chore: remove Identifier (once known as Identities) from codebase.

[jpivarski]

Identifiers were motivated by PartiQL, an experimental combinatorics language that uses set-like referential identity to track particles through a calculation. It would require our data to track a surrogate index through all of its operations, like a Pandas Index.

The Identities (Awkward v1) and Identifier (Awkward v2) were a "foot in the door" implementation to add such a feature. Originally, it was supposed to be implemented before the Awkward 1.0.0 release, but other features were more pressing. I left them in on the theory that it's easier to take them out than add them later, but it's been 3 years now and 2.0.0 is almost ready to ship—it's not going to happen. Moreover, 2.0.0 is almost ready to ship—removing them would mean changing the public API (albeit for the low-level layer, intended for downstream dependencies, rather than data analysts). So it must be done now.

Hopefully, this won't be too hard to merge with the other PRs. The most likely complication is that all Content subclasses and Form subclasses have 1 fewer argument: instead of "identifier, parameters", it's just "parameters". "None" is a likely value to be passed as identifier, and it's a likely value to be passed to parameters, so it shows up as "None, parameters" or "None, None".

---

📚 The documentation for this PR will be available at https://awkward-array.readthedocs.io/en/jpivarski-remove-identifiers-aka-identities/ once Read the Docs has finished building 🔨

[codecov]

Codecov Report

Merging #1845 (e2008d8) into main (6b5d46e) will increase coverage by 0.10%.
The diff coverage is 93.95%.

Additional details and impacted files

Impacted Files	Coverage Δ
src/awkward/__init__.py	97.05% <ø> (-0.09%)	⬇️
src/awkward/_broadcasting.py	93.41% <ø> (ø)
src/awkward/_connect/pyarrow.py	88.46% <ø> (ø)
src/awkward/_util.py	82.15% <ø> (-0.18%)	⬇️
src/awkward/_v2.py	100.00% <ø> (ø)
src/awkward/operations/ak_full_like.py	100.00% <ø> (ø)
src/awkward/operations/ak_singletons.py	96.00% <ø> (ø)
src/awkward/operations/ak_with_name.py	100.00% <ø> (ø)
src/awkward/forms/unionform.py	79.67% <50.00%> (ø)
src/awkward/contents/emptyarray.py	72.28% <57.14%> (+0.23%)	⬆️
... and 38 more

[jpivarski]

Remove 1345 lines that were never tested (Identifiers were almost always None) and the coverage goes up by... 0.1%.

Merging #1845 (10bede1) into main (178b4e9) will increase coverage by 0.10%.
The diff coverage is 93.75%.

[jpivarski]

The docs have (correctly) lost all references to identifier.

[agoose77]

This was a big PR! It will be nice to drop the identities mechanism given that we don't properly leverage it; less code to read and type in future!

I went line-by-line. I'll run a few greps offline to see if we've missed anything, but once that's done I'll approve!

[agoose77]

Also, I took the liberty of removing the v1 content from the Doxygen index, as you already have to touch that file in this PR.

[agoose77]

I checked NumpyArray, IndexedOptionArray, and ListOffsetArray, which I expected to be the most common layout types, and found no misuse.

Do you know what the is_identifier.match excerpt refers to in _prettyprint.py (used in a few places)?

[jpivarski]

Do you know what the is_identifier.match excerpt refers to in _prettyprint.py (used in a few places)?

That checks to see if a field name of a record fits the regex that most languages use as identifiers—in particular, the Datashape language (which inherits it from whatever parser they're using). That's /[A-Za-z_][A-Za-z_0-9]*/.

This PR was not a sed substitution. I checked each usage of strings matching /identi/i (because I wanted to be sure there wasn't anything left-over from the "Identity"/"Identities" days).

I'll also simplify the two f-string substitutions. It used to be all format method calls because we were originally allowing for Python 2. So, little by little, I've been converting them to f-strings, and it seemed to be a good time/place to do that if there were identifiers to remove.

[agoose77]

Do you know what the is_identifier.match excerpt refers to in _prettyprint.py (used in a few places)?

That checks to see if a field name of a record fits the regex that most languages use as identifiers—in particular, the Datashape language (which inherits it from whatever parser they're using). That's /[A-Za-z_][A-Za-z_0-9]*/.

Fab, this was my read of things, but I never closely looked at identifiers before you removed them, so I wanted to clarify.

This PR was not a sed substitution. I checked each usage of strings matching /identi/i (because I wanted to be sure there wasn't anything left-over from the "Identity"/"Identities" days).

You can tell; lots of tricky things to find beyond a guided exploration. Even just reading all the files took a while, nice effort!

I'll also simplify the two f-string substitutions. It used to be all format method calls because we were originally allowing for Python 2. So, little by little, I've been converting them to f-strings, and it seemed to be a good time/place to do that if there were identifiers to remove.

I'm 100% on-board with f-strings. In these two rare cases, I'd either assign the long statements to local variables, or just use a multi-line string.format, hence the suggestion.

[jpivarski]

In this case, I'm going to leave these strings as f-strings, but predefine the variables to substitute. That way, all of the Numba type strings will be generated in the same way.

[jpivarski]

Auto-merging.

Thanks!