[spec] Parsing and validation in a single pass seems to be discouraged

[axic]

This test case in binary.wast (https://github.com/WebAssembly/spec/blob/v1.1/test/core/binary.wast#L739-L759) tests for assert_malformed, e.g. parser error. However it also contains a validation error.

This turns out to pose a difficulty to engines, which would try to do parsing and (or a part of) validation in a single pass.

Is this intentional?

[rossberg]

No, every negative test should contain only one mistake. But it's easy to miss cases where that's not the case. Happy to accept fixes. :)

[chfast]

Problem

The problematic test checks br_table "with inconsistent target count (2 declared, 1 given)". But the intention is quite distant from how binary parser works. I.e. it assumes the parser fails because there is not enough targets provided

0e02 00 (missing LEB128 encoded target)

but the parser will just consume the next byte being the end 0b.

0e02 00 0b

and will fail because of the missing final 0b later. Even the test error message suggests that.

The problem is that in the middle we have br_table with 2 targets: 00 and 0b, where 0b one is invalid. The module is considered invalid before we are able to realize it is malformed.

Solution

1. We can just drop the test.
2. Or to preserve more/less the current cause of being malformed we can add some workaround/hacks preventing the additionally consumed byte seen as invalid br_table target. The only idea I have to achieve that is to make 0b target valid by adding 10 more functions to the module.

[binji]

The only idea I have to achieve that is to make 0b target valid by adding 10 more functions to the module.

Right, though you would need to add blocks, not functions. I think it should work, as long as you made the br_table target length long enough too, so it attempts to access outside of the code section.