Bangle watch: Unable to retrieve board information. Connection Error?

[espruino-discuss]

Posted at 2019-11-25 by bertyhell

When i try connecting to my bangle watch i get this error:
Unable to retrieve board information.
Connection Error?
Connected to Web Bluetooth, Bangle.js 0f49 (No response from board)

I tried resetting the device by holding buttons 1 and 2 according to this guide: https://nodewatch.dev/troubleshooting

I'm trying to get a new game to run on this watch. It's a maze runner based on the wolfenstein3d raytracing algorthm.
You can try it on desktop here: https://bertyhell.github.io/nodewatch-mazerunner/

The github repo: https://github.com/bertyhell/nodewatch-mazerunner

Attachments:

• error.jpg

[gfwilliams]

Posted at 2019-11-25 by @gfwilliams

Try setting 'programmable' to true in settings - 99% of the time that's the issue...

A proper 2 button reset should have done it but maybe you didn't hold BTN1 down for long enough that the Bangle.js logo showed?

It'd be great if you got the maze working!

[espruino-discuss]

Posted at 2019-11-30 by bertyhell

Yes that worked

I reworked the project structure a bit so i can have a desktop (canvas) and a bangle version to build automatically. Had to write my own packaging algorithm since the webpack wrapper code didn't play well with the espruino interpreter.

I also added minification using unglifyJS which replaces all the function names and variables by single letters. With an exclusion list for the Bangle. and g. functions

Even with all that it seems the Bangle goes out of memory when i start the game.

How can i inspect what is eating memory in my app?

I tried with the chrome dev tools on the desktop version to look at the heap size, but i think too much stuff is loading to support to the canvas rendering, that i won't translate well to the heap inside the Bangle.

If you want to try and run it, i've included the built file in the repository:
minified: https://github.com/bertyhell/nodewatch-mazerunner/blob/master/dist/bangle.min.js
non-minified: https://github.com/bertyhell/nodewatch-mazerunner/blob/master/dist/bangle.js

Repo: https://github.com/bertyhell/nodewatch-mazerunner

Attachments:

• memory.jpg

[espruino-discuss]

Posted at 2019-11-30 by AkosLukacs

There is E.getSizeOf() that gives the size of a given thing (object, function, etc). Maybe start with a really small maze and print process.memory().free at the startup, so you have an idea of how many memory you have.
You can try Settings -> Communication -> Modules uploaded as functions. and Settings -> Minification -> Pretokenise code.

[espruino-discuss]

Posted at 2019-11-30 by @allObjects

@bertyhell

welcome to the 70'... regarding memory consumption... rather say: non-consumption!

All the good stuff we learned since then how to build good, robust, extensible, tool supported and all the virtues that come to your mind about software construction ... just aren't good enough anymore when it comes to resourcefulness - in the context of Espruino context / interpreter implementation.

Minification is a good start, and so is pre-tokenizing, and storing modules as functions, and saving to flash on upload...

Since Espruino interprets on the source, you may buy some space with some cycles - the old law still applies that nothing is for free - and make 'global'/public names as short as possible... even references to very often used functions that have a nice but not really helpful name when it comes to space.

I did not use pre-tokenize yet, but I used alternate variable names when it comes even to elements of the JS language... for example I used a lot of Math.min, max, floor, ceil, abs functions... and made them short named properties of the object that uses it. Furthermore, I abandoned nice object orientation and replaced it with indexed arrays where bit encoding was not an option (Have though no real figure for what I saved... but I got it working). Another thing I did - a while ago - I described the pac man maze in a coded way... otherwise, just now way to get it into the space of the Original Espruino... Storing stuff as functions and save to Flash on upload helps a lot to get the code out of the way... but all the other variable things have to fit into the memory.

Looking forward to see your maze game working!

[espruino-discuss]

Posted at 2019-12-01 by bertyhell

fyi Minification -> Pretokenise code break this code sample:

function getValue() {
  return "test";
}

const val = getValue();

console.log(val);

 ____                 _
|  __|___ ___ ___ _ _|_|___ ___
|  __|_ -| . |  _| | | |   | . |
|____|___|  _|_| |___|_|_|_|___|
         |_| espruino.com
 2v04.218 (c) 2019 G.Williams
>Uncaught SyntaxError: BREAK statement outside of SWITCH, FOR or WHILE loop
 at line 1 col 1
return"test";
     ^
in function "getValue" called from line 3 col 15
const val=getValue();
                   ^
undefined
>

This might be something that should still be fixed in the beta

[espruino-discuss]

Posted at 2019-12-01 by @allObjects

...forgot to mention:

Initially, in my setup for a ui I had a lot of level 0 code that built the ui, which means: code that is immediately executed on upload. To have the build in level 0 intended to make what is built to be saved. But it made me run out of memory. After packing it into a function and run it onInit() overcame the memory issue. Whit that I conclude that Espruino deals better with functions invoked at or after onInit() than with lengthy level 0 code. Could be a double buffering issue or a-like, but have no further details.

[espruino-discuss]

Posted at 2019-12-01 by bertyhell

@allObjects
i printed out the free memory throughout the app and it shows clearly where the issues lie. I have a path forward now to optimize the code. Thx for the help.

At startup i use about 40% of the available memory.
Then the render function starts and gets through around 100 of the 240 horizontal screen rays before running out of memory.

So i only have to make my program around 2.5 times more efficient :p

Attachments:

• graph.jpg

[espruino-discuss]

Posted at 2019-12-01 by bertyhell

ok, quick update

skipping the postprocessing step needed for drawing the vertical walls, i can get the memory under control.

Then the game runs at about 1 frame every 8 seconds.

If i reduce the number of rays that are cast to 240/8, then it runs at 1 frame per second.

I guess i can optimize it more by keeping track of the time it takes to do the render function and see where the bottlenecks are.

If you want to try it: https://github.com/bertyhell/nodewatch-mazerunner/blob/master/dist/bangle.min.js

---

Attachments:

• maze-runner.jpg
  
• memory-2.jpg

[espruino-discuss]

Posted at 2019-12-01 by @allObjects

Post made right after #8 (before #9 just showed).

Quite interesting graph...

Have to take a closer look at the code to understand what it does and why all of a sudden such a degrade happens.

Is there a lot of data recursion depth? - It cannot be code, because the stack is 'very' shallow...

[espruino-discuss]

Posted at 2019-12-01 by @allObjects

Since the 'geometry' is quite simple, could a different approach - with limited / course granularity regarding rate of stepping thru the maze and viewing direction - not yield higher frame rate?

[espruino-discuss]

Posted at 2019-12-01 by @MaBecker

Wow, this gone be fun!

[gfwilliams]

Posted at 2019-12-04 by @gfwilliams

Nice - thanks for the graph! Looks like you're dumping scanlines in an array? Using Uint8Array or some other typed array may be a lot faster and more memory efficient.

I'll look into the pretokenise - pretty sure it's an IDE issue with it using the wrong IDs for tokens, so I should be able to clear that up without a firmware update