Are there not spec tests that cover this case ?

Unfortunately I don't think there's a spec test that would make us parse the Wasm 1.0 format when using multi-memory by itself. We could have caught this with a roundtrip test, but I don't think we roundtrip every spec test -- only the actual roundtrip tests. And I think it would have had to be a roundtrip test that uses multi-memory and has --debug-names enabled on the test to catch this (and not --stdout because then we don't do the full roundtrip comparison).

we probably should roundtrip every test - it's what the spec does, as far as we can tell. but that's a whole undertaking...

I agree that would be great (at least for valid modules) -- if it doesn't blow up testing time too much. We could make it an optional test chosen by an environment variable and run it in CI.

If we want to catch everything, ideally we'd be roundtripping with and without folding, with and without debug-names, with and without generate-names, with and without a roundtrip through the spec interpreter's (or wasm-tools's) text-to-binary and binary-to-text pipelines (since we sometimes have an out-of-date binary encoding of something that we don't catch because WABT's writer and reader are consistent).