Type safety/annotation in fast ai library

gballardin · March 27, 2019, 7:08pm

What is the effect of the typed signature in the fast ai library? E.g.

fastai/fastai/blob/master/fastai/core.py#L118


else:
    try: a = len(p)
    except: p = [p]
n = q if type(q)==int else len(p) if q is None else len(q)
if len(p)==1: p = p * n
assert len(p)==n, f'List len mismatch ({len(p)} vs {n})'
return list(p)


_camel_re1 = re.compile('(.)([A-Z][a-z]+)')
_camel_re2 = re.compile('([a-z0-9])([A-Z])')
def camel2snake(name:str)->str:
"Change `name` from camel to snake style."
s1 = re.sub(_camel_re1, r'\1_\2', name)
return re.sub(_camel_re2, r'\1_\2', s1).lower()


def even_mults(start:float, stop:float, n:int)->np.ndarray:
"Build log-stepped array from `start` to `stop` in `n` steps."
mult = stop/start
step = mult**(1/(n-1))
return np.array([start*(step**i) for i in range(n)])

Is it to make functions/methods actually strongly typed, or is it more like type annotation?

sgugger · March 27, 2019, 7:16pm

It’s just for type annotation, to help the reader understand what the function takes and returns.

gballardin · March 27, 2019, 8:02pm

Got it.

@sgugger since I know you’ve been actively developing the fast ai library, I was curious to know whether you’ve found the lack of strict type safety to be a hinderance in maintaining and developing further this code base given that it has reached a decent level of complexity at this point. Another way to put it is, how much of your dev/maintenance work could be prevented if Python was strongly typed?

sgugger · March 27, 2019, 8:56pm

It’s not really been a problem, no. Bugs I see are rarely related to types so I don’t think Python being strongly typed would help us much.