lazily imports more modules for performance

[dmalan]

Improves performance of using just get_* by lazily loading more of SQL and Flask. E.g., on a recent MBP, reduced performance of

import cs50

i = cs50.get_int()
print(i)

from ~500ms to ~300ms.

Cf. https://wiki.python.org/moin/PythonSpeed/PerformanceTips#Import_Statement_Overhead

We could alternatively break a deliberate abstraction barrier and hardcode awareness of / support for Flask inside of SQL, which might further delay expensive imports.

[Jelleas]

Unless I'm missing something, all this does is alias cs50.sql to cs50.SQL. Flask indirectly import cs50.sql anyway, so does this not help in lazy loading.

[Jelleas]

At this point cs50.flask has already imported cs50.sql, so there's not much to gain

[Jelleas]

There is indeed a significant overhead. From testing it's Flask adding most of the work, then SQLAlchemy and to a very small degree sqlparse. Lazily importing just those shrinks the import time oncs50 from 300ms to 50ms for me. Lazily importing anything else did not make a difference.

It is cs50.flask that adds the most overhead. Sadly I see no way around it, because overwriting the logger is a task that you cannot do lazily. But, is this a problem that needs a hack? I'd say the Pythonic way is to import a submodule separately if it does not need to be loaded. Such as matplotlib and matplotlib.pyplot, or even check50 and check50.c ;). You would just import cs50 as normal for get_* and such, and import cs50.flask once you are working with flask. Conceptually this makes the most sense to me, because they are actual separate things.

[kzidane]

Also as an aside, I think both find_module and load_module are deprecated fwiw.

[Jelleas]

@kzidane you're right! (https://docs.python.org/3/library/importlib.html#importlib.abc.Loader.load_module)