Hashing strategy, looking forward

[petosa]

Although the current hashing strategy works, it has a bit of a drawback - anytime you add a new field to the molecule json, the hash for each molecule entry is effectively changed. This means that any time you add/remove a field in all molecules, either you:

• Recalculate all hashes for all molecules, and update all manual references from pages/databases, or
• Ignore the fact that the hash is incorrect and treat it like a unique key, even if that hash no longer reflects the JSON data. This can have problematic drawbacks: if you only have a molecule's JSON, you can no longer just hash it to find its location in the DB. Also, duplicates can potentially slip in.

I'm not sure how often we will need to change the JSON structure for a molecule once we go into production, but it's a good idea to take some proactive measures.

I'm wondering, @dgasmith , what are the minimal fields that define a unique molecule? You mentioned that geometry is not enough, but would hashing symbol + geometry or maybe name + geometry be sufficient? In this way, by basing our hashes on 2 or 3 essential fields, minor changes by adding or removing peripheral fields to all molecule JSONs won't have an avalanche effect on our hashes.

[dgasmith]

From our current list:

  "symbols": ["C", "O", "O"],
  "masses": [16.0, 18.0, 18.0],
  "name": "Carbon Dioxide",
  "charge": 0.0,
  "multiplicity": 1,
  "ghost": [false, false, false],
  "geometry": [ .. ]

This should be enough. We also need one more field called "fragments" that will be a list of list which will need to be included in the hash as well.

I hope that once we build a molecule there will be no reason to rebuild it. However, I cannot definitively say this.

[petosa]

@dgasmith Similar question for pages and databases, what fields define the uniqueness of a page and database document?

For example, is it possible to have multiple databases with the same name? Or multiple pages with the same molecule and method?

[petosa]

In the mean time, as of 357f2803556739a2745e312424904dd0910de143 the hash functions are calculated with the fields described in the readme. The fragments field is also added to molecules.