Optimization to gufe_to_digraph to avoid repeated traversals, calls to GufeTokenizable.to_shallow_dict

[dotsdl]

Builds on the solution to #216.

Added use of a shallow_dicts index to avoid potentially expensive repeated calls of GufeTokenizable.to_shallow_dict, such as for ProteinComponents. Also used as our check if we've already processed the object being visited to avoid going down its dependency tree again, avoiding repeated traversals.

For an AlchemicalNetwork with ~1000 ChemicalSystems and ~2000 Transformations, this optimization reduces gufe_to_digraph execution time from 6min 31s to 811ms.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (16b2776) 82.10% compared to head (1d9c3d1) 81.77%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #219      +/-   ##
==========================================
- Coverage   82.10%   81.77%   -0.33%     
==========================================
  Files          23       23              
  Lines        2939     2941       +2     
==========================================
- Hits         2413     2405       -8     
- Misses        526      536      +10     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ianmkenney]

I think we could move a lot of the logic added to modifier to add_edges instead, allowing add_edges to check if the shallow dict has already been processed. On the other hand, keeping add_edges smaller makes sense (at the expense of code duplication).

By combining you could remove one of the

sd = gufe_obj.to_shallow_dict()
shallow_dicts[gufe_obj.key] = sd

blocks.

Either way works for me, though.

[dotsdl]

Simplified based on your suggestion @ianmkenney! Looks much better now!