Fix event symbolics cache

.github/workflows/main.yml

@@ -204,8 +204,7 @@ jobs:
         pip install -r requirements/crosshair.txt
         pip install hypothesis-python/
     - name: Run tests
-      # remove this ignore when https://github.com/pschanely/hypothesis-crosshair/issues/40 is fixed
-      run: python -m pytest -W "ignore:hypothesis.internal.observability.TESTCASE_CALLBACKS is deprecated" --numprocesses auto ${{ matrix.whichtests == 'nocover' && 'hypothesis-python/tests/nocover' || 'hypothesis-python/tests/ --ignore=hypothesis-python/tests/nocover/ --ignore=hypothesis-python/tests/quality/ --ignore=hypothesis-python/tests/ghostwriter/ --ignore=hypothesis-python/tests/patching/' }}
+      run: python -m pytest --numprocesses auto ${{ matrix.whichtests == 'nocover' && 'hypothesis-python/tests/nocover' || 'hypothesis-python/tests/ --ignore=hypothesis-python/tests/nocover/ --ignore=hypothesis-python/tests/quality/ --ignore=hypothesis-python/tests/ghostwriter/ --ignore=hypothesis-python/tests/patching/' }}
 
   test-osx:
     runs-on: macos-latest

hypothesis-python/RELEASE.rst

@@ -0,0 +1,3 @@
+RELEASE_TYPE: patch
+
+Fix a bug where we persisted symbolics from solver-based :ref:`alternative backends <alternative-backends>` in |event|.

hypothesis-python/pyproject.toml

@@ -111,7 +111,7 @@ pandas = ["pandas>=1.1"]
 pytest = ["pytest>=4.6"]
 dpcontracts = ["dpcontracts>=0.4"]
 redis = ["redis>=3.0.0"]
-crosshair = ["hypothesis-crosshair>=0.0.24", "crosshair-tool>=0.0.93"]
+crosshair = ["hypothesis-crosshair>=0.0.25", "crosshair-tool>=0.0.93"]
 # zoneinfo is an odd one: every dependency is platform-conditional.
 zoneinfo = ["tzdata>=2025.2; sys_platform == 'win32' or sys_platform == 'emscripten'"]
 # We only support Django versions with upstream support - see
@@ -121,7 +121,7 @@ zoneinfo = ["tzdata>=2025.2; sys_platform == 'win32' or sys_platform == 'emscrip
 django = ["django>=4.2"]
 watchdog = ["watchdog>=4.0.0"]
 # Avoid changing this by hand. This is automatically updated by update_changelog_and_version
-all = ["black>=20.8b0", "click>=7.0", "crosshair-tool>=0.0.93", "django>=4.2", "dpcontracts>=0.4", "hypothesis-crosshair>=0.0.24", "lark>=0.10.1", "libcst>=0.3.16", "numpy>=1.19.3", "pandas>=1.1", "pytest>=4.6", "python-dateutil>=1.4", "pytz>=2014.1", "redis>=3.0.0", "rich>=9.0.0", "tzdata>=2025.2; sys_platform == 'win32' or sys_platform == 'emscripten'", "watchdog>=4.0.0"]
+all = ["black>=20.8b0", "click>=7.0", "crosshair-tool>=0.0.93", "django>=4.2", "dpcontracts>=0.4", "hypothesis-crosshair>=0.0.25", "lark>=0.10.1", "libcst>=0.3.16", "numpy>=1.19.3", "pandas>=1.1", "pytest>=4.6", "python-dateutil>=1.4", "pytz>=2014.1", "redis>=3.0.0", "rich>=9.0.0", "tzdata>=2025.2; sys_platform == 'win32' or sys_platform == 'emscripten'", "watchdog>=4.0.0"]
 
 [tool.setuptools.dynamic]
 version = {attr = "hypothesis.version.__version__"}

hypothesis-python/src/hypothesis/control.py

@@ -247,22 +247,34 @@ def event(value: str, payload: Union[str, int, float] = "") -> None:
     """
     context = _current_build_context.value
     if context is None:
-        raise InvalidArgument("Cannot make record events outside of a test")
+        raise InvalidArgument("Cannot record events outside of a test")
 
-    payload = _event_to_string(payload, (str, int, float))
-    context.data.events[_event_to_string(value)] = payload
+    avoid_realization = context.data.provider.avoid_realization
+    payload = _event_to_string(
+        payload, allowed_types=(str, int, float), avoid_realization=avoid_realization
+    )
+    value = _event_to_string(value, avoid_realization=avoid_realization)
+    context.data.events[value] = payload
 
 
 _events_to_strings: WeakKeyDictionary = WeakKeyDictionary()
 
 
-def _event_to_string(event, allowed_types=str):
+def _event_to_string(event, *, allowed_types=str, avoid_realization):
     if isinstance(event, allowed_types):
         return event
+
+    # _events_to_strings is a cache which persists across iterations, causing
+    # problems for symbolic backends. see
+    # https://github.com/pschanely/hypothesis-crosshair/issues/41
+    if avoid_realization:
+        return str(event)
+
     try:
         return _events_to_strings[event]
     except (KeyError, TypeError):
         pass
+
     result = str(event)
     try:
         _events_to_strings[event] = result

hypothesis-python/src/hypothesis/core.py

@@ -1308,6 +1308,16 @@ def _execute_once_for_engine(self, data: ConjectureData) -> None:
         finally:
             # Conditional here so we can save some time constructing the payload; in
             # other cases (without coverage) it's cheap enough to do that regardless.
+            #
+            # Note that we have to unconditionally realize data.events, because
+            # the statistics reported by the pytest plugin use a different flow
+            # than observability, but still access symbolic events.
+
+            try:
+                data.events = data.provider.realize(data.events)
+            except BackendCannotProceed:
+                data.events = {}
+
             if observability_enabled():
                 if runner := getattr(self, "_runner", None):
                     phase = runner._current_phase
@@ -1332,11 +1342,6 @@ def _execute_once_for_engine(self, data: ConjectureData) -> None:
                 except BackendCannotProceed:
                     self._string_repr = "<backend failed to realize symbolic arguments>"
 
-                try:
-                    data.events = data.provider.realize(data.events)
-                except BackendCannotProceed:
-                    data.events = {}
-
                 data.freeze()
                 tc = make_testcase(
                     run_start=self._start_timestamp,

hypothesis-python/src/hypothesis/internal/conjecture/engine.py

@@ -556,6 +556,10 @@ def test_function(self, data: ConjectureData) -> None:
             if not interrupted:  # pragma: no branch
                 assert data.cannot_proceed_scope is None
                 data.freeze()
+
+                if self.settings.backend != "hypothesis":
+                    realize_choices(data, for_failure=data.status is Status.INTERESTING)
+
                 call_stats: CallStats = {
                     "status": data.status.name.lower(),
                     "runtime": data.finish_time - data.start_time,
@@ -566,8 +570,6 @@ def test_function(self, data: ConjectureData) -> None:
                     ),
                 }
                 self.stats_per_test_case.append(call_stats)
-                if self.settings.backend != "hypothesis":
-                    realize_choices(data, for_failure=data.status is Status.INTERESTING)
 
                 self._cache(data)
                 if data.misaligned_at is not None:  # pragma: no branch # coverage bug?

hypothesis-python/tests/cover/test_control.py

@@ -232,4 +232,5 @@ def step(self):
 
 
 def test_can_convert_non_weakref_types_to_event_strings():
-    _event_to_string(())
+    _event_to_string((), avoid_realization=True)
+    _event_to_string((), avoid_realization=False)

hypothesis-python/tests/cover/test_statistical_events.py

@@ -26,6 +26,7 @@
     strategies as st,
     target,
 )
+from hypothesis.control import current_build_context
 from hypothesis.statistics import collector, describe_statistics
 
 from tests.common.utils import Why, xfail_on_crosshair
@@ -92,6 +93,8 @@ def test_formats_are_evaluated_only_once():
 
     @given(st.integers())
     def test(i):
+        if current_build_context().data.provider.avoid_realization:
+            pytest.skip("event() cache is disabled under avoid_realization = True")
         event(Foo())
 
     stats = call_for_statistics(test)
@@ -254,8 +257,6 @@ def threshold(error):
     assert stats["targets"]["error"] > 10
 
 
-# describe_statistics causes not-deterministic crosshair errors for some reason?
-@xfail_on_crosshair(Why.other)
 def test_statistics_with_events_and_target():
     @given(st.integers(0, 10_000))
     def test(value):

hypothesis-python/tests/cover/test_validation.py

@@ -13,7 +13,7 @@
 
 import pytest
 
-from hypothesis import find, given, settings, strategies as st
+from hypothesis import find, given, strategies as st
 from hypothesis.errors import HypothesisWarning, InvalidArgument
 from hypothesis.internal.validation import check_type
 from hypothesis.strategies import (
@@ -278,14 +278,6 @@ def test_check_strategy_might_suggest_sampled_from():
     check_strategy_(integers(), "passes for our custom coverage check")
 
 
-@pytest.mark.xfail(
-    settings._current_profile == "crosshair",
-    reason=(
-        "conflicting warning filters. Remove when "
-        "https://github.com/pschanely/hypothesis-crosshair/issues/42 "
-        "is fixed"
-    ),
-)
 @pytest.mark.parametrize("codec", ["ascii", "utf-8"])
 def test_warn_on_strings_matching_common_codecs(codec):
     with pytest.warns(

hypothesis-python/tests/crosshair/test_crosshair.py

@@ -232,3 +232,11 @@ def test(n):
         test()
 
     assert saw_myevent
+
+
+# see https://github.com/pschanely/hypothesis-crosshair/issues/41
+@given(st.integers())
+@settings(backend="crosshair")
+def test_event_with_realization(value):
+    event(value)
+    float(value)

hypothesis-python/tox.ini

@@ -159,8 +159,6 @@ allowlist_externals =
     bash
 setenv=
     HYPOTHESIS_PROFILE=crosshair
-    # remove this ignore when https://github.com/pschanely/hypothesis-crosshair/issues/40 is fixed
-    PYTEST_ADDOPTS=-W "ignore:hypothesis.internal.observability.TESTCASE_CALLBACKS is deprecated"
 commands =
     # invoke with `./build.sh check-crosshair-cover -- -x -Wignore`
     cover: python -bb -X dev -m pytest -n auto tests/cover/ tests/pytest/ {posargs}

requirements/crosshair.txt

@@ -26,9 +26,9 @@ exceptiongroup==1.2.2 ; python_version < "3.11"
     #   pytest
 execnet==2.1.1
     # via pytest-xdist
-hypothesis==6.135.14
+hypothesis==6.137.0
     # via hypothesis-crosshair
-hypothesis-crosshair==0.0.24
+hypothesis-crosshair==0.0.25
     # via -r requirements/crosshair.in
 importlib-metadata==8.6.1
     # via crosshair-tool

tooling/src/hypothesistooling/__main__.py

@@ -576,7 +576,7 @@ def run_tox(task, version, *args):
     "3.12": "3.12.11",
     "3.13": "3.13.5",
     "3.13t": "3.13t-dev",
-    "3.14": "3.14.0rc1",
+    "3.14": "3.14.0rc2",
     "3.14t": "3.14t-dev",
     "3.15": "3.15-dev",
     "3.15t": "3.15t-dev",