Re: [Jython-users] Strange behavior of Integer and Double

[<bc...@wo...>]

[Will]

>I have run into some surprising behavior of Java Integer objects while using
>jython:
><example snipped>
>What am I doing wrong?  How can I keep Integer and Double objects from being
>converted to Java primitives?

[Finn]

> You can't. The eagerness with which Jython converts the java.lang.*
> wrapper classes to instances of native python classes have been
> discussed previousely:
>
> http://mail.python.org/pipermail/jpython-interest/1999-December/002590.html

[Ben Hutchison]

>My main concern is: will this obstruct the use of Java APIs which require Boolean
>parameters? In the example William gives the expected method is not invoked. Will
>developers using Jython to script or integrate existing Java systems encounter
>problems; either hard to find bugs, or method signatures that cant be invoked?
>
>> It is basicly the same problem you have re-discovered. I'm still not
>> convinved that it is a net win to reduce this eagerness.
>>
>> http://mail.python.org/pipermail/jpython-interest/1999-December/002596.html
>
>From the above article, Finn Bock wrote:
>
>     One of the downsides of this patch, is that the simple if statement
>     below will no longer work. I am afraid to think about the
>     incompabilities this change will introduce in existing jpython
>     programs.
>
>     if not lang.Boolean.FALSE:
>        print "FALSE"
>
>Equivalently, the statement below is not legal Java. 

So? <0.7 wink>

>Is there need to support semantics that arent in Java? 

Yes. One of the problem, as I see it, is that we can evaluate the
booleaness of a lot of objects. Testing lists and dicts are done all the
time:

   l = ['a']
   if l: print "list is not empty"

Since that is possible, common sense dictate that testing a j.l.Boolean
in an "if" statement should actually test the value the Boolean is
wrapping. 

The same extends to arithmetic on j.l.Number subclasses, where is should
still be possible to add a j.l.Short and 42 together.

>I cant imagine many developers using expressions
>similar to above, since they couldnt in Java anyway.
>
>     if (! Boolean.FALSE) {}

This lack of flexibility in java will not convince me that jython don't
need it either <wink>.

I said that I will not spearhead this change. I think jython can be
modified in such a way that the wrapper objects are keept as java
objects but still maintain their usefullness when used from a python
script. It's just not something I will do.

>As I understand the issues, I *think* I prefer a less eager conversion policy,
>but would be happy to hear more counter-arguments.

I would be more happy to see a patch <wink>.

regards,
finn