Fix multithreaded wasm crash (solves #164)

[chemicstry]

This solves #164 by using a preallocated Uint8Array for calling crypto.getRandomValues.

The stdweb implementation seems fine, as it already uses Uint8Array.

Should this fix be backported to 0.15? There are a lot of crates that still use that version.

[chemicstry]

The CI seems to have failed due to unrelated reason. Can someone rerun?

[josephlr]

Looks good, a few minor changes. Also, now that we are depending on js-sys can you add a TODO where we use Global and Self_ to replace that functionality with js_sys::global()

[josephlr]

Instead of having this structure, we could just do:

enum RngSource {
    Node(NodeCrypto),
    Browser(BrowserCrypto, Uint8Array),
}

and then add the comment about using the temporary buffer when we actually use the temp buffer:

RngSource::Browser(crypto, buf) => {
    // getRandomValues does not work with all types of WASM memory, so
    // we initially write to browser memory to avoid an exception.
    for chunk in dest.chunks_mut(BROWSER_CRYPTO_BUFFER_SIZE) {
        ...
    }
}

[chemicstry]

Sorry for the late response, but do you still want me to fix this? Since you force pushed to my branch and already approved it

[josephlr]

Yes, fixing this would be preferable (I can also do it when i have time).

[josephlr]

This then becomes:

let buf = Uint8Array::new_with_length(BROWSER_CRYPTO_BUFFER_SIZE as u32);
return Ok(RngSource::Browser(crypto, buf));

[josephlr]

Should this fix be backported to 0.15? There are a lot of crates that still use that version.

Ya that should be fine, once we make sure this works, I'll backport it (maybe once we also cleanup the Self usage).