[Python-ideas] Trailing-based state restoration

[Andrew Barnert]

On Dec 17, 2015, at 09:50, Andrew Barnert via Python-ideas <python-ideas at python.org> wrote:
> 
>> On Dec 17, 2015, at 09:12, SAVEANT Pierre <pierre.saveant at thalesgroup.com> wrote:
>> 
>> Hi,
>>  
>> I got a proposal for a new feature in Python in order to perform hypothetical reasoning in a very simple way.
>> Hypothetical reasoning is mainly used in problem solving search algorithms.
>>  
>> In such a process, variables are temporally assigned and later restored to their preceding value.
>> With the assumption that only a small fraction of the variables changes at the same time, it is more efficient to store undo information instead of copying the entire state.
>> The basic data structure to handle these moves efficiently is called a trail which is mainly a stack with markers.
>> Potentially this mechanism can be provided to any object type of Python.
>> Only few lines are needed to implement the mechanism using the reflexive power of Python.
>> A deeper implementation could hide the process in the internal data representation in order to offer this new feature in a seamless way.
>>  
>> This new feature is activated with the specification of thee primitives:
>> - push()
>> - back()
>> - assign(object, attribute, value)
>>  
>> push : declare a new hypothetical state.
>> back : restore the previous state.
>> assign(object, attribute, value) : assign the value to the attribute of the object in the current state.
>>  
>> In addition a primitive can be provided to go back to a specific state:
>> backtrack(n): restore the state number n.
>>  
>> EXAMPLES
>>  
>> class Var:
>>     def __init__(self, value):
>>         self.value = value
>>     def __str__(self): return"{}".format(self.value)
>>  
>> def test1():
>>     V1 = Var(10)
>>     assert V1.value == 10
>>     push()
>>     assign(V1, 'value', 100)
>>     assert V1.value == 100
>>     back()
>>     assert V1.value == 10
>>  
>> def test2():
>>     V1 = Var(0)
>>     for i in range(8):
>>         push()
>>         for j in range(5):
>>             assign(V1, 'value',(j+1)+i*5-1)
>>     assert V1.value == 39
>>     assert current() == 8
>>    backtrack(6)
>>     assert V1.value == 29
>>     assert current() == 6
>>     backtrack(4)
>>     assert V1.value == 19
>>     assert current() == 4
>>     backtrack(0)
>>     assert V1.value == 0
>>     assert current() == 0
>>  
>> The assign procedure is used in place of the standard assignment denoted by the "=" operator.
> 
> It seems like all of that can be implemented in Python today (presumably storing the set of trailed Vars as a global or as a class attribute of Var, since there doesn't seem to be any scoping involved). So, why not just build it and put it on PyPI?
> 
>> In a seamless integration the "=" assignment operator could be overridden when the variable has been previously declared as "recordable" (require a declaration primitive or a type declaration).
> 
> That's one thing that couldn't be implemented in Python today, because = can't be overridden (it's a binding-and-possibly-declaring statement, not an assignment operator, in Python). But if you just made all these things attributes on some object, then you could hook its __setattr__. Plus, you could have different sets of trails in parallel, and implement whatever kind of scoping you want instead of everything being sort of global but sort of Python-scoped, and so on.

Actually, reading things a bit more carefully, that's just as doable. Just define Var.__setattr__ to call assign and you're done. Now "v1.value = 10" gets trailed.

(I'm a bit confused about why the "value" attribute is constructed at __init__ time from the constructor argument, but assign lets you assign any attribute you want even though you only "value". If you only want "value" to be trailed, skip the __setattr__ and make it a @property.)

>> Is anybody interested by this new feature?
>>  
>>  
>> Pierre Savéant
>>  
>>  
>>  
>> _______________________________________________
>> Python-ideas mailing list
>> Python-ideas at python.org
>> https://mail.python.org/mailman/listinfo/python-ideas
>> Code of Conduct: http://python.org/psf/codeofconduct/
> _______________________________________________
> Python-ideas mailing list
> Python-ideas at python.org
> https://mail.python.org/mailman/listinfo/python-ideas
> Code of Conduct: http://python.org/psf/codeofconduct/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-ideas/attachments/20151217/a934e086/attachment.html>