put= replace property by key / path / index, preserving as many references as possible.remove= remove property by key / path / index, preserving as many references as possible.patch= combine dicts, preserving as many references as possible.merge= deep patch that combines dicts at all levels.nil= null | undefined.value= primitive | list | plain dict.
These terms are used for some internal functions analogous to the external API but are more specialized, don't validate inputs as rigorously, etc.
assoc= same asput, but checks equality only by reference; used internally byputandputIn.
Treat non-value objects atomically: include and replace them entirely, without cloning.
Emerge seeks a balance of performance and simplicity.
put, patch and other functions tend to speculatively create a new data structure before testing it for equality, possibly throwing it away. Should revisit the code and avoid this where possible.
putIn could be defined as a recursive put, but would be significantly slower due to redundant equality checks on every level. A single nested put, with a single equality check, followed by multiple assoc, is much faster.
In modern V8, our simple fold and fold1 seem to perform as a well as an inline loop, and much better than Array.prototype.reduce.
In modern V8, native rest and spread has almost no meaningful overhead compared to hardcoding a fixed set of additional parameters.
Add benchmarks with large real-world data.