Stop a call in an attribute value swallowing the tag's body

[ghedwards]

Closes #45.

query name="q" cachedWithin=t { echo("x"); }                        // ok
query name="q" cachedWithin=createTimeSpan(0,0,0,5) { echo("x"); }  // mismatched input '<EOF>'

Not the parenthesis, as the issue guessed

I suspected prediction committing to the parenthesised-attribute branch. It's simpler: functionCall carries an optional trailing block.

functionCall : (identifier | specialWord) LEFTPAREN argumentList RIGHTPAREN
               body=compoundStatement? ;

So createTimeSpan(0,0,0,5) { echo("x"); } parses as a call with a body, taking the block the tag needed and leaving cfmlfunctionStatement staring at EOF.

That label was never read

The ctx.body handling in CFExpressionVisitor belongs to visitQualifiedFunctionCall, whose own rule keeps its body. That's why deleting this one compiles without touching a line of Java, and why no fixture's recorded tree changes. It existed only to consume input.

The forms that genuinely want a body still have their own:

cfhttp(url="a" method="GET") { }                // cfmlfunctionStatement
cfmail(to="a", from="b") { writeOutput("x"); }  // tagFunctionStatement

Both verified unchanged.

Verification

• 313 cfparser tests, ./gradlew build

• Differential harness 8 → 7 against master, nothing newly broken

  Worth noting how I checked that: my first diff was against the No arrow function parses inside <cfset>, and the failure is silent #44 branch's results, which flagged common: thin-arrow lambda as newly broken. It isn't — No arrow function parses inside <cfset>, and the failure is silent #44 isn't on this branch, so that case is still failing exactly as it does on master. Re-diffed against master's actual eight. A baseline from the wrong branch is as misleading as diffing the count.

• CFLint 675 against a clean build

• tagscript/tag_attribute_call_45.cfc fails with cfml.parsing/src/main stashed

---

Generated by Claude Code