Improve patches for np.array

Author: Zac-HD

hypothesis-python/scripts/other-tests.sh

@@ -51,11 +51,6 @@ pip uninstall -y lark
 if [ "$(python -c $'import platform, sys; print(sys.version_info.releaselevel == \'final\' and platform.python_implementation() not in ("PyPy", "GraalVM"))')" = "True" ] ; then
   pip install ".[codemods,cli]"
   $PYTEST tests/codemods/
-  pip install "$(grep -E 'black(==| @)' ../requirements/coverage.txt)"
-  if [ "$(python -c 'import sys; print(sys.version_info[:2] >= (3, 9))')" = "True" ] ; then
-    $PYTEST tests/patching/
-  fi
-  pip uninstall -y libcst
 
   if [ "$(python -c 'import sys; print(sys.version_info[:2] == (3, 8))')" = "True" ] ; then
     # Per NEP-29, this is the last version to support Python 3.8
@@ -64,6 +59,12 @@ if [ "$(python -c $'import platform, sys; print(sys.version_info.releaselevel ==
     pip install "$(grep 'numpy==' ../requirements/coverage.txt)"
   fi
 
+  pip install "$(grep -E 'black(==| @)' ../requirements/coverage.txt)"
+  if [ "$(python -c 'import sys; print(sys.version_info[:2] >= (3, 9))')" = "True" ] ; then
+    $PYTEST tests/patching/
+  fi
+  pip uninstall -y libcst
+
   $PYTEST tests/ghostwriter/
   pip uninstall -y black
 

hypothesis-python/src/hypothesis/extra/_patching.py

@@ -18,11 +18,13 @@
 during fuzzing.
 """
 
+import ast
 import difflib
 import hashlib
 import inspect
 import re
 import sys
+import types
 from ast import literal_eval
 from contextlib import suppress
 from datetime import date, datetime, timedelta, timezone
@@ -31,6 +33,7 @@
 import libcst as cst
 from libcst import matchers as m
 from libcst.codemod import CodemodContext, VisitorBasedCodemodCommand
+from libcst.metadata import ExpressionContext, ExpressionContextProvider
 
 from hypothesis.configuration import storage_directory
 from hypothesis.version import __version__
@@ -148,18 +151,64 @@ def get_patch_for(func, failing_examples, *, strip_via=()):
     except Exception:
         return None
 
+    modules_in_test_scope = sorted(
+        (
+            (k, v)
+            for (k, v) in module.__dict__.items()
+            if isinstance(v, types.ModuleType)
+        ),
+        key=lambda kv: len(kv[1].__name__),
+    )
+
     # The printed examples might include object reprs which are invalid syntax,
     # so we parse here and skip over those.  If _none_ are valid, there's no patch.
     call_nodes = []
     for ex, via in set(failing_examples):
         with suppress(Exception):
-            node = cst.parse_expression(ex)
-            assert isinstance(node, cst.Call), node
+            node = cst.parse_module(ex)
+            the_call = node.body[0].body[0].value
+            assert isinstance(the_call, cst.Call), the_call
             # Check for st.data(), which doesn't support explicit examples
             data = m.Arg(m.Call(m.Name("data"), args=[m.Arg(m.Ellipsis())]))
-            if m.matches(node, m.Call(args=[m.ZeroOrMore(), data, m.ZeroOrMore()])):
+            if m.matches(the_call, m.Call(args=[m.ZeroOrMore(), data, m.ZeroOrMore()])):
                 return None
+
+            # Many reprs use the unqualified name of the type, e.g. np.array()
+            # -> array([...]), so here we find undefined names and look them up
+            # on each module which was in the test's global scope.
+            names = {}
+            for anode in ast.walk(ast.parse(ex, "eval")):
+                if (
+                    isinstance(anode, ast.Name)
+                    and isinstance(anode.ctx, ast.Load)
+                    and anode.id not in names
+                    and anode.id not in module.__dict__
+                ):
+                    for k, v in modules_in_test_scope:
+                        if anode.id in v.__dict__:
+                            names[anode.id] = cst.parse_expression(f"{k}.{anode.id}")
+                            break
+
+            # LibCST doesn't track Load()/Store() state of names by default, so we have
+            # to do a bit of a dance here, *and* explicitly handle keyword arguments
+            # which are treated as Load() context - but even if that's fixed later
+            # we'll still want to support older versions.
+            with suppress(Exception):
+                wrapper = cst.metadata.MetadataWrapper(node)
+                kwarg_names = {
+                    a.keyword for a in m.findall(wrapper, m.Arg(keyword=m.Name()))
+                }
+                node = m.replace(
+                    wrapper,
+                    m.Name(value=m.MatchIfTrue(names.__contains__))
+                    & m.MatchMetadata(ExpressionContextProvider, ExpressionContext.LOAD)
+                    & m.MatchIfTrue(lambda n, k=kwarg_names: n not in k),
+                    replacement=lambda node, _, ns=names: ns[node.value],
+                )
+            node = node.body[0].body[0].value
+            assert isinstance(node, cst.Call), node
             call_nodes.append((node, via))
+
     if not call_nodes:
         return None
 

hypothesis-python/tests/patching/callables.py

@@ -12,7 +12,10 @@
 
 from pathlib import Path
 
+import numpy as np
+
 from hypothesis import example, given, strategies as st
+from hypothesis.extra import numpy as npst
 
 WHERE = Path(__file__).relative_to(Path.cwd())
 
@@ -36,4 +39,9 @@ def covered(x):
     """A test function with a removable explicit example."""
 
 
+@given(npst.arrays(np.int8, 1))
+def undef_name(array):
+    assert sum(array) < 100
+
+
 # TODO: test function for insertion-order logic, once I get that set up.

hypothesis-python/tests/patching/test_patching.py

@@ -23,7 +23,7 @@
 )
 from hypothesis.internal.compat import WINDOWS
 
-from .callables import WHERE, Cases, covered, fn
+from .callables import WHERE, Cases, covered, fn, undef_name
 from .toplevel import WHERE_TOP, fn_top
 
 SIMPLE = (
@@ -52,6 +52,11 @@
     + "\n"
     + indent('@example(x=0).via("covering example")', prefix="+"),
 )
+UNDEF_NAME = (
+    undef_name,
+    ("undef_name(\n    array=array([100], dtype=int8),\n)", FAIL_MSG),
+    '+@example(array=np.array([100], dtype=np.int8)).via("discovered failure")',
+)
 
 
 def strip_trailing_whitespace(s):
@@ -76,7 +81,7 @@ def test_adds_simple_patch(tst, example, expected):
 SIMPLE_PATCH_BODY = f'''\
 --- ./{WHERE}
 +++ ./{WHERE}
-@@ -18,6 +18,7 @@
+@@ -21,6 +21,7 @@
 
 
  @given(st.integers())
@@ -88,7 +93,7 @@ def fn(x):
 CASES_PATCH_BODY = f'''\
 --- ./{WHERE}
 +++ ./{WHERE}
-@@ -25,6 +25,9 @@
+@@ -28,6 +28,9 @@
  class Cases:
      @example(n=0, label="whatever")
      @given(st.integers(), st.text())
@@ -111,7 +116,7 @@ def fn_top(x):
 COVERING_PATCH_BODY = f'''\
 --- ./{WHERE}
 +++ ./{WHERE}
-@@ -31,7 +31,7 @@
+@@ -34,7 +34,7 @@
 
  @given(st.integers())
  @example(x=2).via("not a literal when repeated " * 2)
@@ -121,6 +126,19 @@ def covered(x):
 
 '''
 
+UNDEF_NAME_PATCH_BODY = f"""\
+--- ./{WHERE}
++++ ./{WHERE}
+@@ -40,6 +40,7 @@
+
+
+ @given(npst.arrays(np.int8, 1))
+{{0}}
+ def undef_name(array):
+     assert sum(array) < 100
+
+"""
+
 
 @pytest.mark.parametrize(
     "tst, example, expected, body, remove",
@@ -131,6 +149,7 @@ def covered(x):
         pytest.param(
             *COVERING, COVERING_PATCH_BODY, ("covering example",), id="covering"
         ),
+        pytest.param(*UNDEF_NAME, UNDEF_NAME_PATCH_BODY, (), id="undef_name"),
     ],
 )
 def test_make_full_patch(tst, example, expected, body, remove):