Warn about arguments the callee never declared - #92

Passing a named argument no parameter matches, or more positional arguments than there are parameters, was silently dropped: the call ran, the parameter kept its default, and a typo'd argument name was undetectable. Real OpenSCAD warns for both (Parameters.cc's parse_without_defaults), so this port now does too, with the same message text:

WARNING: variable b not specified as parameter in file t.scad, line 3
WARNING: Too many unnamed arguments supplied in file t.scad, line 5

Where the check goes

Five places, because none of the argument-binding paths share a choke point:

That last row is the one that isn't obvious: the transform/color/hull/minkowski/render/extrude/projection/offset/roof/CSG family never reaches evalModularCall once the bytecode compiler turns it into a wrap op. tests/test_unexpected_args.cpp runs every case twice — VM off and VM on — and asserts the two agree, which is what surfaced it.

Rules ported

• $-prefixed names other than $children are exempt (isConfigVariable, mirroring ContextFrame::is_config_variable) — a $-name is a dynamic-scope override, never a parameter.
• "Too many unnamed arguments" fires once per call, like upstream's warned_for_extra_arguments latch.
• Builtin parameter names live in builtinParamNames (registry.cpp), each list the union of upstream's own Parameters::parse declaration and any extra name this port reads via getArg, so it never warns about an argument it goes on to honour. Add to it whenever a builtin gains a parameter.
• Builtin functions get no entries beyond textmetrics/fontmetrics: upstream reads their arguments positionally without Parameters::parse, so sin(bogus=30) warns about nothing there (verified against 2022.08.22) and warning would be a divergence, not a fix.
• Not ported: argument X supplied more than once and argument X overrides positional argument, separate conditions this port's simpler positional-matching rule doesn't track.

Verification

• Message-for-message parity with OpenSCAD 2022.08.22 on a script covering user modules, user functions, function literals, nested/compiled calls, and every builtin group.
• False-positive sweep: the 901 scripts extracted from BOSL2's own tests/*.scadtest, plus BOSL2's examples/, plus Dalek.scad — zero unexpected-argument warnings from any of them.
• Full suite green both ways: 1477/1477 with the VM off and with OSCAD_BYTECODE_VM=1.
• No measurable render-time change on Dalek.scad.

🤖 Generated with Claude Code