They still fail to validate after replacing those invalid section IDs with valid ones, such as 01 or 02. For example, Look at the (malformed) module in the first line, which says 0d (13) is an invalid section ID:
(module binary "\00asm" "\01\00\00\00" "\0d\00")
Actually now all major web browsers and Node have implemented EH and expose it by default (= without a flag). So that should be 0e now, even before we merge the EH proposal.
But anyway, this module fails to validate not because 0d is an invalid ID. If you replace it with 01, it still fails to validate, because even if it is an empty section, there should be the number 0 saying it has elements, and to emit that number 0, the length of the payload should be at least 1. But in the example, the length of the payload, which is the next byte after the section ID (0d) is 00, not leaving a space for the number 0. All known sections, even if they are empty, need at least one byte of payload to say there is 0 elements. (Start section doesn't emit a number of elements, but it needs an index instead, so it needs at least a byte of payload as well.)
0d has changed to 0e, and the next byte is the length of the payload of that section, and the next byte says the section has 0 elements, whatever that element is. This way, now we replace 0e with a known section ID, such as 01, the module validates.
I think the purpose of these tests are to have an invalid section with a valid section (01). In that case, we also need not only \00 after the section ID but at least \01\00 to properly emit an empty type section.)
And this updates Node version to v19, because, for a reason I don't understand, Node v18 accepts section ID 0x0e(14)~0x13(19) as valid. I guess they are treated as reserved section number or something. So we cannot test 0x0e as an invalid section ID with Node v18. Node v19 seems to have fixed it.
Current
assert_malformedinvalid section ID tests are passing, meaning they fail to validate, but not because of invalid section IDs. https://github.com/WebAssembly/spec/blob/721dd6ce2cf53028133bd2ceeea9b0655570c641/test/core/binary.wast#L47-L52They still fail to validate after replacing those invalid section IDs with valid ones, such as
01or02. For example, Look at the (malformed) module in the first line, which says0d(13) is an invalid section ID:Actually now all major web browsers and Node have implemented EH and expose it by default (= without a flag). So that should be
0enow, even before we merge the EH proposal.But anyway, this module fails to validate not because
0dis an invalid ID. If you replace it with01, it still fails to validate, because even if it is an empty section, there should be the number0saying it has elements, and to emit that number0, the length of the payload should be at least 1. But in the example, the length of the payload, which is the next byte after the section ID (0d) is00, not leaving a space for the number0. All known sections, even if they are empty, need at least one byte of payload to say there is 0 elements. (Start section doesn't emit a number of elements, but it needs an index instead, so it needs at least a byte of payload as well.)So the line above should change to
0dhas changed to0e, and the next byte is the length of the payload of that section, and the next byte says the section has 0 elements, whatever that element is. This way, now we replace0ewith a known section ID, such as01, the module validates.For these lines:
I think the purpose of these tests are to have an invalid section with a valid section (
01). In that case, we also need not only\00after the section ID but at least\01\00to properly emit an empty type section.)And this updates Node version to v19, because, for a reason I don't understand, Node v18 accepts section ID 0x0e(14)~0x13(19) as valid. I guess they are treated as reserved section number or something. So we cannot test
0x0eas an invalid section ID with Node v18. Node v19 seems to have fixed it.This was the reason for the CI failure in https://github.com/WebAssembly/exception-handling/pull/262.
cc @chfast