Is it possible to lazy load the repr handlers?

[laike9m]

When profiling my program, I find that importing cheap_repr is very expensive. Here's one profiling result using py-spy:

https://laike9m.github.io/images/af9b0d4.svg

Importing cheap_repr took ~8% of the total time. In __init__.py, it seems that it should be possible to make @try_register_repr calls lazy. Especially for registering Pandas handlers, they took most of the time. This becomes worse when users have Pandas installed but didn't really call cheap_repr on Pandas objects.

[alexmojaki]

I am interested in this (#3) but I don't know the best way to do it or how easy it would be. Happy to accept a PR. Since it seems like it requires an import hook it could be a Python 3 only enhancement.

[laike9m]

I think it would be helpful to write down what you have in mind about import hooks, and we can see if somebody will have time to work on this.

[alexmojaki]

I'm imagining an import hook that is triggered whenever a module name that was registered with try_register_repr is imported for the first time, and only then is the repr registered against the actual class. So instead of importing pandas, it waits until someone else imports pandas, and at that moment registers the reprs for DataFrame etc.

it seems that it should be possible to make @try_register_repr calls lazy

What did you have in mind?

[laike9m]

I was more thinking of doing reigsitering when it is actually used

[alexmojaki]

How?

[laike9m]

I don't know if that's possible, I'll have to take a closer look at the code. It's just a thought for now.