#3540 causes RecursionError in particular draw pattern

[rsokl]

#3540 introduces a change that causes a RecursionError in MyGrad's test suite.

This is a weird bug. It seems to be specific to drawing from an array-based strategy that spans multiple lines and that includes an identity mapping (I said it was weird!).

The following tests pass (note the xfail)

import pytest

import hypothesis.extra.numpy as hnp
import hypothesis.strategies as st
from hypothesis import given

@given(st.data())
def test_one_line_identity_passes(data):
    x = data.draw(hnp.arrays(shape=(2,), dtype=float).map(lambda x: x))


@pytest.mark.xfail(raises=RecursionError, strict=True)
@given(st.data())
def test_multiline_identity_fails(data):
    x = data.draw(
        hnp.arrays(
            shape=(2,),
            dtype=float,
        ).map(lambda x: x)
    )


@given(st.data())
def test_multiline_non_identity_passes(data):
    x = data.draw(
        hnp.arrays(
            shape=(2,),
            dtype=float,
        ).map(lambda x: 1 * x)
    )
    
    
@given(st.data())
def test_multiline_non_array_passes(data):
    x = data.draw(
        st.lists(
            st.integers(
                1,
                5,
            )
        ).map(lambda x: x)
    )

[Zac-HD]

Wow, finally sniped this and feel kinda stupid: the following call can recurse, but doesn't copy the stack used for cycle-detection.

hypothesis/hypothesis-python/src/hypothesis/vendor/pretty.py

Lines 359 to 361 in 48165d9

	p = RepresentationPrinter()
	p.known_object_printers = self.known_object_printers
	p.repr_call("_" * self.output_width, args, kwargs, force_split=False)

Bleeding obvious once you see it, and it's a single-line fix🤦‍♂️