`frozendict`'s hash is not initialised when created through pipe operator

[KowalskiThomas]

Bug report

Bug description:

As far as I know, the expectation is a == b => hash(a) == hash(b), but currently, the following code fails:

a = frozendict({"a": 1})
b = frozendict({"b": 2})

c = frozendict({"a": 1, "b": 2})
c_union = a | b

assert c == c_union
assert hash(c) == hash(c_union)

with

  File "t.py", line 8, in <module>
    assert hash(c) == hash(c_union)
           ^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError

I have a fix ready in #149675.

CPython versions tested on:

3.15

Operating systems tested on:

macOS

Linked PRs

• gh-149676: fix: properly initialise frozendict hash value #149675
• [3.15] gh-149676: Fix hash(frozendict | frozendict) (GH-149675) #149717

[sergey-miryanov]

cc @vstinner @corona10

[vstinner]

Code:

a = frozendict({"a": 1})
b = frozendict({"b": 2})
h = hash(a | b)
print(h)

I confirm that Valgrind detects usage of uninitialized memory: PYTHONMALLOC=malloc valgrind ./python x.py

==100111== Conditional jump or move depends on uninitialised value(s)
==100111==    at 0x4DCA23: frozendict_hash (dictobject.c:8214)
==100111==    by 0x4F093D: PyObject_Hash (object.c:1162)
==100111==    by 0x5C865E: builtin_hash (bltinmodule.c:1857)
==100111==    by 0x4EB4AB: cfunction_vectorcall_O (methodobject.c:536)
==100111==    by 0x4984EE: _PyObject_VectorcallTstate (pycore_call.h:144)
==100111==    by 0x4985B6: PyObject_Vectorcall (call.c:327)
==100111==    by 0x5CF6FD: _Py_VectorCallInstrumentation_StackRefSteal (ceval.c:775)
==100111==    by 0x5D6783: _PyEval_EvalFrameDefault (generated_cases.c.h:1846)
==100111==    by 0x5EEADA: _PyEval_EvalFrame (pycore_ceval.h:122)
==100111==    by 0x5EEC8C: _PyEval_Vector (ceval.c:2156)
==100111==    by 0x5EED5F: PyEval_EvalCode (ceval.c:686)
==100111==    by 0x6683DB: run_eval_code_obj (pythonrun.c:1369)
==100111== 
==100111== Conditional jump or move depends on uninitialised value(s)
==100111==    at 0x5C8663: builtin_hash (bltinmodule.c:1858)
==100111==    by 0x4EB4AB: cfunction_vectorcall_O (methodobject.c:536)
==100111==    by 0x4984EE: _PyObject_VectorcallTstate (pycore_call.h:144)
==100111==    by 0x4985B6: PyObject_Vectorcall (call.c:327)
==100111==    by 0x5CF6FD: _Py_VectorCallInstrumentation_StackRefSteal (ceval.c:775)
==100111==    by 0x5D6783: _PyEval_EvalFrameDefault (generated_cases.c.h:1846)
==100111==    by 0x5EEADA: _PyEval_EvalFrame (pycore_ceval.h:122)
==100111==    by 0x5EEC8C: _PyEval_Vector (ceval.c:2156)
==100111==    by 0x5EED5F: PyEval_EvalCode (ceval.c:686)
==100111==    by 0x6683DB: run_eval_code_obj (pythonrun.c:1369)
==100111==    by 0x6685A2: run_mod (pythonrun.c:1472)
==100111==    by 0x668D8B: pyrun_file (pythonrun.c:1296)
==100111== 
==100111== Conditional jump or move depends on uninitialised value(s)
==100111==    at 0x4D36C4: PyLong_FromSsize_t (longobject.c:1576)
==100111==    by 0x5C866C: builtin_hash (bltinmodule.c:1860)
==100111==    by 0x4EB4AB: cfunction_vectorcall_O (methodobject.c:536)
==100111==    by 0x4984EE: _PyObject_VectorcallTstate (pycore_call.h:144)
==100111==    by 0x4985B6: PyObject_Vectorcall (call.c:327)
==100111==    by 0x5CF6FD: _Py_VectorCallInstrumentation_StackRefSteal (ceval.c:775)
==100111==    by 0x5D6783: _PyEval_EvalFrameDefault (generated_cases.c.h:1846)
==100111==    by 0x5EEADA: _PyEval_EvalFrame (pycore_ceval.h:122)
==100111==    by 0x5EEC8C: _PyEval_Vector (ceval.c:2156)
==100111==    by 0x5EED5F: PyEval_EvalCode (ceval.c:686)
==100111==    by 0x6683DB: run_eval_code_obj (pythonrun.c:1369)
==100111==    by 0x6685A2: run_mod (pythonrun.c:1472)
==100111== 
==100111== Conditional jump or move depends on uninitialised value(s)
==100111==    at 0x4C9E0B: get_small_int (longobject.c:63)
==100111==    by 0x4D36F0: PyLong_FromSsize_t (longobject.c:1576)
==100111==    by 0x5C866C: builtin_hash (bltinmodule.c:1860)
==100111==    by 0x4EB4AB: cfunction_vectorcall_O (methodobject.c:536)
==100111==    by 0x4984EE: _PyObject_VectorcallTstate (pycore_call.h:144)
==100111==    by 0x4985B6: PyObject_Vectorcall (call.c:327)
==100111==    by 0x5CF6FD: _Py_VectorCallInstrumentation_StackRefSteal (ceval.c:775)
==100111==    by 0x5D6783: _PyEval_EvalFrameDefault (generated_cases.c.h:1846)
==100111==    by 0x5EEADA: _PyEval_EvalFrame (pycore_ceval.h:122)
==100111==    by 0x5EEC8C: _PyEval_Vector (ceval.c:2156)
==100111==    by 0x5EED5F: PyEval_EvalCode (ceval.c:686)
==100111==    by 0x6683DB: run_eval_code_obj (pythonrun.c:1369)

[vstinner]

Fixed. Thanks for the fix and the bug report.