[Jython-dev] PyShadowString test, ideas, questions

[Jeff A. <ja...@fa...>]

I've added a test for PyShadowString, our string that can have two 
values at once. Trying to design a test raises some questions.

 >>> s = PyShadowString("hello", "bonjour")
 >>> s == 'bonjour'
False
 >>> s.addtarget(r"org\.python\.util\.InteractiveConsole")
 >>> s == 'bonjour'
True

So far so good.

1. Should I be able to do this?

 >>> s.gettargets().pop()
('org\\.python\\.util\\.InteractiveConsole',)
 >>> s == 'bonjour'
False

2. This seems like a useful "unconditional", but I wonder if it is harmful:

 >>> s.addtarget(None)
 >>> s == 'bonjour'
True

3. Confusing subject, but I'm not sure the exposure of __eq__ is 
"canonical". I think the convention is __eq__ just wraps 
shadowstr___eq__, as done for startswith and __repr__, but here there is 
"added value" in __eq__ (also in PyString.__eq__). shadowstr___eq__ will 
be exposed as __eq__, and shouldn't PyShadowString.__eq__ do the same 
thing? What we need is complicated by the wrapper PyObject._eq.

4. Single-stepping through __eq__ I notice that a call like s == 
'bonjour' (i.e. a call to __eq__) is quite costly, even when there are 
no targets that match. The problem is in isTarget() and so I think I 
will try changing the logic to:

return other==string || (other==shadow && isTarget())

but also, we can make isTarget() cheaper.

5. I tried the tests of str on PyShadowString, to prove it is also a 
str, but they fail because it cannot be sub-classed. Is there a reason 
to prevent sub-classing?

6. And relatedly, when I was working on PyType I noticed that it is 
possible to expose a Java sub-class as the same Python type as its 
parent. This would mean that type(PyShadowString("a", "b")) would be str 
even though the behaviour is different. This sounds worth trying.

Jeff

-- 
Jeff Allen