[pypy-issue] Issue #2015: Mysterious integer overflow bug when performing lots of float operations (pypy/pypy)

[David MacIver]

New issue 2015: Mysterious integer overflow bug when performing lots of float operations
https://bitbucket.org/pypy/pypy/issue/2015/mysterious-integer-overflow-bug-when

David MacIver:

I unfortunately don't have a reproducible case for this bug. I have seen it precisely once and not been able to get it to happen again, let alone produce a minimal example.

On pypy3-2.4.0 I've seen the following crash:


```
RPython traceback:
File "pypy_module_pypyjit_interp_jit.c", line 105, in jump_absolute__AccessDirect_None
File "rpython_jit_metainterp_warmstate.c", line 6552, in maybe_compile_and_run__star_5_1
File "rpython_jit_metainterp_warmstate.c", line 14027, in execute_assembler__star_2_2
File "rpython_rlib_rbigint.c", line 43665, in fromfloat_1
~~~ Crash in JIT! <OverflowError object at 0x1ce1ae0> 
```

This happened when running the following command:


```
#!bash

PYTHONPATH=src python -m pytest tests/test_example_quality.py --capture=no -klist_of_floats_on

```

against git at github.com:DRMacIver/hypothesis.git checked out to the branch repro_for_pypy_1

This command should result in something that looks like the following:

```
============================================ test session starts =============================================
platform linux2 -- Python 3.2.5[pypy-2.4.0-final] -- py-1.4.26 -- pytest-2.7.0 -- /home/david/.pyenv/versions/pypy3-2.4.0/bin/python
rootdir: /home/david/projects/hypothesis, inifile: pytest.ini
collected 28 items 

tests/test_example_quality.py::test_minimize_list_of_floats_on_large_structure Falsifying example: everything_is_terrible(x=[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0, 3.0])
PASSED

```

It is highly likely that this is a case where an OverflowError really should be occuring - there are quite a few of them that will be hit when running this, with the code handling it appropriately.

I've tried a bunch of things I thought might trigger this and nothing seems to have, so it's either quite subtle or I'm missing something. This is why you're getting this slightly useless bug report, sorry.