sympy
sympy copied to clipboard
sympy
dummify argument in lambdify should be smarter
The default option in lambdify is dummy=True, which is handles names that aren't valid Python identifiers, like functions, or symbols with weird symbols in them.
However, this option breaks introspection for the lambdified function.
>>> x, y = symbols('x, y')
>>> f = lambdify((x, y), x + y**2)
>>> help(f)
Shows
Help on function <lambda> in module numpy:
<lambda> lambda _Dummy_123, _Dummy_124
Created with lambdify. Signature:
func(x, y)
Expression:
x + y**2
Whereas with
>>> f = lambdify((x, y), x + y**2, dummify=False)
>>> help(f)
you get
Help on function <lambda> in module numpy:
<lambda> lambda x, y
Created with lambdify. Signature:
func(x, y)
Expression:
x + y**2
I think the default should be to dummify only if the names are not valid Python identifiers.
It also seems to break numba.jit (although that has other issues too, see https://github.com/numba/numba/issues/2328).
As I said above, it should only apply when it needs to, i.e., when the names aren't valid Python identifiers.
@asmeurer shouldn't it supposed to throw errors with invalid python identifiers, If you want to set dummify True only for non-python identifiers then we should add a check to see if the identifier is valid or invalid.
I am working on documenting this at https://github.com/sympy/sympy/pull/13485. It's odd that the default is False instead of None for lambdify, even though the default behavior seems to be to only dummify when necessary. For lambdastr, the default is None.
Probably, this and this suggest that the work on this issue is complete. Though I am adding a Could Close label. @asmeurer
Closing this in reference to the above comment. Please let us know if there is a need to re-open this. Thanks for your contributions.
The default for dummify right now is False, but it should be None, which would automatically dummify names that require it, but nothing else (True would still dummify everything).