Memory Leak?

[RKBoss6]

I have a slight problem - I'm trying to develop/modify my existing clock face, Daily Color Clock, and I ran into a problem. The memory is not getting cleared when remove() is called. It's gotten to the point where switching from the clock and launcher 2-3 times will cause a memory hang.
Some numbers I got from process.memory():
When bangle is reset: 1804 bytes
When app is loaded: 7543 bytes
When app remove() is called: 6430 bytes.

In the fast loading tutorials, it mentions that some increase before and after is expected, but surely not >4000 bytes. Can anyone tell me what's going wrong here, and why? I have little experience developing with a fixed memory constraint in mind, and I could not solve this, no matter what I tried. Any help is appreciated!
Code

[gfwilliams]

I took a quick look and I can't see anything obvious. One odditty is:

E.toString(require('heatshrink').deproccompress(

Shouldn't deproccompress be decompress? I tried to run the code and it seems to fail for me (partially because of decompress, but also it doesn't like settings.colorQueue.length=0; because Espruino doesn't support setting array length.

After fixing that, I don't see any memory leaks at all... so I wonder whether the code you're having trouble with isn't actually that code?

About testing for leaks: after loading something there will be some increased memory use (libraries/etc get loaded). You can *really *tell by unloading and reloading though:

• Upload app via IDE to "dailycolorclk.app.js"
• Run Bangle.setUI() to remove the clock
• Check process.memory().usage
• Now run eval(require("Storage").read("dailycolorclk.app.js")) which will effectively 'fast load' the app
• Run Bangle.setUI() to remove the clock
• Check process.memory().usage

And the two usages you see should be kind of the same. You can then try the eval..setUI again and see if usage goes up but at that point it shouldn't

[RKBoss6]

So.. They aren't the same. The normal load once removed gives 6199 bytes, and the eval load once removed clocks in at 9199 bytes. I fixed the decompress and array length errors, and I really have no clue what's going on. I've updated the code in the link to reflect my changes, anything strange you see?

[gfwilliams]

Well, given I'm not seeing the memory leak, but you load ClockInfos, it makes me wonder whether you have some ClockInfo that is leaking data?

Clockinfos do get loaded and unloaded, so if one of them allocated something that didn't get freed then that could cause a problem.

What I'd suggest is you:

• run it twice (load and then eval) as you did so you have 9199 vars used.
• disconnect the Web IDE
• Go to https://espruino.github.io/EspruinoMemView/vis.html and connect

It might take a while to load all those vars, but you should see a visual graph of all the allocated memory which you can drag around.

It'll be pretty big and a bit of a pain to wade through, but I imagine you'll see two big blobs that kind of match (from the two times you ran the clock) and based on the names of those variables that were allocated it might give you an idea of where the leak is.

... you could also uninstall any non-standard clockinfos and see if that fixes it?

[RKBoss6]

I tried the mem viewer and I could not find anything related to clock infos anywhere in my search - even though it definitely is there. How easy is it to add a search to the mem viewer so we can search things up?

[gfwilliams]

Well, it's all on GitHub so you can take a look :) But really it's not about finding ClockInfo, it's trying to find out what's there twice.

Probably if you can't see two bits of memory that shouldn't be there, the best thing is connect with the Web IDE:

• Load your clock, call setUI
• Run E.dumpVariables() and save the output to a file
• eval the app, then call setUI again
• Run E.dumpVariables() and save the output to a second file
• Diff the 2 files - and you should see what's new, and maybe some variable names will catch your eye

[RKBoss6]

The IDE only allows a set amount of lines outputting, so as a result I can save the last third of the output to a file. Is there a way to disable this?

[gfwilliams]

IDE -> settings -> General -> Terminal Log Icon = true

Then you can keep a log of all the data you receive

[RKBoss6]

Thanks! I checked it out, and the majority of items that were new with the eval were variables from this clock, and, for some reason, a bunch of weather variables (drawSnow, DrawCloudy, chooseiconbytxt...) and many widgets. I saw nothing related to clockInfos...

[gfwilliams]

Ok, well that's something (even after using .setUI to unload the clock?) - could be a bug in the weather app that causes it to keep being allocated for some reason? Do you see a memory leak with other clocks that use Clockinfos? Maybe try uninstalling the weather app and see if that helps?

I did just update https://espruino.github.io/EspruinoMemView/vis.html so you can now:

• Load the clock, do Bangle.setUI() in the IDE to unload it
• Click connect
• In the IDE (hopefully you can have both connected at once? I can here) eval the clock and then do Bangle.setUI() to unload it again
• Click 'Diff' in the memory viewer

So now hopefully it'll make all the previously allocated variables grey, so the ones in colour are new ones (eg the memory leak). It's not foolproof as vars could have moved around but it should give you an idea.

Also now clicking on a node will try and show you a 'path' which might point to where the variable is allocated if it's easy for it to find out

[RKBoss6]

Well, I've no idea what's going on here, but somehow, deleting and reinstalling non-essential clockInfos has worked some magic, and I don't get that much of a leak anymore, certainly not the the point of crashing my watch. Thank you so much for all the help in finding the problem, I'll update this with anything new that pops up in the future, but for now, I think we've solved the issue. I've modified countless apps without changing the github code itself, so it's possible one of those mods caused the leak. Everything seems to be normal now.

Thanks once again!