Add no-cgo support

[webmaster128]

This is a follow-up of #373. It implements the same ideas, but with the following additions:

• Clarify that internal/api cannot be built without cgo
• Move some types from api to wasmvm/types
• Add documentation for the different Go packages
• Add CI testing for CGO_ENABLED=0 builds

[jhernandezb]

I only get 1 issue now

# github.com/CosmWasm/wasmd/x/wasm/keeper
../../../go/pkg/mod/github.com/!cosm!wasm/wasmd@v0.29.2/x/wasm/keeper/keeper.go:128:24: undefined: cosmwasm.NewVM

[webmaster128]

Thank you for checking this, @jhernandezb

keeper/keeper.go:128:24: undefined: cosmwasm.NewVM

This is expected IMO. The moment wasmvm is built without cgo it cannot provide functionality that requires cgo. Creating a VM instance is not possible. So I assume if wasmd also wants to provide a no-cgo build it has to deactivate such calls in wasmd.

[jhernandezb]

I guess that there should be a keeper.go for cgo and non-cgo at least the NewKeeper() method

[jhernandezb]

I got it working with these changes https://github.com/public-awesome/wasmd/commit/cffe53fa0eeee0a9735ac2dd6d419d9d1d25eccf

I think the LibwasmvmVersion() can actually be fixed from this repo and it will essentially be just the keeper change.

[webmaster128]

I think the LibwasmvmVersion() can actually be fixed from this repo

Can it? LibwasmvmVersion() calls into the libwasmvm at runtime to get its version via C/cgo.

[jhernandezb]

It could error for non_cgo but maybe it's something consumers would not expect?

Consumers could also handle the cgo/non cgo build as I did in my branch.

[webmaster128]

It could error for non_cgo but maybe it's something consumers would not expect?

It would be a bit inconsistent to return an error. In theory all the other functions should then also return an error instead of being unavailable. I think we should not mix

• unavailable
• panicing implementation
• erroring implementation

But one thing that we could do is make "none" success output option. Then "none" means no version of libwasmvm was loaded and any other version is coming from the library. Looking at how this is used it might be a nice way to reduce the burden on the wasmd codebase.

[webmaster128]

Hmm, creating a "none" output is also adding a new case. Looking at the wasmd diff again I think it is very bad having to duplicate the whole CLI command file just because you don't want to compile one line. Since you want an error there, I think the most pragmatic solution is is to have an erroring implementation here.

[faddat]

Hey guys, this is amazing.

Thank you!