Writing multiple values quickly on a characteristic fails

[saiimons]

I use bleno to emulate a BLE device, which writes values on a characteristic as data packets.

When I switched to Mojave, I started using bleno-mac, and my emulator stopped working. The only output I got is peripheralManagerIsReadyToUpdateSubscribers. Seems it is due to too much data being written.

I was able to verify this by:

• setting up breakpoints before writing data, thus slowing the rate
• setting up a delay in between the writes

[notjosh]

Interesting! I won't have a chance to look at it this week I don't think, and nothing comes immediately to mind.

I'll try setting up a purely native harness and see if it's a macOS issue, or an issue with the dispatching across to JS.

How much data are you writing? Do you have a small repro you can provide?

[saiimons]

I'll try to provide an example, though basically I call updateValueCallback twice in a row.

From my understanding, there is some backpressure management in the Apple framework. So, there should probably be some queueing in place in order to avoid this.

As a reference : https://developer.apple.com/documentation/corebluetooth/cbperipheralmanagerdelegate/1393248-peripheralmanagerisreadytoupdate

[notjosh]

Ahh right, of course, that would make sense! That should be (hopefully) be simple enough. I remember thinking about that, and if(& how) that would be exposed back to the JS side without complicating the non-Apple implementations. I think a "fire and forget" from the JS side is probably fine since it's all async anyway, but I'm not sure if there are cases where you'd want to know when a value is queued (and eventually fired). Thoughts?

[saiimons]

My views are the following: keep it simple (fire and forget) and hide it in the Apple implementation, seems to make quite a lot of sense.

Source: I do it this way on Android ;)

The alternative would be to return a Promise, letting us know when and if the write completes

[notjosh]

I agree! 👍 I'll see what I can manage, I'll let you know when I get a chance to look at it.

[saiimons]

Awesome, if I get some time I'll try to dig in the native code and documentation too.

[cwhittl]

Hey @saiimons and @notjosh did you guys ever figure out how to get this to work? I'm hitting it now too.

[cwhittl]

Well I experimented and found this works, not sure I'm happy about it but if it helps... I tried promises but not very hard, it seems like it's a timing issue because timeout doesn't work.

function chunkData(str, size) {
  const chunks = new Array(Math.round(str.length / size));
  let newo = 0;
  for (let i = 0, o = 0; i < chunks.length; i += 1, o = newo) {
    newo += size;
    chunks[i] = str.substr(o, size);
  }

  return chunks;
}

function wait(ms) {
  const waitTill = new Date(new Date().getTime() + ms);
  while (waitTill > new Date()) { /* Don nothing */ }
}

const chunks = chunkData('Bob is your uncle and stuff like that', maxMessageResponseValueSize);
      chunks.forEach((chunk) => {
        const data = Buffer.from(chunk, 'utf-8');
        console.log(chunk);
        messageUpdateValueCallback(data);
        wait(10);
      });

[notjosh]

Do either of you happen to have a simple repro project? I'm sure it's easy enough, but I don't know how many "multiple" is, or how quickly we're talking.

I think I know the issue, but not certain.

[cwhittl]

@notjosh here is a gist, it's a dumbed down version of what I am working on and has not actually been run as is but should work https://gist.github.com/cwhittl/2bc19de82df28f2b415b8cd77180d2f4

[cwhittl]

If you remove the wait you only see the first callback on my android app, if you leave the wait it does all of them