Performance of json_schema generation about mashumaro

[Fatal1ty]

Hi @gshank

But the performance was substantially worse (2x slower)

How did you measure performance? You were running your json_schema method for the same dataclass sequentially, right? If this is the case, then you can simply cache the result once, since it will not change. The easiest way with lru_cache:

from functools import lru_cache

def foo():
    print('run foo!')
    return 42

class Bar:
    @classmethod
    @lru_cache(maxsize=1)
    def json_schema(cls):
        return foo()

print(Bar.json_schema())
print(Bar.json_schema())

# run foo!
# 42
# 42

I see hologram uses a class-level cache when you apply JsonSchemaMixin to the class. This is a wise decision but it will make it impossible to use the same dataclass in multiple root dataclasses with overriden serialization.

[gshank]

This is working. I will compare the performance and see how it does.

[Fatal1ty]

@gshank to make lru_cache do its job for unknown number of subclasses I can suggest you set max_size=None explicitly since it's limited to 128 by default. In my original example it was set to 1, so it didn't work for you.

[Fatal1ty]

@gshank looks like you’ve sorted out the caching problem? Can we close this issue?

[Fatal1ty]

I’m going to close this issue as stale. Feel free to reopen it if needed.