How to support the updated EH spec

[dschuff]

The new EH spec will eventually need to be supported in WABT. One new thing in the update is that the opcode for catch_all is 0x05, the same as else. This is not ambiguous because else must occur with an if and not a try, but wabt seems to have the assumption (e.g. in opcode.def ) that there's a 1:1 correspondence.
I haven't looked at how to fix it yet, but wondered if anyone (perhaps @aardappel or @sbc100) had any good ideas about the best way to make it work.

[sbc100]

I'm not too familiar with part of the code I'm afraid.

[lars-t-hansen]

Will anything good come from saving that one opcode, apart from saving one byte in the opcode space?

SpiderMonkey currently assumes a 1:1 mapping in a couple of places. Resolving the opcode overloading is likely not a real hardship, but I still wonder whether the overloading is not simply adding complexity to the spec.

[wingo]

PR here, fwiw #1596

[sbc100]

Will anything good come from saving that one opcode, apart from saving one byte in the opcode space?

SpiderMonkey currently assumes a 1:1 mapping in a couple of places. Resolving the opcode overloading is likely not a real hardship, but I still wonder whether the overloading is not simply adding complexity to the spec.

Agreed, unless there is some significant upside it seems like a bad idea.

[takikawa]

I agree that there are downsides to sharing the opcode. In particular, it's not great for error messages (you could get errors expressed in terms of else but the source had catch_all or vice versa).

Historically, I think this punning was inherited from the an early version of the proposal.

[dschuff]

Thanks @wingo that's handy for just local hacking like I've been doing.
It sounds like we might want to reconsider the opcode though. I opened WebAssembly/exception-handling#147 to discuss that.

[aardappel]

I guess end is technically already overloaded, but there we can just use the single identifier.

Supporting WABT to output different names may be somewhat messy, it would mean that any calls to Opcode::GetName would need context parameter to know which control structure we're currently in.

[aheejin]

We decided to change the catch_all's opcode (WebAssembly/exception-handling#147) and @takikawa landed a PR doing that (#1608). Closing.