polyhedron(vnf), and object()'s single-element delete entry

[revarbat]

Two independent fixes, both reported while using render() expressions in anger.

object(): a single-element entry deletes that key

a = object(a=42, b=53, c=8);
b = object(a, [["d", 18], ["b"]]);   // -> { a = 42; c = 8; d = 18; }

An unnamed list argument holds entries: [key, value] sets, and the single-element [key] deletes. Only the two-element form was implemented, so any script using a delete got undef for the whole call — silently, with no diagnostic pointing at why.

Checking the rest of the semantics against OpenSCAD 2026.02.01 turned up more than the reported gap:

	before	now
[["b"]] delete	undef	deletes
[["zz"]] delete a missing key	undef	silent no-op
malformed entries	undef, silent	undef + warning

Deletion removes the key rather than blanking it, which is observable because ValueObject is insertion-ordered and oscEqual is order-sensitive: object(a, [["b"], ["b", 99]]) puts b last, while object(a, [["b", 99], ["b"]]) has no b at all. Both are pinned.

All 19 cases — values and warning text — were diffed character-for-character against the reference binary with --enable=object-function. The diff is empty. The warnings reproduce two of upstream's own inconsistencies: the "not a list" case puts spaces inside its parens where every other case does not, and the "unnamed argument" case ends with a trailing space. Those strings are what a user greps for, so matching them is the point.

mergeObjectArgs gained an Evaluator and a position so it can warn; both call sites — the interpreter's builtinObject and the VM's Op::CallFn — pass them through.

polyhedron() accepts a VNF 2-list

[vertices, faces] is what every BOSL2 function passes around and what obj.vnf holds. All four of these are now the same call:

polyhedron(obj);
polyhedron(obj.vnf);
polyhedron(obj.vertices, obj.faces);
polyhedron(spheroid(d=30));          // any BOSL2 VNF

Only considered when faces was not supplied separately, so the two-argument form always wins and can never be reinterpreted. The discriminator is BOSL2's own is_vnf test: a VNF's second element is a list of lists, where a points list that happens to have length 2 has a point there — bare numbers. Such a list is left alone and still errors as the invalid polyhedron it is.

Verified both directions against real BOSL2: polyhedron(spheroid(...)) matches the volume vnf_polyhedron() produces, and vnf_polyhedron(obj.vnf) accepts ours.

Testing

967 tests pass under both OSCAD_BYTECODE_VM=0 and =1.

🤖 Generated with Claude Code