Jimmy Jia

Hmm, I don't see a way to adjust the cost of `__typename` in the current API, actually. Do you have an idea for how this sort of API should look?

With Apollo Server, creating a new validation rule per request is tricky. In general, it's also not easy to get the relevant variables for a batch request. I'm going to...

Is this purely because the parameters are decoded with e.g. ``` python variable['min'] + u[index]*(variable['max']-variable['min']) ```

I did some casual profiling a while ago... it looks like you need to strip away a ridiculous amount of logic to get to toasted-marshmallow-level performance. Just stuff like inlining...

@deckar01 Hmm, I think in general, most of our time spent in Marshmallow is actually for doing a `many` dump on list endpoints. I guess I'm just a little surprised...

I thought they were working off a v3 beta. I'm just really surprised at the differences here. If anything I feel like they're worth understanding for their own sake. This...

well, `ma.utils.get_value` does a lot more than just `getattr`, anyway. dotted getter notation and all that. if serialization is actually 5x faster than it used to be, then maybe there's...

The complexity is that the accessor is a function of both the object and the field name. The same schema with a `name` field currently can handle both an instance...

Yup, exactly. You'd need to cache/memoize on the type of the object being serialized to be reasonably safe.

@deckar01 Sure, so I think a starting point would be to just memoize the last accessor based on the `type` of the object being dumped.