Remove the unreachable VAR (FINAL | STATIC) alternative

[ghedwards]

Closes #68.

 localAssignmentExpression
-  : (VAR | (FINAL | STATIC) VAR? | VAR (FINAL | STATIC)) left=startExpression ...
+  : (VAR | (FINAL | STATIC) VAR?) left=startExpression ...

The trailing form could not be reached by any input. FINAL and STATIC are both in the identifier rule — put there by #46 so code naming a variable final keeps working — so var final matches the plain VAR alternative with final as the identifier, prediction commits, and the third alternative is never considered.

Removed rather than repaired

var final already means something, and cfparser gets it right:

final = getValue();       ->  final = getValue()
var final = getValue();   ->  var final = getValue()

tree-sitter-cfml's corpus tests exactly these under a case named "keywords as variable names", with the expected tree showing left: (identifier). Making var final x = 1; parse would resolve that ambiguity against a documented, tested reading in favour of one with nothing behind it.

I could find no support for var final x = 1, var static x = 1 or static final x = 1 in:

Source	Result
tree-sitter-cfml corpus	contradicts it — var final = getValue() is a variable named final
tree-sitter-cfml grammar.js	no final/static modifier modelled
Shipped dictionaries (cf9–cf11, lucee5)	no entries
CFLint's real-world test corpus	no usage
cfdocs (foundeo/cfdocs)	cannot answer — 806 functions, 170 tags, no language keywords at all, not even var
Adobe / Lucee docs	unreachable from this environment

final x = 1; is unaffected and stays — Lucee's immutable local, and the reason FINAL is in this rule at all.

Behaviour-neutral, and verified as such

Parse and decompile output across thirteen forms is byte-identical before and after, including the three that fail:

diff before.txt after.txt
IDENTICAL — removing the alternative changed nothing

That is the proof the alternative was dead. It also means the new fixture passes with cfml.parsing/src/main stashed, which is by design rather than a weak test: it exists to pin the keyword-as-name readings against a future change to this rule, not to demonstrate a fix. Worth saying plainly, since every other grammar PR in this series used the revert-check as evidence and this one cannot.

Provenance

VAR FINAL came in with 292ba9c, whose message says final "takes VAR's shape and may combine with it" — an assertion with no corpus case behind it. VAR STATIC was mine in #65, written to mirror it, so half of what this removes I added three PRs ago.

Verification

• 326 tests, ./gradlew build
• Differential harness unchanged at 1, nothing newly broken case-by-case
• CFLint 675 against a clean build

Still open

static final x = 1; remains unparsed. Unlike the above it was never declared, so it is a gap rather than dead grammar, and it has the same evidence problem. If someone can confirm it against a live Lucee or ACF, it is a small change to this same line.

---

Generated by Claude Code