Failing to fetch() Wasm retries the network request

[verhovsky]

If the fetch() fails (no network, 404, whatever) here

emscripten/src/preamble.js

Lines 607 to 608 in 7ed425e

	var response = fetch(binaryFile, {{{ makeModuleReceiveExpr('fetchSettings', "{ credentials: 'same-origin' }") }}});
	var instantiationResult = await WebAssembly.instantiateStreaming(response, imports);

it will be retried, regardless of why it failed here

emscripten/src/preamble.js

Line 619 in 7ed425e

return instantiateArrayBuffer(binaryFile, imports);

emscripten/src/web_or_worker_shell_read.js

Line 44 in 7ed425e

var response = await fetch(url, {{{ makeModuleReceiveExpr('fetchSettings', "{ credentials: 'same-origin' }") }}});

The retry logic shouldn't retry the network call, because the end result is that if you pass a bad URL or serve a broken Wasm file it gets requested twice, just to make sure.

(See also tree-sitter/tree-sitter#4775 )

[sbc100]

I agree this logic is seem wrong. I imagine it dates back to to some ancient time when perhaps instantiateStreaming was new and not always available.

[verhovsky]

The only reason I can see for retrying at all if instantiateStreaming() fails is that it's strict about Content-Type being application/wasm and loading the bytes and instantiating them is a way to avoid that check.

[sbc100]

The only reason I can see for retrying at all if instantiateStreaming() fails is that it's strict about Content-Type being application/wasm and loading the bytes and instantiating them is a way to avoid that check.

I see, so any replacement logic may still have to retry in the case of that specific error. And in theory there could be other error types that might be currently handled by this fallback.

However, I think we should probably find a way to make the fallback optional, and opt-in since instantiateStreaming should work for most users.