AW: AW: [Jython-dev] work in progress

[Schmidmeier, A. <Arn...@si...>]

> -----Ursprüngliche Nachricht-----
> Von:	Samuele Pedroni [SMTP:pe...@in...]
> Gesendet am:	Donnerstag, 9. November 2000 13:55
> An:	jyt...@li...
> Betreff:	Re: AW: [Jython-dev] work in progress
> 
> Hi.
> 
> [Schmidmeier, Arno]
> >  Yes they can be different. e.g. py passing
> > -Duser.dir=~/something on the commandline of the java VM.
> > 	I do not know which should be preferred, but my feeling says me that
> > "" eqalsto System.getProperty("user.dir")should be preffered.
> > 	However there might be a problem is the access to System.getProperty
> > is restricted or if this property is not defined.
> > 	I suggest therefore something like:
> > 
> > 	String FileName=null;
> > 	try{
> > 		FileName=System.getProperty("user.dir",null);
> > 	}catch(SecurityException sec){}
> > 	if (FileName==null)
> > 		FileName=".";
> > 	...
> I should admit I had missed these aspects. In any case the expansion
> happen when interpreting classpath and sys.path, so "." is preferable.
	[Schmidmeier, Arno]  I see here a problem with a bug in old versions
of JPython.
	I do not know if it is fixed. It is pretty likely that . is on the
classpath and also on sys.path.
	Older Versions of Jpython require a totally distinct classpath and
sys.path.
	I am a bit scarry, only if this bug still exist, 
	that you might add unwanted . to the classpath or sys.path.
	If this bug is fixed . is ok for me.

> But I really worry about what happens on a Mac?
> Unfortunately I have no access to such a machine and its jvms.
> 
>  	[Schmidmeier, Arno]  
> > 	I do not understand this topic fully, I want to ask is following
> > scenario possible:
> > 	I have a java-package X with classes a and b.
> > 	on the classpath (sys or jython) there are two distinct entries
> > j1.jar and j2.jar. 
> > 	j1.jar contains a and j2.jar contains b. Is this possible?
> > 	Same for .py filesand directories.
> 
> The semantics for py modules will remain as in python, if one not plays
> with 
> __path__ a py packages span over just one directory and its subdirs.
> sys.path can contain only dirs (an extension for allowing zips and jars
> should be not too difficult but it depends on what happens on CPython side
> and for the moment I like to see what happen with the patch and get rid of
> old and (hopefully few) new bugs, extending things should be easier
> with a clean (so I hope) codebase.
	[Schmidmeier, Arno]  I agree, extending things is much easier with a
clean codebase.
	I vote for cleaning up first, then adding features.
> The support for relative imports inside py pkgs should work as in CPython.
> This should still be fixed in jythonc.
> Concerning java with the patch:
> * classpath is virtually extended with sys.path
> classpath take over sys.path.
> * py pkgs take over java pkgs, but the java classes in the shadowed pkg
>  can be accessed through the py package
	[Schmidmeier, Arno]  I am a bit scary about this. Will following
scenario work, with this design and implementation:
	I subclass a Java-Interface with Jython and create an instance a. 
	(Jython needs to load the Java Interface), it is currently not in
the Java VM. (instance of java.lang.Class called IJP
	A second Java Thread needs the class for the Java-Interface so it
query the Java-Classloader after the interface, the classloader returns IJ.
	Jython passes A to a Java-Object which was loaded by the
Java-Classloader.
	Now is IJP==IJ?

> * dir and from import * for a java package now should really consider
> the union of everything that's is in the dirs/jar on classpath and in the
> dirs on sys.path.
> 
> regards, Samuele
> 
> 
	[Schmidmeier, Arno]  
	regards
		Arno