyException e) {
return super.getProperty(property); // <== property does not exist -> value from the super class which delegates to the meta class
}
}
public void setProperty(String property, Object newValue) {
if ("binding".equals(property))
setBinding((Binding) newValue);
else if("metaClass".equals(property))
setMetaClass((MetaClass)newValue);
else
binding.setVariable(property, newValue); // <== change/set value in binding regardless of the existence of the variable in the binding or not
}
The solution is indeed close to what Jochen suggested. I just think that for the sake of symmetry all we need to do is to check whether the variable exists in the binding and behave accordingly. If it exists change its value in the binding, if it does not exist delegate to the meta class (or call super.super). This would ensure a consistent (symmetrical) behaviour between getProperty() and setProperty().
@Override
public void setProperty(String property, Object newValue)
{
if ("binding".equals(property)) setBinding((Binding) newValue);
else if("metaClass".equals(property)) setMetaClass((MetaClass)newValue);
else if(getBinding().hasVariable(property)) getBinding().setVariable(property, newValue); // <== if variable exist => set binding value
else getMetaClass().setProperty(this, property, newValue); // <== variable does not exist => delegate to meta class
}
Note that the origin of the dissymmetry is actually a consequence of the dissymmetry between the getVariable() and the setVariable() methods in the Binding class. getVariable() throws an exception when the variable does not exist where as setVariable() does not (it adds the variable to the linked map). The former exception is used by getProperty() in Script to trigger the delegation to the meta class but this can be used by setProperty() since no exception is thrown by setVariable() in Binding.
Finally to be I would suggest to not use the exception of the getVariable() method (in Binding) to decide articulate the behaviour of the getProperty() method (in Script). To be fully symmetrical and consistent I would suggest to also fix the getProperty() method in the following manner:
@Override
public Object getProperty(String property)
{
if(getBinding().hasVariable(property)) return getBinding().getVariable(property);
else return getMetaClass().getProperty(this, property);
}
Another positive consequence is that the binding variables become consistent since script level remain separate from the dynamically introduced binding.
If we wanted both variable/property spaces to be automatically merged and in sync at assignment time then we would have to remove the existence test in the setProperty() code (but I am not sure this is what we want; it is a higher level design decision)
@Override
public void setProperty(String property, Object newValue)
{
if ("binding".equals(property)) setBinding((Binding) newValue);
else if("metaClass".equals(property)) setMetaClass((MetaClass)newValue);
else
{
getBinding().setVariable(property, newValue); // <== update binding
getMetaClass().setProperty(this, property, newValue); // <== notify meta class
}
}
Unless I missed out some other subtle aspects, I think this should have been the original implementation of the Script setProperty().
Could this be a candidate for a fix? (BTW, this solution works and provides the same behaviour with and without the CompileStatic option activated)
Thank you all for your help.
Thierry
-----Original Message-----
From: Thierry Hanser [mailto:Thierry.Hanser@lhasalimited.org]
Sent: 05 April 2016 17:24
To: users@groovy.apache.org
Subject: RE: Failing to change the value of a Java variable from wihtin a Groovy script (potential issue?)
Thank you for your explanations Jochen.
Your assumption is correct except that I think you meant to say that *The subsequent "println '2 = ' + x" will then get the value property instead of asking for the binding.* since the value ('one') is the one from the unchanged variable/property 'x'.
I am confused by the lack of symmetry in the handling of the binding/properties value.
"print x" in a script will call getX() (if available) whereas "x = 'two'" will call setProperty() which will add 'x' (if not present) to the binding but will not subsequently call setX(). As a result when you use the value of a property 'x' in a script you are guaranteed to call getX() but if you assign a value to 'x' you are not guaranteed that setX() is called !
This dissymmetry seems odd to me (and obviously leads to unexpected behaviour).
Note that without static compilation the same dissymmetry is observed, however for some reason the property 'x' is then properly updated but still without calling setX()
Really confusing to me !!!
I understand there are issues with the CompileStatic option. Do we know what level of prioritization are these issues in the bigger Groovy roadmap?
Thanks a lot,
Thierry
--
Groovy script:
"println '1 = ' + x; x = 'two'; println binding.variables; println '2 = ' + x\n";
With static compilation:
init x: one
Getting x : one
1 = one
Setting property : x = two
[x:two]
Getting x : one
2 = one
Without static compilation:
init x: one
Getting x : one
1 = one
Setting property : x = two
[x:two]
2 = two
Java:
public abstract class JavaScriptDemo extends Script {
public String x;
public JavaScriptDemo()
{
x="one";
System.out.println("init x: " + x);
}
public Object process(Closure code)
{
System.out.println("Processing with x : " + x);
return code.call();
}
public void setX(String text)
{
System.out.println("Setting x : " + x);
x = text;
}
public String getX()
{
System.out.println("Getting x : " + x);
return x;
}
@Override
public void setProperty(String property, Object value)
{
System.out.println("Setting property : " + property + " = " + value);
super.setProperty(property, value);
}
/**
* Demo main
* @param args
* @throws Throwable
*/
public static void main(String...args) throws Throwable
{
////
// Compilation configuration
CompilerConfiguration configuration = new CompilerConfiguration();
//configuration.addCompilationCustomizers(new ASTTransformationCustomizer(CompileStatic.class));
configuration.setScriptBaseClass(JavaScriptDemo.class.getName());
GroovyShell shell = new GroovyShell(JavaScriptDemo.class.getClassLoader(), configuration);
// source code
String scriptSource= "println '1 = ' + x; x = 'two'; println binding.variables; println '2 = ' + x\n";
// compile the source code and run the compiled script
JavaScriptDemo compiledScript = (JavaScriptDemo)shell.parse(scriptSource);
compiledScript.run();
}
}
-----Original Message-----
From: Jochen Theodorou [mailto:blackdrag@gmx.org]
Sent: 09 March 2016 13:37
To: users@groovy.apache.org
Subject: Re: Failing to change the value of a Java variable from wihtin a Groovy script (potential issue?)
On 08.03.2016 17:31, Thierry Hanser wrote:
[...]
> *In Groovy*
>
> println '1 = ' + x
> x = 'two'
> println '2 = ' + x
>
> *Output*:
>
> init x: one <- initial value assignement OK
>
> 1 = one <- successfully accessing 'x' from within the compiled script
> OK
>
> the Groovy script has picked up the value of the Java
> variable;
>
> the implicit getX() has been called
>
> 2 = one<- should be 'two' as per Groovy code (second line)
>
> but is unchanged ???
I think I know the problem...
When using a normal script your code above would have failed, because the binding does not contain x. But what exactly happens if the binding does not contain x? Groovy will ask the script class for a property of that name. Just to point it out very much: The property lookup is done only after binding is asked!
What happens in the set case then? The set case will always set a variable in the binding!
This explaines why for you "println '1 = ' + x " gives "one", but the important part is that then "x = 'two'" will leave that property untouched and goes directly to the binding. The subsequent "println '2 = ' + x" will then get the value from the binding instead of asking for the property.
The solution would be to overwrite setProperty to first try setting the property on the script class, basically: a try with this.getMetaClass().setProperty(this, property, newValue); and a super.setProperty in the exception case.
And it seems the static compiler has big issues with handling a script base class
bye Jochen
________________________________
Switchboard: +44 (0)113 394 6020
Technical Support: +44 (0)113 394 6030
________________________________
Lhasa Limited, a not-for-profit organisation, promotes scientific knowledge & understanding through the development of computer-aided reasoning & information systems in chemistry & the life sciences. Registered Charity Number 290866. Registered Office: Granary Wharf House, 2 Canal Wharf, Leeds, LS11 5PS. Company Registration Number 01765239. Registered in England and Wales.
This communication, including any associated attachments, is intended for the use of the addressee only and may contain confidential, privileged or copyright material. If you are not the intended recipient, you must not copy this message or attachment or disclose the contents to any other person. If you have received this transmission in error, please notify the sender immediately and delete the message and any attachment from your system. Except where specifically stated, any views expressed in this message are those of the individual sender and do not represent the views of Lhasa Limited. Lhasa Limited cannot accept liability for any statements made which are the sender's own. Lhasa Limited does not guarantee that electronic communications, including any attachments, are free of viruses. Virus scanning is recommended and is the responsibility of the recipient.
Re: Failing to change the value of a Java variable from wihtin a
Groovy script (potential issue?)
Posted by Jochen Theodorou <bl...@gmx.org>.
On 05.04.2016 18:23, Thierry Hanser wrote:
> Thank you for your explanations Jochen.
>
> Your assumption is correct except that I think you meant to say that *The subsequent "println '2 = ' + x" will then get the value property instead of asking for the binding.* since the value ('one') is the one from the unchanged variable/property 'x'.
>
> I am confused by the lack of symmetry in the handling of the binding/properties value.
That's because a script is not intended to be a class. In a script the
binding is the important instance anything else is just a fallback. A
simple example would be "this.class". If there is no "class" in the
binding you get the result of getClass(), which is normally what you
intend to get. On the other hand, if you did "foo=1", why would you
expect setFoo() being called instead? You just set something in the
binding, so why not get it from the binding?
> "print x" in a script will call getX() (if available) whereas "x = 'two'" will call setProperty() which will add 'x' (if not present) to the binding but will not subsequently call setX().
That is write, but don't forget that "print x" will go through
getProperty, which is written to ask the binding. So there is a symmetry
with getProperty and setProperty, it just does not extend to getX and
setX, because the implementation of the setter breaks that.
So the question is really if we should not first look for a setter and
only use the binding for setting, if the setter is not available.
Well... traditionally scripts don't really have a lot of setters, that's
why it normally it does not matter. And if you need that kind of logic,
you can still use a class in the script... normally... of course base
script changed that story a bit, but then maybe it should redefine the
binding logic in this case as well. That's fully up to the custom base
class actually.
> As a result when you use the value of a property 'x' in a script you are guaranteed to call getX() but if you assign a value to 'x' you are not guaranteed that setX() is called !
wrong "x=1; print x" will get you the value from the binding. so getX()
is not guaranteed to be called. and setX() is actually never called
through a "x=1".. unless you did mean with static compilation
> This dissymmetry seems odd to me (and obviously leads to unexpected behaviour).
>
> Note that without static compilation the same dissymmetry is observed, however for some reason the property 'x' is then properly updated but still without calling setX()
I would call that a bug in the static compilation. But I wonder how the
static compiler is actually supposed to behave. It cannot know the
values of the binding, thus it would need to compile against the setters
and getters. It would then be wrong to ask the binding for anything,
unless you do it directly. Which means all scripts using the binding
normally won't compile statically. But hey, I think that is ok.
[...]
> I understand there are issues with the CompileStatic option. Do we know what level of prioritization are these issues in the bigger Groovy roadmap?
we normally use only blocker, major and minor. A blocker is a bug you
cannot work around easily. Since in your case you can call the setter
instead, it would be no blocker. minor are bugs, that show more little
inconsistencies. And major priority is the vast majority of the bugs.
Sometimes we also use blocker to ensure it is in the next release... but
we have gotten away from this.
bye Jochen