lists with unique_by → one_of → keyword-only composite = TypeError

[SydneyUni-Jim]

Hypothesis 4.24.0, pytest 4.6.0, Python 3.7.3, macOS 10.14.5

A TypeError is raised when a lists strategy with the unique_by argument is chained to a one_of strategy chained to a composite strategy that only has keyword-only arguments.

The following code demonstrates the issue. It raises TypeError: g1() got an unexpected keyword argument 'kwarg1'.

from hypothesis import given
import hypothesis.strategies as st

@st.composite
def g1(draw, *, kwarg1=None):
  return draw(
    st.fixed_dictionaries({'kwarg1': st.just(kwarg1), 'i': st.integers()})
  )

def g2(*, kwarg1=None):
  return st.one_of(g1(kwarg1=kwarg1))

def g3(*, kwarg1=None):
  return st.lists(
    g2(kwarg1=kwarg1),
    min_size=1,
    unique_by=lambda x: x['i']
  )

@given(g3())
def test_1(a):
   assert True

The TypeError is being raised by convert_keyword_arguments in reflection.py.

    if kwargs and not argspec.varkw:
        if len(kwargs) > 1:
            raise TypeError(
                "%s() got unexpected keyword arguments %s"
                % (function.__name__, ", ".join(map(repr, kwargs)))
            )

When the TypeError is raised, convert_keyword_arguments is being called (indirectly) by event_to_string in engine.py as it tries to str a lazyformat object that has a __format_string of 'Retried draw from %r to satisfy filter'.

After applying any one of the following patches to the demonstration code, the TypeError is no longer raised.

Don't use a keyword-only argument in the composite.

@@ -5 +5 @@
-def g1(draw, *, kwarg1=None):
+def g1(draw, kwarg1=None):

Don't use a composite at the end of the chain.

@@ -4,3 +4,2 @@
-@st.composite
-def g1(draw, *, kwarg1=None):
-  return draw(
+def g1(*, kwarg1=None):
+  return (

Don't use one_of in the middle of the chain.

@@ -11 +11 @@
-  return st.one_of(g1(kwarg1=kwarg1))
+  return g1(kwarg1=kwarg1)

Don't use unique_by in the lists stratergy.

@@ -17 +17 @@
-    unique_by=lambda x: x['i']
+    # unique_by=lambda x: x['i']

[Zac-HD]

Wow! Thanks for the fantastic report - I imagine it took a while to track down. It's also detailed enough that I narrowed it down to the following patch pretty quickly:

diff --git a/hypothesis-python/src/hypothesis/internal/reflection.py b/hypothesis-python/src/hypothesis/internal/reflection.py
index ef070ab..3a157a0 100644
--- a/hypothesis-python/src/hypothesis/internal/reflection.py
+++ b/hypothesis-python/src/hypothesis/internal/reflection.py
@@ -182,7 +182,7 @@ def convert_keyword_arguments(function, args, kwargs):
             else:
                 raise TypeError("No value provided for argument %r" % (arg_name))

-    if kwargs and not argspec.varkw:
+    if kwargs and not (argspec.varkw or argspec.kwonlyargs):
         if len(kwargs) > 1:
             raise TypeError(
                 "%s() got unexpected keyword arguments %s"

If this fixes it for you, would you like to open a PR on that basis?

[SydneyUni-Jim]

That does fix my tests.

I'm new to both property-based testing and Hypothesis. So I wanted to make sure it wasn't something I was doing wrong. And I was happy to get off the quick sand for a bit.

I'd be to organise a PR. But this something you've solved, not me.

[Zac-HD]

Awesome! All you'll need to do for a pr: apply the patch above, put the test case in a new file in tests/nocover/ (with a descriptive name), and add a new file hypothesis-python/RELEASE.rst with the following contents:

RELEASE_TYPE: patch 

This patch fixes :issue:`1999`, a spurious bug raised when a :func:`@st.composite <hypothesis.strategies.composite>` function was passed a keyword-only argument. 

Thanks to Jim Nicholls for his fantastic bug report.