Use iterable range without conversion to list as domain in add_rand examples

[mariushegele]

The add_rand method of the crv.Randomized class expects an iterable domain. All given examples transform iterable ranges to lists, e.g.:

def Point(crv.Randomized):
   def __init__(self, x, y):
        # ...
        self.add_rand('x', list(range(10)))

Why is this transformation necessary? It is not scalable to large ranges, because Python will construct the whole list in memory. The use of domain within constraint resolving appears to be only iteration. This works on ranges even more efficiently:

$ python3 -m timeit 'for i in range(0xFFFF): pass'
50 loops, best of 5: 8.57 msec per loop
$ python3 -m timeit 'for i in list(range(0xFFFF)): pass'
20 loops, best of 5: 11 msec per loop

[mariushegele]

https://github.com/mciepluc/cocotb-coverage/blob/e4b171189a84b2c14daa619093e7155a6595c852/cocotb_coverage/crv.py#L569

appears to be a related instance:

        # Cartesian product of above
        ducSolutions = list(itertools.product(*ducDomains))

        # merge solutions: constrained ones and all possible distribution 
        # values
        for sol in solutions:
            for ducsol in ducSolutions:
            # ...

[mciepluc]

Hmm, I don't remember just now what was the reason behind. Could be a Python 2-related stuff (which does not matter at the moment) or some internal conversion requirement. Anyway, if you need to create a big domain, it is not recommented to do it using add_rand. Please use pre/post_randomize instead if you see performance issues. Have a look here.

[mariushegele]

I see the point, thank you for the link!