Drawing vector text stops working after 400~800 updates

[sjlongland]

Hi,

I've got a project where we're using a Interstate75W board (RP2350-based) to display a 2 or 3-digit number. The use case here is to display the number of a horse rider that passes through a checkpoint, so they know they've been recorded as arriving through the checkpoint.

For this we've got it paired up with a 64×32 LED panel. The actual application receives this rider number via (IPv4!) UDP multicast over WiFi. I managed to convert the Terminus TTF font to give us a clear readable numerical display. Via WiFi, we can send a UDP packet that encodes the foreground colour, background colour, and the text to display, and the display updates.

Aside from it jumping around horizontally (it's supposed to be a monospace font, so shouldn't do that), it's perfect, except there's one problem. What I observe is after between 400 ~ 800 updates, the text stops being displayed. At most the background colour changes, but no text appears.

For the sake of application, I've stripped out the network code to focus on the text display:

• demo.py -- application
• terminus.af.gz -- font

Running this application on the i75w_rp2350-v0.0.5-micropython-with-filesystem.uf2 image, I see the number count roughly to around 750 or so, before disappearing. With the network-related code, it only gets to about 400.

There's no exception thrown, functions return None, no apparent change in behaviour, just the text isn't getting drawn on the screen anymore. If I re-start the application, re-drawing happens once more, until it suddenly doesn't some 400+ screen updates later.

If there's an error, we should be either handling it or throwing it, not silently failing and relying on the user to press a reset button.

[sjlongland]

It appears, importing gc and calling gc.collect() immediately after calling i75.update() seems to work around this issue, so there's possibly a memory leak somewhere.

[Gadgetoid]

Any difference with the newer builds here: #42

There were some issues with vector graphics and the opposite of a memory leak- ie: GC collecting things it shouldn't. This may have been fixed by pimoroni/pimoroni-pico@fba0f1c but the latest i75 release predates this fix.

[sjlongland]

Hi,

Dug the board out and gave it a try this morning, the newer build seems to correct this issue. I first ran that demo.py unmodified, and after seeing it do a few cycles, tried speeding up the loop.

No issues counting well past 4000 or so. Looks like the memory leak has been resolved, I'm just testing with the actual firmware I developed to see if there are any more gremlins hiding.

[sjlongland]

Testing for 24 hours using the new MicroPython build having backed out my gc.collect() call, the display is working exactly as expected.

Prior to this, I'd start it, walk away, and a few hours it'd be just a solid colour that changed each time it received a packet; the actual numbers that were supposed to be drawn on top weren't appearing. This seems to be fixed now by this build.

[Gadgetoid]

Good news. Thank you for testing and feeding back, should help me expedite a (long overdue) release!