Merge pull request #4049 from Zac-HD/more-crosshair-compat

More crosshair compatibility

Author: Zac-HD

.github/workflows/main.yml

@@ -83,7 +83,12 @@ jobs:
           - check-pandas13
           - check-py39-pandas12
           - check-py39-pandas11
-          # - check-crosshair-cover
+          ## `-cover` is too slow under crosshair; use a custom split
+          # - check-crosshair-custom-cover/test_[a-d]*
+          # - check-crosshair-custom-cover/test_[e-i]*
+          # - check-crosshair-custom-cover/test_[j-r]*
+          # - check-crosshair-custom-cover/test_[s-z]*
+          # - check-crosshair-custom-pytest/test_*
           # - check-crosshair-nocover
           # - check-crosshair-niche
           - check-py38-oldestnumpy
@@ -121,7 +126,14 @@ jobs:
           sudo apt-get update && \
           sudo apt-get install -y dotnet-sdk-6.0
     - name: Run tests
-      run: TASK=${{ matrix.task }} ./build.sh
+      run: |
+        export TASK=${{ matrix.task }}
+        if [[ $TASK == check-crosshair-custom-* ]]; then
+          GROUP="${TASK#check-crosshair-custom-}"
+          ./build.sh check-crosshair-custom -- -n auto $(cd hypothesis-python ; echo tests/$GROUP)
+        else
+          ./build.sh
+        fi
     - name: Upload coverage data
       uses: actions/upload-artifact@v2
       # Invoke the magic `always` function to run on both success and failure.

hypothesis-python/RELEASE.rst

@@ -0,0 +1,4 @@
+RELEASE_TYPE: patch
+
+This patch disables :func:`hypothesis.target` on alternative
+backends where it would not work.

hypothesis-python/setup.py

@@ -60,7 +60,7 @@ def local_file(name):
     "pytest": ["pytest>=4.6"],
     "dpcontracts": ["dpcontracts>=0.4"],
     "redis": ["redis>=3.0.0"],
-    "crosshair": ["hypothesis-crosshair>=0.0.7", "crosshair-tool>=0.0.60"],
+    "crosshair": ["hypothesis-crosshair>=0.0.7", "crosshair-tool>=0.0.61"],
     # zoneinfo is an odd one: every dependency is conditional, because they're
     # only necessary on old versions of Python or Windows systems or emscripten.
     "zoneinfo": [

hypothesis-python/src/hypothesis/control.py

@@ -297,6 +297,11 @@ def target(observation: Union[int, float], *, label: str = "") -> Union[int, flo
             "Calling target() outside of a test is invalid.  "
             "Consider guarding this call with `if currently_in_test_context(): ...`"
         )
+    elif context.data.provider.avoid_realization:
+        # We could in principle realize this in the engine, but it seems more
+        # efficient to have our alternative backend optimize it for us.
+        # See e.g. https://github.com/pschanely/hypothesis-crosshair/issues/3
+        return observation  # pragma: no cover
     verbose_report(f"Saw target({observation!r}, {label=})")
 
     if label in context.data.target_observations:

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

@@ -2273,10 +2273,9 @@ def draw_boolean(
         #
         # Note that even if we lift this 64 bit restriction in the future, p
         # cannot be 0 (1) when forced is True (False).
-        if forced is True:
-            assert p > 2 ** (-64)
-        if forced is False:
-            assert p < (1 - 2 ** (-64))
+        eps = 2 ** (-64) if isinstance(self.provider, HypothesisProvider) else 0
+        assert (forced is not True) or (0 + eps) < p
+        assert (forced is not False) or p < (1 - eps)
 
         kwargs: BooleanKWargs = self._pooled_kwargs("boolean", {"p": p})
 

hypothesis-python/tests/cover/test_type_lookup.py

@@ -17,14 +17,11 @@
 import pytest
 
 from hypothesis import given, infer, settings, strategies as st
-from hypothesis.errors import (
-    HypothesisDeprecationWarning,
-    InvalidArgument,
-    ResolutionFailed,
-)
+from hypothesis.errors import InvalidArgument, ResolutionFailed
 from hypothesis.internal.compat import get_type_hints
 from hypothesis.internal.reflection import get_pretty_function_description
 from hypothesis.strategies._internal import types
+from hypothesis.strategies._internal.lazy import LazyStrategy
 from hypothesis.strategies._internal.types import _global_type_lookup
 from hypothesis.strategies._internal.utils import _strategies
 
@@ -36,42 +33,11 @@
 )
 from tests.common.utils import fails_with, temp_registered
 
-# Build a set of all types output by core strategies
-blocklist = {
-    "builds",
-    "data",
-    "deferred",
-    "from_regex",
-    "from_type",
-    "ip_addresses",
-    "iterables",
-    "just",
-    "nothing",
-    "one_of",
-    "permutations",
-    "random_module",
-    "randoms",
-    "recursive",
-    "runner",
-    "sampled_from",
-    "shared",
-    "timezone_keys",
-    "timezones",
+types_with_core_strat = {
+    type_
+    for type_, strat in _global_type_lookup.items()
+    if isinstance(strat, LazyStrategy) and strat.function in vars(st).values()
 }
-assert set(_strategies).issuperset(blocklist), blocklist.difference(_strategies)
-types_with_core_strat = set()
-for thing in (
-    getattr(st, name)
-    for name in sorted(_strategies)
-    if name in dir(st) and name not in blocklist
-):
-    for n in range(3):
-        try:
-            ex = find_any(thing(*([st.nothing()] * n)))
-            types_with_core_strat.add(type(ex))
-            break
-        except (TypeError, InvalidArgument, HypothesisDeprecationWarning):
-            continue
 
 
 @pytest.mark.skipif(sys.version_info[:2] >= (3, 14), reason="FIXME-py314")
@@ -91,7 +57,44 @@ def inner(ex):
 
 
 def test_lookup_knows_about_all_core_strategies():
-    cannot_lookup = types_with_core_strat - set(types._global_type_lookup)
+    # Build a set of all types output by core strategies
+    blocklist = {
+        "builds",
+        "data",
+        "deferred",
+        "from_regex",
+        "from_type",
+        "ip_addresses",
+        "iterables",
+        "just",
+        "nothing",
+        "one_of",
+        "permutations",
+        "random_module",
+        "randoms",
+        "recursive",
+        "runner",
+        "sampled_from",
+        "shared",
+        "timezone_keys",
+        "timezones",
+    }
+    assert set(_strategies).issuperset(blocklist), blocklist.difference(_strategies)
+    found = set()
+    for thing in (
+        getattr(st, name)
+        for name in sorted(_strategies)
+        if name in dir(st) and name not in blocklist
+    ):
+        for n in range(3):
+            try:
+                ex = find_any(thing(*([st.nothing()] * n)))
+                found.add(type(ex))
+                break
+            except Exception:
+                continue
+
+    cannot_lookup = found - set(types._global_type_lookup)
     assert not cannot_lookup
 
 

hypothesis-python/tests/nocover/test_collective_minimization.py

@@ -16,17 +16,11 @@
 
 from tests.common import standard_types
 from tests.common.debug import minimal
-from tests.common.utils import flaky
 
 
 @pytest.mark.parametrize("spec", standard_types, ids=list(map(repr, standard_types)))
-@flaky(min_passes=1, max_runs=2)
 def test_can_collectively_minimize(spec):
-    """This should generally exercise strategies' strictly_simpler heuristic by
-    putting us in a state where example cloning is required to get to the
-    answer fast enough."""
     n = 10
-
     try:
         xs = minimal(
             lists(spec, min_size=n, max_size=n),

hypothesis-python/tests/nocover/test_conjecture_engine.py

@@ -10,7 +10,6 @@
 
 from hypothesis import given, settings, strategies as st
 from hypothesis.database import InMemoryExampleDatabase
-from hypothesis.internal.compat import int_from_bytes
 from hypothesis.internal.conjecture.data import ConjectureData
 from hypothesis.internal.conjecture.engine import ConjectureRunner
 from hypothesis.internal.conjecture.shrinker import Shrinker, node_program
@@ -80,29 +79,6 @@ def x(data):
     assert len(x) == n
 
 
-def test_regression_1():
-    # This is a really hard to reproduce bug that previously triggered a very
-    # specific exception inside one of the shrink passes. It's unclear how
-    # useful this regression test really is, but nothing else caught the
-    # problem.
-    #
-    # update 2024-01-15: we've since changed generation and are about to
-    # change shrinking, so it's unclear if the failure case this test was aimed
-    # at (1) is still being covered or (2) even exists anymore.
-    # we can probably safely remove this once the shrinker is rebuilt.
-    @run_to_buffer
-    def x(data):
-        data.draw_bytes(2, forced=b"\x01\x02")
-        data.draw_bytes(2, forced=b"\x01\x00")
-        v = data.draw_integer(0, 2**41 - 1)
-        if v >= 512 or v == 254:
-            data.mark_interesting()
-
-    assert list(x)[:-2] == [1, 2, 1, 0, 0, 0, 0, 0, 0]
-
-    assert int_from_bytes(x[-2:]) in (254, 512)
-
-
 @given(st.integers(0, 255), st.integers(0, 255))
 def test_cached_with_masked_byte_agrees_with_results(byte_a, byte_b):
     def f(data):

tooling/src/hypothesistooling/__main__.py

@@ -520,7 +520,7 @@ def standard_tox_task(name, py=ci_version):
 standard_tox_task("py39-pandas11", py="3.9")
 standard_tox_task("py39-pandas12", py="3.9")
 
-for kind in ("cover", "nocover", "niche"):
+for kind in ("cover", "nocover", "niche", "custom"):
     standard_tox_task(f"crosshair-{kind}")
 
 standard_tox_task("py38-oldestnumpy", py="3.8")