Moving private doctests to pytests. Fix the compatibility issue 0. I-> 0. #895
Conversation
| @@ -999,7 +875,7 @@ class RealNumberQ(Test): | |||
| >> RealNumberQ[0 * I] | |||
| = True | |||
| >> RealNumberQ[0.0 * I] | |||
| = True | |||
| = False | |||
There was a problem hiding this comment.
Recall there is no RealNumberQ in WMA.
Eventually I'd like to remove RealNumberQ from Mathics3. It would be okay to have an eval_RealNumberQ though. Can we revise this revised to use that?
There was a problem hiding this comment.
We have this as a hack to put it in a rule. I agree, we should remove this at some point. By now, I just adjusted the test according to the actual behavior of 0. I
There was a problem hiding this comment.
Ok - but please consider doing this before contemplating anything that includes bigger picture design work.
There was a problem hiding this comment.
I was investigating a little bit about this, and I found this discussion:https://mathematica.stackexchange.com/questions/18647/how-to-check-if-an-expression-is-a-real-valued-number
So, in WMA, there is an internal symbol that makes exactly what RealNumberQ does, but with a slightly different name.
There was a problem hiding this comment.
And actually, in the last version, RealValuedNumberQ is a system symbol.
| @@ -18,7 +18,21 @@ | |||
| from mathics.core.systemsymbols import SymbolIndeterminate | |||
|
|
|||
|
|
|||
| @lru_cache(maxsize=1024) | |||
| # Another issue with the lru_cache: for mpmath, mpmath.mpc(0.,0.) and | |||
There was a problem hiding this comment.
The problem is putting the lru_cache decorator any any function that has an optional parameter.
If we rewrite such function to call another function that does not use optional parameters, but has all of the parameters filled out and put the lru decorator there, I don't think there will be problem.
| @@ -42,8 +56,8 @@ def from_mpmath( | |||
| # HACK: use str here to prevent loss of precision | |||
| return PrecisionReal(sympy.Float(str(value), precision=precision - 1)) | |||
| elif isinstance(value, mpmath.mpc): | |||
| if value.imag == 0.0: | |||
| return from_mpmath(value.real, precision=precision) | |||
| # if value.imag == 0.0: | |||
There was a problem hiding this comment.
If this is going to stay as a comment, then it would be helpful to explain why this does not work. For example, list an example where the evaluation does the wrong thing.
There was a problem hiding this comment.
I asked in the mpmath project mpmath/mpmath#714, and indeed is the expected behavior for mpmath and Python.
In any case, the problem with the lru_cache is fixed by passing the parameter typed=True.
There was a problem hiding this comment.
In another PR then let's reinstate the lru_cache adding the typed=True parameter. This is a better solution than the one proposed before. But there a other places in the code where lru_cache was used, and the split function technique was used. That then should be adjusted not to split the function and use typed=True instead.
There was a problem hiding this comment.
This is weird: I made the change, commit and push it to this branch, but for some reason, it never appeared here. I put this in another PR with the missing lines...
See comments in code for how this can probably be addressed keeping lru_cache. Overall, this looks good. Thanks for undertaking this! LGTM - but see the other small comments. |
In this PR, I start to move some private doctests to Pytest. I also propose some extra related tests and check that the expected behavior matches with obtained in WMA.
From this, it came up that we are handling wrong how
0. Iis evaluated: In master, it is evaluated to0., but in WMA, it is evaluated toComplex[0.`, 0.`].To match this behavior, I had to modify
from_mpmath, and deactivate its lru_cache. The problem now with lru_cache is thatmpmath.mpf(0.)produces the same hash thanmpmath.mpc(0., 0.), which looks wrong.