I have implemented starting the tests in a separate subprocess at
https://github.com/sympy/sympy/pull/1379. With that request, hash
randomization will be enabled by default in the tests and doctests whenever
it is supported. A seed is also given, so that all tests are reproducible.
Please review this, so that we can push it in and start fixing the hash
randomization related test failures.
Aaron Meurer
On Monday, May 28, 2012, Aaron Meurer wrote:
> So I've determined the situation with reproducibility of the hash
> randomization.
>
> You can set the hash randomization seed using the environment variable
> PYTHONHASHSEED, like:
>
> $PYTHONHASHSEED=42 python3.3
> Python 3.3.0a3 (v3.3.0a3:0b53b70a40a0, May 1 2012, 11:39:35)
> [GCC 4.2.1 (Apple Inc. build 5666) (dot 3)] on darwin
> Type "help", "copyright", "credits" or "license" for more information.
> >>> hash('a')
> -5486454195961207828
> >>> ^D
>
> $PYTHONHASHSEED=42 python3.3
> Python 3.3.0a3 (v3.3.0a3:0b53b70a40a0, May 1 2012, 11:39:35)
> [GCC 4.2.1 (Apple Inc. build 5666) (dot 3)] on darwin
> Type "help", "copyright", "credits" or "license" for more information.
> >>> hash('a')
> -5486454195961207828
>
> That's the good news. The bad news is that it's impossible to get
> that seed when it is set automatically by the interpreter. That's
> because the seed is used to generate a secret, and that secret is what
> the interpreter keeps track of. You can access the secret with
>
> >>> from ctypes import c_int, pythonapi
> >>> c_int.in_dll(pythonapi, "_Py_HashSecret").value
> 1758302383
>
> but that for example corresponds to the seed of 42 above.
>
> So if we want to print the seed for testing, we have to manually set
> it, which means that we would have to spawn a new Python process to
> run the tests. For SymPy bot, this is not a problem, because it does
> that anyway. For just running the tests normally, we would have to
> change the test runner to do this. Do people think that this is worth
> the trouble? If a test failure comes out of the blue that is caused
> by some rare hash randomization seed, it would be nice to be able to
> reproduce it (similar to failures with the random module seed). I'm
> not sure how much of a burdon this would be on the test runner (maybe
> it wouldn't be any at all, I haven't tried it yet).
>
> Also, can anyone comment on how py.test might handle this situation?
>
> By the way, the information above comes from
> http://mail.python.org/pipermail/python-porting/2012-May/000317.html.
>
> Aaron Meurer
>
--
You received this message because you are subscribed to the Google Groups
"sympy" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to
[email protected].
For more options, visit this group at
http://groups.google.com/group/sympy?hl=en.
