Merge pull request #3799 from HypothesisWorks/create-pull-request/patch

Update pinned dependencies

Author: Zac-HD

hypothesis-python/.coveragerc

@@ -28,3 +28,4 @@ exclude_lines =
     if TYPE_CHECKING:
     if sys\.version_info
     if "[\w\.]+" in sys\.modules:
+    if .+ := sys\.modules\.get\("[\w\.]+"\)

hypothesis-python/RELEASE.rst

@@ -0,0 +1,6 @@
+RELEASE_TYPE: minor
+
+This release adds an optional ``payload`` argument to :func:`hypothesis.event`,
+so that you can clearly express the difference between the label and the value
+of an observation.  :ref:`statistics` will still summarize it as a string, but
+future observability options can preserve the distinction.

hypothesis-python/setup.py

@@ -117,6 +117,7 @@ def local_file(name):
         "Programming Language :: Python :: 3.9",
         "Programming Language :: Python :: 3.10",
         "Programming Language :: Python :: 3.11",
+        "Programming Language :: Python :: 3.12",
         "Programming Language :: Python :: Implementation :: CPython",
         "Programming Language :: Python :: Implementation :: PyPy",
         "Topic :: Education :: Testing",

hypothesis-python/src/hypothesis/control.py

@@ -11,6 +11,7 @@
 import math
 from collections import defaultdict
 from typing import NoReturn, Union
+from weakref import WeakKeyDictionary
 
 from hypothesis import Verbosity, settings
 from hypothesis._settings import note_deprecation
@@ -168,18 +169,38 @@ def note(value: str) -> None:
         report(value)
 
 
-def event(value: str) -> None:
-    """Record an event that occurred this test. Statistics on number of test
+def event(value: str, payload: Union[str, int, float] = "") -> None:
+    """Record an event that occurred during this test. Statistics on the number of test
     runs with each event will be reported at the end if you run Hypothesis in
     statistics reporting mode.
 
-    Events should be strings or convertible to them.
+    Event values should be strings or convertible to them.  If an optional
+    payload is given, it will be included in the string for :ref:`statistics`.
     """
     context = _current_build_context.value
     if context is None:
         raise InvalidArgument("Cannot make record events outside of a test")
 
-    context.data.note_event(value)
+    payload = _event_to_string(payload, (str, int, float))
+    context.data.events[_event_to_string(value)] = payload
+
+
+_events_to_strings: WeakKeyDictionary = WeakKeyDictionary()
+
+
+def _event_to_string(event, allowed_types=str):
+    if isinstance(event, allowed_types):
+        return event
+    try:
+        return _events_to_strings[event]
+    except (KeyError, TypeError):
+        pass
+    result = str(event)
+    try:
+        _events_to_strings[event] = result
+    except TypeError:
+        pass
+    return result
 
 
 def target(observation: Union[int, float], *, label: str = "") -> Union[int, float]:

hypothesis-python/src/hypothesis/core.py

@@ -80,10 +80,10 @@
 from hypothesis.internal.conjecture.shrinker import sort_key
 from hypothesis.internal.entropy import deterministic_PRNG
 from hypothesis.internal.escalation import (
+    InterestingOrigin,
     current_pytest_item,
     escalate_hypothesis_internal_error,
     format_exception,
-    get_interesting_origin,
     get_trimmed_traceback,
 )
 from hypothesis.internal.healthcheck import fail_health_check
@@ -970,7 +970,7 @@ def _execute_once_for_engine(self, data):
 
                 self.failed_normally = True
 
-                interesting_origin = get_interesting_origin(e)
+                interesting_origin = InterestingOrigin.from_exception(e)
                 if trace:  # pragma: no cover
                     # Trace collection is explicitly disabled under coverage.
                     self.explain_traces[interesting_origin].add(trace)
@@ -1037,7 +1037,9 @@ def run_engine(self):
             info = falsifying_example.extra_information
             fragments = []
 
-            ran_example = ConjectureData.for_buffer(falsifying_example.buffer)
+            ran_example = runner.new_conjecture_data_for_buffer(
+                falsifying_example.buffer
+            )
             ran_example.slice_comments = falsifying_example.slice_comments
             assert info.__expected_exception is not None
             try:

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

@@ -20,7 +20,6 @@
     Callable,
     Dict,
     FrozenSet,
-    Hashable,
     Iterable,
     Iterator,
     List,
@@ -1367,7 +1366,7 @@ def __init__(
         self.testcounter = global_test_counter
         global_test_counter += 1
         self.start_time = time.perf_counter()
-        self.events: "Union[Set[Hashable], FrozenSet[Hashable]]" = set()
+        self.events: Dict[str, Union[str, int, float]] = {}
         self.forced_indices: "Set[int]" = set()
         self.interesting_origin: Optional[InterestingOrigin] = None
         self.draw_times: "List[float]" = []
@@ -1615,10 +1614,6 @@ def stop_example(self, *, discard: bool = False) -> None:
 
             self.observer.kill_branch()
 
-    def note_event(self, event: Hashable) -> None:
-        assert isinstance(self.events, set)
-        self.events.add(event)
-
     @property
     def examples(self) -> Examples:
         assert self.frozen
@@ -1643,7 +1638,6 @@ def freeze(self) -> None:
         self.frozen = True
 
         self.buffer = bytes(self.buffer)
-        self.events = frozenset(self.events)
         self.observer.conclude_test(self.status, self.interesting_origin)
 
     def draw_bits(self, n: int, *, forced: Optional[int] = None) -> int:
@@ -1729,7 +1723,7 @@ def mark_interesting(
 
     def mark_invalid(self, why: Optional[str] = None) -> NoReturn:
         if why is not None:
-            self.note_event(why)
+            self.events["invalid because"] = why
         self.conclude_test(Status.INVALID)
 
     def mark_overrun(self) -> NoReturn:

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

@@ -15,7 +15,6 @@
 from datetime import timedelta
 from enum import Enum
 from random import Random, getrandbits
-from weakref import WeakKeyDictionary
 
 import attr
 
@@ -101,8 +100,6 @@ def __init__(
         self.statistics = {}
         self.stats_per_test_case = []
 
-        self.events_to_strings = WeakKeyDictionary()
-
         self.interesting_examples = {}
         # We use call_count because there may be few possible valid_examples.
         self.first_bug_found_at = None
@@ -209,7 +206,9 @@ def test_function(self, data):
                     "status": data.status.name.lower(),
                     "runtime": data.finish_time - data.start_time,
                     "drawtime": math.fsum(data.draw_times),
-                    "events": sorted({self.event_to_string(e) for e in data.events}),
+                    "events": sorted(
+                        k if v == "" else f"{k}: {v}" for k, v in data.events.items()
+                    ),
                 }
                 self.stats_per_test_case.append(call_stats)
                 self.__data_cache[data.buffer] = data.as_result()
@@ -1055,20 +1054,6 @@ def kill_branch(self):
             self.__data_cache[buffer] = result
         return result
 
-    def event_to_string(self, event):
-        if isinstance(event, str):
-            return event
-        try:
-            return self.events_to_strings[event]
-        except (KeyError, TypeError):
-            pass
-        result = str(event)
-        try:
-            self.events_to_strings[event] = result
-        except TypeError:
-            pass
-        return result
-
     def passing_buffers(self, prefix=b""):
         """Return a collection of bytestrings which cause the test to pass.
 

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

@@ -11,10 +11,11 @@
 import contextlib
 import os
 import sys
+import textwrap
 import traceback
 from inspect import getframeinfo
 from pathlib import Path
-from typing import Dict
+from typing import Dict, NamedTuple, Optional, Type
 
 import hypothesis
 from hypothesis.errors import (
@@ -105,32 +106,46 @@ def get_trimmed_traceback(exception=None):
     return tb
 
 
-def get_interesting_origin(exception):
+class InterestingOrigin(NamedTuple):
     # The `interesting_origin` is how Hypothesis distinguishes between multiple
     # failures, for reporting and also to replay from the example database (even
     # if report_multiple_bugs=False).  We traditionally use the exception type and
     # location, but have extracted this logic in order to see through `except ...:`
     # blocks and understand the __cause__ (`raise x from y`) or __context__ that
     # first raised an exception as well as PEP-654 exception groups.
-    tb = get_trimmed_traceback(exception)
-    if tb is None:
+    exc_type: Type[BaseException]
+    filename: Optional[str]
+    lineno: Optional[int]
+    context: "InterestingOrigin | tuple[()]"
+    group_elems: "tuple[InterestingOrigin, ...]"
+
+    def __str__(self) -> str:
+        ctx = ""
+        if self.context:
+            ctx = textwrap.indent(f"\ncontext: {self.context}", prefix="    ")
+        group = ""
+        if self.group_elems:
+            chunks = "\n  ".join(str(x) for x in self.group_elems)
+            group = textwrap.indent(f"\nchild exceptions:\n  {chunks}", prefix="    ")
+        return f"{self.exc_type.__name__} at {self.filename}:{self.lineno}{ctx}{group}"
+
+    @classmethod
+    def from_exception(cls, exception: BaseException, /) -> "InterestingOrigin":
         filename, lineno = None, None
-    else:
-        filename, lineno, *_ = traceback.extract_tb(tb)[-1]
-    return (
-        type(exception),
-        filename,
-        lineno,
-        # Note that if __cause__ is set it is always equal to __context__, explicitly
-        # to support introspection when debugging, so we can use that unconditionally.
-        get_interesting_origin(exception.__context__) if exception.__context__ else (),
-        # We distinguish exception groups by the inner exceptions, as for __context__
-        tuple(
-            map(get_interesting_origin, exception.exceptions)
+        if tb := get_trimmed_traceback(exception):
+            filename, lineno, *_ = traceback.extract_tb(tb)[-1]
+        return cls(
+            type(exception),
+            filename,
+            lineno,
+            # Note that if __cause__ is set it is always equal to __context__, explicitly
+            # to support introspection when debugging, so we can use that unconditionally.
+            cls.from_exception(exception.__context__) if exception.__context__ else (),
+            # We distinguish exception groups by the inner exceptions, as for __context__
+            tuple(map(cls.from_exception, exception.exceptions))
             if isinstance(exception, BaseExceptionGroup)
-            else []
-        ),
-    )
+            else (),
+        )
 
 
 current_pytest_item = DynamicVariable(None)

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

@@ -163,8 +163,10 @@ def draw_naive_datetime_and_combine(self, data, tz):
         try:
             return replace_tzinfo(dt.datetime(**result), timezone=tz)
         except (ValueError, OverflowError):
-            msg = "Failed to draw a datetime between %r and %r with timezone from %r."
-            data.mark_invalid(msg % (self.min_value, self.max_value, self.tz_strat))
+            data.mark_invalid(
+                f"Failed to draw a datetime between {self.min_value!r} and "
+                f"{self.max_value!r} with timezone from {self.tz_strat!r}."
+            )
 
 
 @defines_strategy(force_reusable_values=True)

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

@@ -12,7 +12,6 @@
 from contextlib import contextmanager
 
 from hypothesis.errors import InvalidArgument
-from hypothesis.internal.lazyformat import lazyformat
 from hypothesis.internal.reflection import get_pretty_function_description
 from hypothesis.internal.validation import check_type
 from hypothesis.strategies._internal.strategies import (
@@ -112,13 +111,7 @@ def do_draw(self, data):
                 with self.limited_base.capped(self.max_leaves):
                     return data.draw(self.strategy)
             except LimitReached:
-                # Workaround for possible coverage bug - this branch is definitely
-                # covered but for some reason is showing up as not covered.
-                if count == 0:  # pragma: no branch
-                    data.note_event(
-                        lazyformat(
-                            "Draw for %r exceeded max_leaves and had to be retried",
-                            self,
-                        )
-                    )
+                if count == 0:
+                    msg = f"Draw for {self!r} exceeded max_leaves and had to be retried"
+                    data.events[msg] = ""
                 count += 1

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

@@ -43,7 +43,6 @@
     combine_labels,
 )
 from hypothesis.internal.coverage import check_function
-from hypothesis.internal.lazyformat import lazyformat
 from hypothesis.internal.reflection import (
     get_pretty_function_description,
     is_identity_function,
@@ -550,7 +549,7 @@ def do_filtered_draw(self, data):
                 if element is not filter_not_satisfied:
                     return element
                 if not known_bad_indices:
-                    FilteredStrategy.note_retried(self, data)
+                    data.events[f"Retried draw from {self!r} to satisfy filter"] = ""
                 known_bad_indices.add(i)
 
         # If we've tried all the possible elements, give up now.
@@ -940,9 +939,6 @@ def do_draw(self, data: ConjectureData) -> Ex:
         data.mark_invalid(f"Aborted test because unable to satisfy {self!r}")
         raise NotImplementedError("Unreachable, for Mypy")
 
-    def note_retried(self, data):
-        data.note_event(lazyformat("Retried draw from %r to satisfy filter", self))
-
     def do_filtered_draw(self, data):
         for i in range(3):
             start_index = data.index
@@ -954,7 +950,7 @@ def do_filtered_draw(self, data):
             else:
                 data.stop_example(discard=True)
                 if i == 0:
-                    self.note_retried(data)
+                    data.events[f"Retried draw from {self!r} to satisfy filter"] = ""
                 # This is to guard against the case where we consume no data.
                 # As long as we consume data, we'll eventually pass or raise.
                 # But if we don't this could be an infinite loop.