Stack overflow caused by incrementing global variable (LVGL display)

ESPHome
1

I have a very large (2500 lines) lvgl based weather display project in the works. The code works fine when running in the host emulator. When running on an ESP32-S3 I unfortunately get a stack overflow.

The following bit of yaml is what is “causing” this:

        - lambda: |-
            id(arrayindex)=0;
        - while:
            condition:
              lambda: return id(arrayindex)<30;
            then:
              # - lvgl.canvas.draw_line:
              #     id: wind_stick
              #     color: LIGHTGREEN
              #     width: 1
              #     points:
              #       - x: !lambda return 90 + (90 * (cos((id(hist_dir)[id(arrayindex)]-90)*PI/180)));
              #         y: !lambda return 90 + (90 * (sin((id(hist_dir)[id(arrayindex)]-90)*PI/180)));
              #       - x: !lambda return 90 + ((90-(70*id(hist_wind)[id(arrayindex)]/id(maxwind))) * (cos((id(hist_dir)[id(arrayindex)]-90)*PI/180)));
              #         y: !lambda return 90 + ((90-(70*id(hist_wind)[id(arrayindex)]/id(maxwind))) * (sin((id(hist_dir)[id(arrayindex)]-90)*PI/180)));
              - lambda: id(arrayindex)++;

Note the commented code is not in play, the stack overflow happens when just this is added:

        - while:
            condition:
              lambda: return id(arrayindex)<30;
            then:
              - lambda: id(arrayindex)++;

Have I just run into yet more memory limitations - or is there something else I am missing? Again - works fine on the host: platform.

EDIT: possibly memory? Reducing the number of loops to 10 instead of 30 stops the stack overflow… Opinions please?

2

Smells like that one way or another.
Anyway, add the quotes here:

lambda: 'return id(arrayindex)<30;'

3

No dice. I also tried with repeat: count: 30 and the same thing happened. Reducing the number of readings displayed to 10 means that it doesn’t fail immediately. Removing a bunch of sensors AND leaving the count at 10 makes it perfectly stable but no better than my existing 3.5" weather display…

Would I be right in saying LVGL is more memory hungry than simply coding a display in lambda?

Anyhow I’ll proceed on the “lack of memory” theory and do yet more optimisation…

4

Quite likely.
Any psram on board?

5

Yes and supposedly configured correctly according to other forum users who have this board/display combo…