How to improve Composition.__repr__

[janosh]

There's a problem with Composition.__repr__.

from pymatgen.core import Composition
from pymatgen.core.periodic_table import Species

comp1 = Composition({Species("Fe2+"): 2})
comp2 = Composition({"Fe2+": 2})
comp3 = Composition({"Fe": 2})

These 3 compositions all print

Comp: Fe2

which isn't ideal given only the first 2 are equal.

comp1 == comp2
>>> True
comp1 == comp3
>>> False

Ideally, the output of Composition.__repr__ should allow to recreate the Composition by pasting that string into the Python interpreter. Current __repr__ is lacking oxi_state information and also doesn't produce valid Python code.

Suggestion

How about we change to:

print(f"{Composition('Fe2+:2')!r}")
>>> Composition('Fe2+:2')

This would strictly speaking be a breaking change since people might have written code that depends on how Composition stringifies. Curious to hear opinion @mkhorton @shyuep.

[janosh]

Another example mp-18767:

Composition('Li+:2 Mn3+:2 O2-:4')

Before

Comp: Li2 Mn2 O4

[chiang-yuan]

This is a good proposal, but what if in the future we consider even more detailed information like partial charge, would it be more complicated? I guess we can just round the partial charge to formal charge and use formal charge as species tag?

[janosh]

For added background, the plan is to keep the current __str__ method. In Python str and repr serve 2 different purposes:

• repr goal is to be unambiguous
• str goal is to be readable

repr is more for devs and str more for users.

[Andrew-S-Rosen]

Yup, retracted my comment! :)

[shyuep]

I think changing repr is fine. No one should be using repr for any purposes.

[janosh]

Cool. @tschaume @yang-ruoxi any downstream implications for mp-web / crystaltoolkit here?