A few points there: * classical shared memory (mapped into the main address space) is very hard to do in the context of Webassembly, especially in a performant

A few points there: * classical shared memory (mapped into the main address space) is very hard to do in the context of Webassembly, especially in a performant way. Not impossible, but complicated. It would be valuable to add, but there are different approaches as well (like sharing additional Webassembly-level memories between instances, for example) * WASI already has a concept similar to select * unix sockets would be a pretty straight-forward addition > Why include error-prone cruft? A big goal here is to make existing software compile to Webassembly, with only a relatively small amount of changes. For writing new software the direction of WASI (going towards the fine-grained capability model) is a very valid approach. But that won't get you compatibility with pretty much all software in existence. And there is space to explore both directions.