No events are given to reply from other than wasm submsg

[loloicci]

The comment

cosmwasm/packages/std/src/results/submessages.rs

Line 204 in 1ce73c6

/// with full Cosmos SDK events.

(SubMsgResponse is a field of Reply) says this contains all the SDK events.
However, x/wasm discards events from other than wasm submsg in /~https://github.com/CosmWasm/wasmd/blob/07c8beea006cd009e0c7b865fb32753f61221e1b/x/wasm/keeper/msg_dispatcher.go#L117 and this is passed to reply in /~https://github.com/CosmWasm/wasmd/blob/07c8beea006cd009e0c7b865fb32753f61221e1b/x/wasm/keeper/msg_dispatcher.go#L147.

Is it intentional? And if it is, I think the reason should be written in spec and cosmwasm's comment for SubMsgResponse.

[webmaster128]

Thank you for bringing this up. The filtering was added for good reason (which I do not know in detail at this point). We'll update the docs to highlight this. Moving to the cosmwasm repo.

[loloicci]

Thank you @webmaster128 for your reply and for updating the docs! If you know who knows this well, could you give this question or tell this ISSUE to them?

[p0mvn]

This leads to the fact that we charge gas for the attributes but not the events here:
/~https://github.com/CosmWasm/wasmd/blob/09c49d81d38990d232080069b9804af35d68a1b0/x/wasm/keeper/gas_register.go#L200-L211

What are the reasons for this behavior? If we filter events, should we also filter out their attributes?

[chipshort]

Events are not part of consensus and therefore we generally cannot assume they are deterministic between systems. We know that wasm events are deterministic, since they are the result of deterministic execution, so in this special case, we can keep them.
I'll update the docs for this now.