Name mangling of acronyms

[andreaTP]

The current algorithm for name mangling is(incorrectly?) considering subsequent uppercase characters as separate terms, e.g.:

externalID -> external_i_d as discovered here.

This is a pretty widely adopted pattern in the presence of acronyms.
I understand that its an edge case but I think that we should improve the algorithm to handle those use-cases:

externalID -> external_id

the tricky part is in the presence of suffixes(although completely deterministic) where we should make sure this conversion is applied:

IDExternal -> id_external

another possible source of issues is single-character terms:

MeADeveloper -> me_a_developer

I hope I captured all of the invariants, let me know if I'm missing something 🙏

[baywet]

Hi @andreaTP
Thanks for bringing this up.
I think your examples outline the challenge pretty well: coming up with a heuristic that works better than what we currently have in all/most cases.
I'd like to avoid a "known acronyms list" kind of solution since this is something we've done in the past in other areas and it lead to confusion and/or collisions, and we've had to remove such lists.
Did you have any idea in mind for the heuristic?

[andreaTP]

I'd like to avoid a "known acronyms list"

Completely agreed.
I think that the examples are completely decidable with an algorithm and without heuristics:

externalID -> trailing uppercase characters should be treated as a single term
IDExternal -> in presence of lowercase characters after 1+ uppercase, the last uppercase is the first character of the next term, and the rest of uppercases will be grouped into a single term
MeADeveloper -> fit into the previous logic

I might be missing something but, so far, the reasoning seems to hold, wdyt?

[baywet]

That makes sense. If I follow your logic, we'd need to implement some back propagation for the first case, correct?
I'm a bit worried about the performance impact this will have, especially on larger descriptions.

[andreaTP]

some back propagation for the first case, correct?

uhm, not sure I get this, do you have an example?

I'm a bit worried about the performance impact this will have

I don't think so, a practical implementation of what I described only needs one character lookahead and can be done with a single character-by-character scan of the string(an alternative would be using regexps).
I expect performance to remain completely comparable.