Ambiguity in placement of metadata.code function annotations

[tlively]

CodeMetadata.md currently says that @metadata.code.XXX annotations must be within function definitions and gives an example:

(func (@metadata.code.hotness "\01") $test (type 0) ...)

But this could lead to ambiguity for annotations that could apply to either functions or instructions:

(func (@metadata.code.func_or_inst? "unknown") nop)

This potential ambiguity could be avoided (and the text format could be made nicer IMO) if the function annotations were placed before function definitions instead:

(@metadata.code.definitely_a_func "yep")
(func nop)

There are no formally specified function annotations, so this text format is only specified here in the tool-conventions repo. If we get stakeholders to agree, we should be able to change this.

cc @yuri91 as well as @fitzgen to route to appropriate wasm-tools folks, @sbc100 for WABT, and @kripken for binaryen. WDYT?

[kripken]

(link to original discussion: WebAssembly/design#1563 (comment))

+1 for putting annotations before the function.

[yuri91]

I don't recall if there was a rationale for putting it after func, but I agree that the proposed change seems more consistent with how it works with instructions (on top of solving the ambiguity).