Can we consider optionally disabling automatic repr-printing and jsonification when TESTCASE_CALLBACKS are present?

[joshuafcole]

Thanks for Hypothesis, it's awesome!

I have a REPL set up for continuously testing the program I'm working on using some hypothesis tests. The inputs are fairly large dataclasses that are unfortunately self-referential.

This works well in general, but when I attempt to tap result generation via a callable in TESTCASE_CALLBACKS, things slow down or crash due to blowing the stack. I have my own pretty-printers that I'd like to call from the repl experience, so I don't have any use for the jsonable or pprinted inputs. I checked over the code to see if I could find a flag to override the behavior (or return my own value for either) but I couldn't find any reasonable solution.

Would you be willing to consider making that behavior optionable via a flag or bump the to_json check earlier in the to_jsonable resolution chain?

I'd be happy to patch it to save you the time if a solution jumps out at you as a good one.

Thanks!
Josh

to_jsonable

only attempts user-configurable to_json if all else fails

impossible to prevent jsonification when TESTCASE_CALLBACKS are present

[Zac-HD]

I'd prefer not to add another config option, but would be happy to shift the .to_json() check to before the attrs/dataclasses/pydantic block in to_jsonable(). PR welcome for that!

You can also improve the interaction with pretty-printing downstream; we support the IPython protocol and/or you can shove a callback directly into the registry if you'd prefer.

[joshuafcole]

I'd prefer not to add another config option

Understandable!

happy to shift the .to_json() check to before the attrs/dataclasses/pydantic block in to_jsonable(). PR welcome for that!

Sure thing, I'll get one cooking today!

You can also improve the interaction with pretty-printing downstream; we support the IPython protocol and/or you can shove a callback directly into the registry if you'd prefer.

That's a solid idea, thanks!