Disabling valgrind for now.

[rwgk]

EDIT 2020-07-10: This PR was rolled back with PR #3090.

---

Some observations:

• gcc + Python are identical between "last working" and "first broken". See below.
• The valgrind version is pinned (3.16.1).
• The test triggering a valgrind error was new in tested PR smart_holder trampoline shared_from_this support #3023 (test_class_sh_trampoline_shared_from_this.py). See below for the valgrind error.
• Google-internal testing with (ASAN, MSAN, UBSAN) x (Python 3.6, 3.7) x (fastbuild, opt, dbg) did not show any errors for test_class_sh_trampoline_shared_from_this.py (to self for future reference: log_test18_2021-07-01+0746).

The root cause for the error could be:

1. A change in pybind11 under PR smart_holder trampoline shared_from_this support #3023.
2. An existing bug in pybind11 triggered only by PR smart_holder trampoline shared_from_this support #3023.
3. An existing bug in Python 3.9.5 triggered only by PR smart_holder trampoline shared_from_this support #3023.
4. An existing bug in gcc 9.3.0 triggered only by PR smart_holder trampoline shared_from_this support #3023.
5. A valgrind false positive.

Based on the ASAN, MSAN, UBSAN testing, 1. and 2. do not seem to be the most likely root causes, although only proving what the root cause actually is can rule that out completely. Unfortunately this is likely to be very time consuming.

../../tests/test_class_sh_trampoline_shared_from_this.py .............. [ 38%]
==7052== Invalid read of size 1
==7052== at 0x485AC8: subtype_dealloc (typeobject.c:1343)
==7052== by 0x46C5C4: _Py_Dealloc (object.c:2209)
==7052== by 0x44354B: _Py_DECREF (object.h:430)
==7052== by 0x44354B: frame_dealloc (frameobject.c:582)
==7052== by 0x46C5C4: _Py_Dealloc (object.c:2209)
==7052== by 0x53950B: _Py_DECREF (object.h:430)
==7052== by 0x53950B: _Py_XDECREF (object.h:497)
==7052== by 0x53950B: tb_dealloc (traceback.c:167)
==7052== by 0x46C5C4: _Py_Dealloc (object.c:2209)
==7052== by 0x4362A5: _Py_DECREF (object.h:430)
==7052== by 0x4362A5: BaseException_clear (exceptions.c:78)
==7052== by 0x4368E1: BaseException_dealloc (exceptions.c:88)
==7052== by 0x46C5C4: _Py_Dealloc (object.c:2209)
==7052== by 0x47DA04: _Py_DECREF (object.h:430)
==7052== by 0x47DA04: _Py_XDECREF (object.h:497)
==7052== by 0x47DA04: tupledealloc (tupleobject.c:237)
==7052== by 0x46C5C4: _Py_Dealloc (object.c:2209)
==7052== by 0x45B2D6: _Py_DECREF (object.h:430)
==7052== by 0x45B2D6: _Py_XDECREF (object.h:497)
==7052== by 0x45B2D6: dict_dealloc (dictobject.c:2018)
==7052== Address 0x7cdd0a9 is 185 bytes inside a block of size 936 free'd
==7052== at 0x483C9FC: free (vg_replace_malloc.c:538)
==7052== by 0x46FA5E: _PyMem_RawFree (obmalloc.c:127)
==7052== by 0x471162: PyObject_Free (obmalloc.c:709)
==7052== by 0x54AB9D: PyObject_GC_Del (gcmodule.c:2329)
==7052== by 0x483B71: type_dealloc (typeobject.c:3486)
==7052== by 0x8544410: pybind11_meta_dealloc (class.h:231)
==7052== by 0x46C5C4: _Py_Dealloc (object.c:2209)
==7052== by 0x8537F7B: _Py_DECREF (object.h:430)
==7052== by 0x8545146: pybind11_object_dealloc (class.h:438)
==7052== by 0x485AC7: subtype_dealloc (typeobject.c:1337)
==7052== by 0x46C5C4: _Py_Dealloc (object.c:2209)
==7052== by 0x44354B: _Py_DECREF (object.h:430)
==7052== by 0x44354B: frame_dealloc (frameobject.c:582)
==7052== Block was alloc'd at
==7052== at 0x483B7FB: malloc (vg_replace_malloc.c:307)
==7052== by 0x46FAD1: _PyMem_RawMalloc (obmalloc.c:99)
==7052== by 0x4710D2: PyObject_Malloc (obmalloc.c:685)
==7052== by 0x549BAD: _PyObject_GC_Alloc (gcmodule.c:2237)
==7052== by 0x54A974: _PyObject_GC_Malloc (gcmodule.c:2264)
==7052== by 0x488CAA: PyType_GenericAlloc (typeobject.c:1047)
==7052== by 0x48FD40: type_new (typeobject.c:2596)
==7052== by 0x488209: type_call (typeobject.c:1014)
==7052== by 0x43396B: _PyObject_MakeTpCall (call.c:191)
==7052== by 0x433C64: _PyObject_FastCallDictTstate (call.c:113)
==7052== by 0x433CF8: PyObject_VectorcallDict (call.c:142)
==7052== by 0x6A0E00: builtin___build_class__ (bltinmodule.c:232)
==7052==
{
<insert_a_suppression_name_here>
Memcheck:Addr1
fun:subtype_dealloc
fun:_Py_Dealloc
fun:_Py_DECREF
fun:frame_dealloc
fun:_Py_Dealloc
fun:_Py_DECREF
fun:_Py_XDECREF
fun:tb_dealloc
fun:_Py_Dealloc
fun:_Py_DECREF
fun:BaseException_clear
fun:BaseException_dealloc
fun:_Py_Dealloc
fun:_Py_DECREF
fun:_Py_XDECREF
fun:tupledealloc
fun:_Py_Dealloc
fun:_Py_DECREF
fun:_Py_XDECREF
fun:dict_dealloc
}

$ grep 'The CXX compiler identification' 0_last_working.txt 1_first_broken.txt 0_last_working.txt:2021-06-29T23:19:15.2998535Z -- The CXX compiler identification is GNU 9.3.0
1_first_broken.txt:2021-06-29T23:32:34.3368079Z -- The CXX compiler identification is GNU 9.3.0

$ grep 'Unpacking python3.9' 0_last_working.txt
2021-06-29T23:16:57.2875405Z Unpacking python3.9-minimal (3.9.5-3~20.04.1) ...
2021-06-29T23:17:00.4808704Z Unpacking python3.9 (3.9.5-3~20.04.1) ...
2021-06-29T23:17:00.5488969Z Unpacking python3.9-dbg (3.9.5-3~20.04.1) ...
2021-06-29T23:17:02.0151383Z Unpacking python3.9-dev (3.9.5-3~20.04.1) ...
2021-06-29T23:17:02.0779172Z Unpacking python3.9-venv (3.9.5-3~20.04.1) ...

$ grep 'Unpacking python3.9' 1_first_broken.txt 2021-06-29T23:30:12.7628298Z Unpacking python3.9-minimal (3.9.5-3~20.04.1) ...
2021-06-29T23:30:15.9816223Z Unpacking python3.9 (3.9.5-3~20.04.1) ...
2021-06-29T23:30:16.0515017Z Unpacking python3.9-dbg (3.9.5-3~20.04.1) ...
2021-06-29T23:30:17.5608930Z Unpacking python3.9-dev (3.9.5-3~20.04.1) ...
2021-06-29T23:30:17.6301530Z Unpacking python3.9-venv (3.9.5-3~20.04.1) ...

[rwgk]

Logging a fresh observation for future reference (thanks a lot to @Yhg1s for making this possible!):

Google-internal testing with (ASAN, UBSAN) x (Python 3.9) x (fastbuild, opt, dbg) is clean

BUT

MSAN x (Python 3.9) x (fastbuild, opt, dbg) fails, pointing to the exact same line number in Objects/typeobject.c as valgrind. See below.

This makes potential root causes 4. and 5. in the PR description very unlikely.

Also note that test_class_sh_trampoline_shared_from_this.py is the only failing test. I had to exclude 5 tests that need numpy/scipy/eigen, but the rest runs clean with all sanitizers.

For completeness, the 5 excluded tests are: test_buffers.py, test_eigen.py, test_numpy_array.py, test_numpy_dtypes.py, test_numpy_vectorize.py.

platform linux -- Python 3.9.5+, pytest-6.2.3, py-1.10.0, pluggy-0.9.0
configfile: pytest.ini
collected 14 items
test_class_sh_trampoline_shared_from_this.py .............. [100%]
============================== 14 passed in 0.52s ==============================
==5483==WARNING: MemorySanitizer: use-of-uninitialized-value
#0 0x5639a075a73d in subtype_dealloc Objects/typeobject.c:1343:46