Support Tail Call extension

[alexbiehl]

Is your feature request related to a problem? Please describe.

I am trying to run a WASM workload that makes use of the return_call instruction (see https://developer.mozilla.org/en-US/docs/WebAssembly/Reference/Control_flow/call).

Currently execution fails with

loadBytes: compile module: invalid function[2]: invalid instruction 0x12

Describe the solution you'd like

Ideally wazero supports that instruction. Some research revealed it’s part of https://github.com/WebAssembly/tail-call/blob/main/proposals/tail-call/Overview.md. After all it would be nice if wazero supported the tail call extension.

Describe alternatives you've considered

Additional context

[evacchi]

interesting, it might be the first time I find this used in the wild. What kind of workload is that? Not ruling out eventually supporting that proposal, but in many cases you are just a compiler flag away from removing that instruction from the output binary

[alexbiehl]

@evacchi Check out the WASM file attached in this release https://github.com/alexbiehl/sqlc-hs/releases/tag/0.1.0.0. It was produced by the WASM backend of GHC and unfortunately it seems like I can't make it stop emitting that instruction.

[anuraaga]

Hi @alexbiehl - I happened to build a GHC binary recently and found the workaround of using an older ghc-wasm-meta to work fine for now in case it helps.

https://github.com/wasilibs/go-hadolint/blob/main/buildtools/wasm/Dockerfile#L6

@evacchi I think one issue with this in particular is haskell seems to compile wasi-libc themselves, and do enable tail call for it, so even if I try disabling it in my downstream build it ends up in the binary.

[evacchi]

@anuraaga FYI I inteded working on tail calls for a while, and in the meantime I implemented them, but it needs a little more testing

I have a working branch on my own fork: evacchi#61
I haven't opened a proper PR here yet because I wanted to make sure everything is in order :)
I confirm that @alexbiehl sqlc-hs works, and all the spec tests pass, but @ncruces found an issue when running a build of sqlite with tail-calls enabled

[evacchi]

addressing all of the edge cases took a bit more than I expected :D #2403