Builtins.lazy for lazy arguments

[dg-pb]

@xitop

I find the usefulness quite limited, because lazy can wrap only a single call. Even a small change is problematic:

v = dct.get(k, default=lazy(factorial, 100_000) + 1) # error

Good point, but lazy isn’t purposed for that. Its sole purpose is to wrap a callable and signal argument being lazy. It can be used in conjunction with other tools to achieve what you have pointed out. Something along the lines of Evaluation graphs would do. E.g.:

v = dct.get(k, default=lazy(dask.deferred(factorial, 100_000) + 1)

Or given the fact, that this case is trivial, simple lambda can do:

v = dct.get(k, default=lazy(lambda: factorial(100_000) + 1))

@xitop

Expressions would require lazy-aware versions of operator.XXX (and some infix → prefix rewriting):

v = dct.get(k, default=lazy(operator.add, lazy(factorial, 100_000), 1))

That is also an option.
In this specific case one could do with something like a pipe instead (see: `functools.pipe`. It didn’t go through though):

v = dct.get(k, default=lazy(pipe(factorial, partial(add, 1)), 1))

But the main point is that for complex cases there are many tools to be used in conjunction.

@xitop

I would then prefer:

v = dct[k] if k in dct else factorial(100_000) + 1

This is a good option, however it only works for this specific case of “default return value argument”, but lazy is more generic, e.g.:

def ifelse(cond, a, b):
    result = a if cond else b
    if isinstance(result, lazy):
        result = result()
    return result