cfscript: support a new expression as a function-listener target (#98)

[ghedwards]

Closes #98 — the residual half, after #102 shipped the arguments-required change.

threadName = new Query():function( result, error ) { … }; parses. That completes the eleven forms in Lucee's Function Listeners recipe; the other ten landed in #96 and #97.

The number this issue was parked on had expired

#98 records the widening at +591 states and parks the issue on that. That figure was measured when new_expression could complete on the bare keyword new — and #102 removed exactly that by requiring its arguments. Re-measured on today's base, the same widening costs +14.

The transferable part: a cost taken before a related change is not evidence about after it. Nothing about this feature got cheaper; the measurement simply stopped describing the grammar it was taken from.

What actually blocked it was precedence, not size

With the target widened, new Foo() before a contested : is either a listener target or a ternary consequence, and only an open ? tells them apart — which is what GLR carries. The rule shipped in #87 sits at prec.right('call', …), and 'call' binds tighter than 'ternary', so the listener reading won inside c ? new Foo() : obj and that ternary stopped parsing. Every earlier attempt failed this way, including the +260 variant that restricted the listener side to a function literal.

The fix is a second arm below 'ternary', plus the conflict [$.expression, $.function_listener_expression] that tree-sitter generate asks for. Where a ? is open the ternary wins; where none is, nothing competes for the colon. The call-target arm keeps 'call' and is untouched.

input	before	after
new Query():function( r, e ) { … }	ERROR	function_listener_expression with a new_expression target
new org.lucee.cfml.Query():callback	ERROR	parses
c ? new Foo() : obj	ternary	ternary
isNull(o) ? new() : o	ternary	ternary
Query():function( … ) (call target)	parses	parses
x = new.foo, var new = 1, obj.new()	identifier reads	identifier reads
a[ f() : g() ] slice, case f():	parse	parse

Verification

check	result
npm test	346/346 (116 cfml, 158 cfscript, 72 cfquery)
npm run probe	50/56, no drift; cfscript/function_listener_new.cfc flips to pass
npm run lint	clean
corpus scan, 15,392 files	640 → 638 error nodes across 121 → 120 files — FunctionListener.cfc to zero, nothing regressed
npm run treediff vs master	zero changed trees in both grammars
STATE_COUNT	5315 → 5368 (+53), one declared conflict
npm run bench	cfscript +1.0% against untouched controls spanning 4.2 points — cfml +2.5%, cfquery −1.7%

The benchmark is the gate CLAUDE.md requires for a declared conflict, and the controls are the point: the subject moved less than two grammars that did not change at all, so the cost is inside the noise floor. That matches what the prefix predicts — this conflict is live on a new_expression followed by : and nowhere else, unlike the identifier ( shapes that cost 2.8×.

Also in the diff

• Corpus test function listener on a new target (#98) pinning both ternary controls beside the construct, because they are what every earlier attempt broke.
• LIMITATIONS.md and docs/FAILING-PATTERNS.md rewritten from rejection to record, keeping the expired-measurement lesson and the still-valid reasons new is not reserved (395 .new( call sites across 160 corpus files; Lucee's reserved-word list is null, true, false).
• CHANGELOG.md under ## [Unreleased], and the in-flight entry in docs/TODO.md.

🤖 Generated with Claude Code

https://claude.ai/code/session_0117EvdX3EgLgDevVNrWZ5hr

---

Generated by Claude Code