Frank Sachsenheim

while you are correct that conditional normalization with the `oneof`-rule would seem to make sense, i still object this approach because of the inconsistency and additional complexity it introduces. since...

don't worry, i'm short of time as well. > but I didn't like it too much because it always sets a default, so it is not possible to insert a...

yes, this looks intersting. how much testable documents for the schema could you provide? are they all valid according to the schema?

that'd be absoulutely sufficient. could you prepare a module that works [like this](https://github.com/pyeve/cerberus/blob/master/cerberus/benchmarks/test_overall_performance.py)? without the document generation, of course.

could you please open a pull request here? there are handy review features here.

i see that there would be no other way to solve your use-case. would it be a good idea to make the lookup-strategy configurable (also regarding `^`) per validator instance,...

this behaviour is to be expected as `str(None)` is not failing. i'd say a custom coercer is the way to go. would *not coercing* values that are `None` and constrained...

sorry for not being very verbose on this. i gave it some thoughts: the behaviour you describe is to be expected because the current implementation works like this: when a...

okay, i investigated the 'cleaner' solution and it is doable. (if anyone wants to take a peek, i can open a pr.) > would that break current uses? yes. or...

cool, after fixing the tests, only `test_default_none_nullable` is failing. i still would argue that the default setter should be called in any case and it is up to itself to...