Canonicalization of reactions and multi-SMILES: component-wise, or global?

[avaucher]

The current implementations of canonicalize_multicomponent_smiles and canonicalize_compounds, are doing molecule-wise canonicalizations, not changing the order of molecules.

As raised in a discussion with @A-Thakkar, one may also expect the canonicalization to take care of reordering the components: one may expect to obtain the same representation when "canonicalizing" two identical reactions where only the order of the compounds is different.

It is still unclear which of the two approaches should be preferred.

[avaucher]

Maybe we could add a flag canonical_ordering: bool to those functions? (with default value False?)

To be investigated: sorting the compounds alphabetically, would we obtain the same ordering as direct canonicalization with a dot-separated SMILES string with RDKit?

[drugilsberg]

I like the idea of adding a dedicated flag to impose this (I would default to False for backward compatibility). One can find useful to have the components sorted in a "standard" way.

Regarding the last question, I think it's not a problem to diverge from the direct canonicalisation behaviour, but maybe I'm overlooking some obvious problems.