Merge pull request #3444 from Zac-HD/annotate-internals

Skip uninformative locations in explain mode, misc internal cleanups

hypothesis-python/RELEASE.rst

@@ -0,0 +1,5 @@
+RELEASE_TYPE: patch
+
+This patch fixes some type annotations for Python 3.9 and earlier (:issue:`3397`),
+and teaches :ref:`explain mode <phases>` about certain locations it should not
+bother reporting (:issue:`3439`).

hypothesis-python/src/hypothesis/core.py

@@ -881,8 +881,7 @@ def run_engine(self):
                 errors_to_report.append((fragments, err))
             except BaseException as e:
                 # If we have anything for explain-mode, this is the time to report.
-                for line in explanations[falsifying_example.interesting_origin]:
-                    fragments.append(line)
+                fragments.extend(explanations[falsifying_example.interesting_origin])
                 errors_to_report.append(
                     (fragments, e.with_traceback(get_trimmed_traceback()))
                 )

hypothesis-python/src/hypothesis/internal/compat.py

@@ -22,6 +22,19 @@
         BaseExceptionGroup as BaseExceptionGroup,
         ExceptionGroup as ExceptionGroup,
     )
+if typing.TYPE_CHECKING:  # pragma: no cover
+    from typing_extensions import Concatenate as Concatenate, ParamSpec as ParamSpec
+else:
+    try:
+        from typing import Concatenate as Concatenate, ParamSpec as ParamSpec
+    except ImportError:
+        try:
+            from typing_extensions import (
+                Concatenate as Concatenate,
+                ParamSpec as ParamSpec,
+            )
+        except ImportError:
+            Concatenate, ParamSpec = None, None
 
 PYPY = platform.python_implementation() == "PyPy"
 WINDOWS = platform.system() == "Windows"

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

@@ -278,8 +278,8 @@ def __init__(self, engine, initial, predicate, allow_transition):
 
         # We keep track of the current best example on the shrink_target
         # attribute.
-        self.shrink_target = None
-        self.update_shrink_target(initial)
+        self.shrink_target = initial
+        self.clear_change_tracking()
         self.shrinks = 0
 
         # We terminate shrinks that seem to have reached their logical
@@ -447,23 +447,15 @@ def s(n):
                     return "s" if n != 1 else ""
 
                 total_deleted = self.initial_size - len(self.shrink_target.buffer)
-
-                self.debug("---------------------")
-                self.debug("Shrink pass profiling")
-                self.debug("---------------------")
-                self.debug("")
                 calls = self.engine.call_count - self.initial_calls
+
                 self.debug(
-                    "Shrinking made a total of %d call%s "
-                    "of which %d shrank. This deleted %d byte%s out of %d."
-                    % (
-                        calls,
-                        s(calls),
-                        self.shrinks,
-                        total_deleted,
-                        s(total_deleted),
-                        self.initial_size,
-                    )
+                    "---------------------\n"
+                    "Shrink pass profiling\n"
+                    "---------------------\n\n"
+                    f"Shrinking made a total of {calls} call{s(calls)} of which "
+                    f"{self.shrinks} shrank. This deleted {total_deleted} bytes out "
+                    f"of {self.initial_size}."
                 )
                 for useful in [True, False]:
                     self.debug("")
@@ -828,22 +820,17 @@ def __changed_blocks(self):
 
     def update_shrink_target(self, new_target):
         assert isinstance(new_target, ConjectureResult)
-        if self.shrink_target is not None:
-            self.shrinks += 1
-            # If we are just taking a long time to shrink we don't want to
-            # trigger this heuristic, so whenever we shrink successfully
-            # we give ourselves a bit of breathing room to make sure we
-            # would find a shrink that took that long to find the next time.
-            # The case where we're taking a long time but making steady
-            # progress is handled by `finish_shrinking_deadline` in engine.py
-            self.max_stall = max(
-                self.max_stall, (self.calls - self.calls_at_last_shrink) * 2
-            )
-            self.calls_at_last_shrink = self.calls
-        else:
-            self.__all_changed_blocks = set()
-            self.__last_checked_changed_at = new_target
-
+        self.shrinks += 1
+        # If we are just taking a long time to shrink we don't want to
+        # trigger this heuristic, so whenever we shrink successfully
+        # we give ourselves a bit of breathing room to make sure we
+        # would find a shrink that took that long to find the next time.
+        # The case where we're taking a long time but making steady
+        # progress is handled by `finish_shrinking_deadline` in engine.py
+        self.max_stall = max(
+            self.max_stall, (self.calls - self.calls_at_last_shrink) * 2
+        )
+        self.calls_at_last_shrink = self.calls
         self.shrink_target = new_target
         self.__derived_values = {}
 

hypothesis-python/src/hypothesis/internal/reflection.py

@@ -64,10 +64,6 @@ def function_digest(function):
         hasher.update(function.__name__.encode())
     except AttributeError:
         pass
-    try:
-        hasher.update(function.__module__.__name__.encode())
-    except AttributeError:
-        pass
     try:
         # We prefer to use the modern signature API, but left this for compatibility.
         # While we don't promise stability of the database, there's no advantage to

hypothesis-python/src/hypothesis/internal/scrutineer.py

@@ -12,6 +12,7 @@
 from collections import defaultdict
 from functools import lru_cache, reduce
 from itertools import groupby
+from os import sep
 from pathlib import Path
 
 from hypothesis._settings import Phase, Verbosity
@@ -45,6 +46,20 @@ def trace(self, frame, event, arg):
                 self._previous_location = current_location
 
 
+UNHELPFUL_LOCATIONS = (
+    # There's a branch which is only taken when an exception is active while exiting
+    # a contextmanager; this is probably after the fault has been triggered.
+    # Similar reasoning applies to a few other standard-library modules: even
+    # if the fault was later, these still aren't useful locations to report!
+    f"{sep}contextlib.py",
+    f"{sep}inspect.py",
+    f"{sep}re.py",
+    f"{sep}re{sep}__init__.py",  # refactored in Python 3.11
+    # Quite rarely, the first AFNP line is in Pytest's assertion-rewriting module.
+    f"{sep}_pytest{sep}assertion{sep}rewrite.py",
+)
+
+
 def get_explaining_locations(traces):
     # Traces is a dict[interesting_origin | None, set[frozenset[tuple[str, int]]]]
     # Each trace in the set might later become a Counter instead of frozenset.
@@ -84,21 +99,25 @@ def get_explaining_locations(traces):
             else:
                 queue.update(cf_graphs[origin][src] - seen)
 
-    return explanations
+    # The last step is to filter out explanations that we know would be uninformative.
+    # When this is the first AFNP location, we conclude that Scrutineer missed the
+    # real divergence (earlier in the trace) and drop that unhelpful explanation.
+    return {
+        origin: {loc for loc in afnp_locs if not loc[0].endswith(UNHELPFUL_LOCATIONS)}
+        for origin, afnp_locs in explanations.items()
+    }
 
 
 LIB_DIR = str(Path(sys.executable).parent / "lib")
 EXPLANATION_STUB = (
     "Explanation:",
     "    These lines were always and only run by failing examples:",
 )
-HAD_TRACE = "    We didn't try to explain this, because sys.gettrace()="
 
 
 def make_report(explanations, cap_lines_at=5):
     report = defaultdict(list)
     for origin, locations in explanations.items():
-        assert locations  # or else we wouldn't have stored the key, above.
         report_lines = [
             "        {}:{}".format(k, ", ".join(map(str, sorted(l for _, l in v))))
             for k, v in groupby(locations, lambda kv: kv[0])
@@ -107,15 +126,14 @@ def make_report(explanations, cap_lines_at=5):
         if len(report_lines) > cap_lines_at + 1:
             msg = "        (and {} more with settings.verbosity >= verbose)"
             report_lines[cap_lines_at:] = [msg.format(len(report_lines[cap_lines_at:]))]
-        report[origin] = list(EXPLANATION_STUB) + report_lines
+        if report_lines:  # We might have filtered out every location as uninformative.
+            report[origin] = list(EXPLANATION_STUB) + report_lines
     return report
 
 
 def explanatory_lines(traces, settings):
     if Phase.explain in settings.phases and sys.gettrace() and not traces:
-        return defaultdict(
-            lambda: [EXPLANATION_STUB[0], HAD_TRACE + repr(sys.gettrace())]
-        )
+        return defaultdict(list)
     # Return human-readable report lines summarising the traces
     explanations = get_explaining_locations(traces)
     max_lines = 5 if settings.verbosity <= Verbosity.normal else 100

hypothesis-python/src/hypothesis/strategies/_internal/core.py

@@ -49,7 +49,14 @@
 from hypothesis.errors import InvalidArgument, ResolutionFailed
 from hypothesis.internal.cathetus import cathetus
 from hypothesis.internal.charmap import as_general_categories
-from hypothesis.internal.compat import ceil, floor, get_type_hints, is_typed_named_tuple
+from hypothesis.internal.compat import (
+    Concatenate,
+    ParamSpec,
+    ceil,
+    floor,
+    get_type_hints,
+    is_typed_named_tuple,
+)
 from hypothesis.internal.conjecture.utils import (
     calc_label_from_cls,
     check_sample,
@@ -122,14 +129,6 @@
 except ImportError:  # < py3.8
     Protocol = object  # type: ignore[assignment]
 
-try:
-    from typing import Concatenate, ParamSpec
-except ImportError:
-    try:
-        from typing_extensions import Concatenate, ParamSpec
-    except ImportError:
-        ParamSpec = None  # type: ignore
-
 
 @cacheable
 @defines_strategy()

hypothesis-python/tests/common/setup.py

@@ -12,7 +12,7 @@
 from tempfile import mkdtemp
 from warnings import filterwarnings
 
-from hypothesis import Verbosity, settings
+from hypothesis import Phase, Verbosity, settings
 from hypothesis._settings import not_set
 from hypothesis.configuration import set_hypothesis_home_dir
 from hypothesis.errors import NonInteractiveExampleWarning
@@ -25,10 +25,7 @@ def run():
     filterwarnings("ignore", category=ImportWarning)
     filterwarnings("ignore", category=FutureWarning, module="pandas._version")
 
-    # Fixed in recent versions but allowed by pytest=3.0.0; see #1630
-    filterwarnings("ignore", category=DeprecationWarning, module="pluggy")
-
-    # See https://github.com/numpy/numpy/pull/432
+    # See https://github.com/numpy/numpy/pull/432; still a thing as of 2022.
     filterwarnings("ignore", message="numpy.dtype size changed")
     filterwarnings("ignore", message="numpy.ufunc size changed")
 
@@ -42,14 +39,6 @@ def run():
         category=UserWarning,
     )
 
-    # Imported by Pandas in version 1.9, but fixed in later versions.
-    filterwarnings(
-        "ignore", message="Importing from numpy.testing.decorators is deprecated"
-    )
-    filterwarnings(
-        "ignore", message="Importing from numpy.testing.nosetester is deprecated"
-    )
-
     # User-facing warning which does not apply to our own tests
     filterwarnings("ignore", category=NonInteractiveExampleWarning)
 
@@ -77,7 +66,11 @@ def run():
             )
 
     settings.register_profile(
-        "default", settings(max_examples=20 if IN_COVERAGE_TESTS else not_set)
+        "default",
+        settings(
+            max_examples=20 if IN_COVERAGE_TESTS else not_set,
+            phases=list(Phase),  # Dogfooding the explain phase
+        ),
     )
 
     settings.register_profile("speedy", settings(max_examples=5))

hypothesis-python/tests/cover/test_phases.py

@@ -11,6 +11,7 @@
 import pytest
 
 from hypothesis import Phase, example, given, settings, strategies as st
+from hypothesis._settings import all_settings
 from hypothesis.database import ExampleDatabase, InMemoryExampleDatabase
 from hypothesis.errors import InvalidArgument
 
@@ -47,7 +48,7 @@ def test_sorts_and_dedupes_phases(arg, expected):
 
 
 def test_phases_default_to_all_except_explain():
-    assert settings().phases + (Phase.explain,) == tuple(Phase)
+    assert all_settings["phases"].default + (Phase.explain,) == tuple(Phase)
 
 
 def test_does_not_reuse_saved_examples_if_reuse_not_in_phases():

hypothesis-python/tests/cover/test_settings.py

@@ -464,7 +464,7 @@ def __repr__(self):
 
 
 def test_show_changed():
-    s = settings(max_examples=999, database=None)
+    s = settings(max_examples=999, database=None, phases=tuple(Phase)[:-1])
     assert s.show_changed() == "database=None, max_examples=999"