D123 - An opcode with no shape row is an error, not a skip

Status: assumed

Builder::check verifies every emitted opcode has a row in the shape table, before it checks anything about identifiers.

Skipping unknown opcodes was the original design and the reasoning was sound in isolation: guessing an unknown instruction's shape would mean reading a literal as an identifier and complaining about a module that is fine. What that reasoning missed is that a skipped instruction's result is never recorded either - so the next instruction to use it is reported as referring to nothing.

That is a false failure naming the wrong instruction, which is worse than either guessing or silence, and it happened the first time an opcode was added without its row. The check now says so directly.

The general shape is worth keeping: a check that degrades on unknown input has to be asked what it does with the parts that depend on the unknown, not only what it does with the unknown itself.