ydylla

I also stumbled across this and agree with mwalser the error message was very irritating. Please mention it in the docs or better fix it 😄 This also happens when...

Yes, I did not want to add instance checks for the deserialization, to keep it as fast as possible. One could workaround this problem by listing the most complex types...

Yes reordering should be possible. It should happen around [here](https://github.com/yukinarit/pyserde/blob/f5c03243e2b988848162249ce979effdcc5d9e11/serde/de.py#L638) before we pass `union_args` into the template. But we need to find a good metric (sort logic). Maybe other dataclasses...

@yukinarit I don't remember why 😃 Probably just because it previously also was at runtime see [431c914 @ de.py:L377](https://github.com/yukinarit/pyserde/commit/431c9146e31210a4d3e999dd2456667a1ea3acb0#diff-733dfe107c23fc8e53eb1400ac38c83b92253885303cdc5d1ca0d679ddb98046L377) which called into the removed `more_types.py`. I agree it would probably...

During testing with a real config I noticed that my 0 bytes fix was flawed, I force pushed a better one.

> I wonder if we can use the MatcherSet without having to pass in a logger Would you prefer it if `layer4.Connection` had a public logger? I already did this...

> Maybe that's a good idea! Can we try it? Sure, I will change it tomorrow.

@mholt I have added a logger to the layer4 connection and also removed the `caddy.Module` requirement for `ConnMatcher`.

Hi @mholt, I am still interested to see this merged. What is missing here to move this forward?

I added some more tests and docs. It should be ready to be merged now.