Cannot convert Rational to float

[Wrzlprmft]

from symengine import Rational
x = Rational(2,3)
print(float(x))

raises a TypeError.

The best workaround I found so far is float(x.n().real_part()).

[certik]

@Wrzlprmft thank you for the bug report. @isuruf fixed it in #265.

[Wrzlprmft]

Unfortunately the fix (or some other recent change) seems to destroy my work-around (see neurophysik/jitcdde#17). More specifically, symengine.Float(42).real_part does not exist (only symengine.Float(42).real).

Can you either make a release (that allows me to pinpoint the version requirements) or implement a real_part attribute for Floats for backwards compatibility? Since everything is hardcoded, I can not monkey-patch this either.

[isuruf]

That fix should be backwards compatible. Which version of symengine are you using?

[Wrzlprmft]

That fix should be backwards compatible. Which version of symengine are you using?

I am comparing a version I pulled a few minutes ago from GitHub and the latest pre-release from PyPI (0.3.1.dev1). What I need something that converts any of SymEngine’s non-complex number types (Integer, Rational, Float, …) to a Python float and works in both of these versions (or a new release/pre-release).

Suppose x is my SymEngine number (e.g., x=symengine.Rational(2,3)). In the current version, I can just use float(x), but that does not work in the pre-release. In the pre-release I can use float(x.n().real_part()) but that does not work in the current version (“'RealDouble' object has no attribute 'real_part'”); the closest I can get is float(x.n().real).

[certik]

@Wrzlprmft is the issue that you are trying to make your code work with multiple versions of symengine?

Could one have a check based on a version of symengine and do what works for each? I understand that's not ideal, but at least it's some way forward.

[isuruf]

@Wrzlprmft, use float(x.n(real=True)) which should work in both versions.

[Wrzlprmft]

@Wrzlprmft is the issue that you are trying to make your code work with multiple versions of symengine?

Correct. Alternatively, I can deal with this using explicitly requiring a version, but that would require a release.

Could one have a check based on a version of symengine and do what works for each? I understand that's not ideal, but at least it's some way forward.

sympy.__version__ returns 0.3.1.dev1 in both cases, so that wouldn’t work. What I could do is to catch the exception, but I see no easy way to avoid catching some real problems due to faulty user input or similar. Anyway:

@Wrzlprmft, use float(x.n(real=True)) which should work in both versions.

Indeed. That solves my problem. Thank you.

[certik]

@Wrzlprmft perfect, I am glad it's solved. Sorry about the inconvenience.