Bobby crashes with trigger phrase

[Noremacam]

Observed on my Pebble Time Steel running last official firmware, and the last official Pebble app.
Phone is Pixel 9 Pro.
Bobby is launched from the center Quick Launch button

If I ask Bobby "What is the speed of light in furlongs per fortnight?" The app will show "Getting a calculator" and then freeze, followed by the Pebble logo as the watch reboots.

Edit: After additional reproduction attempts, it seems it only happens if this is the first question after opening the app. I asked it simpler math questions beforehand, and then asked
"What is the speed of light in furlongs per fortnight?" and it came back with an answer without crashing.

[Katharine]

I can repro this. Watch crashes are always fun…

[Katharine]

The response that apparently kills it is:

c<<!!WIDGET:{"content":{"number":"1,802,614,041,067","unit":"furlongs per fortnight"},"type":"number"}!!>>
d

[Katharine]

Weirdly, I can only repro this if I actually dictate the question. If I add it to the preloaded sample question list, it never crashes.

This doesn't make an awful lot of sense in tandem with it causing a whole system crash (which is supposed to be impossible for userland apps in the first place).

[Katharine]

This seems to be caused by the logic to pick the font to use for a number widget:

data->number_font = fonts_get_system_font(FONT_KEY_LECO_32_BOLD_NUMBERS);
data->number_height = 32;
size = graphics_text_layout_get_content_size(widget->number, data->number_font, GRect(0, 0, bounds.size.w + 100, bounds.size.h), GTextOverflowModeTrailingEllipsis, GTextAlignmentLeft);
if (size.w > inset_bounds.size.w) {
  data->number_font = fonts_get_system_font(FONT_KEY_LECO_26_BOLD_NUMBERS_AM_PM);
  data->number_height = 26;
  size = graphics_text_layout_get_content_size(widget->number, data->number_font, GRect(0, 0, bounds.size.w + 100, bounds.size.h), GTextOverflowModeTrailingEllipsis, GTextAlignmentLeft);
  if (size.w > inset_bounds.size.w) {
    data->number_font = fonts_get_system_font(FONT_KEY_LECO_20_BOLD_NUMBERS);
    data->number_height = 20;
    size = graphics_text_layout_get_content_size(widget->number, data->number_font, GRect(0, 0, bounds.size.w + 100, bounds.size.h), GTextOverflowModeTrailingEllipsis, GTextAlignmentLeft);
    if (size.w > inset_bounds.size.w) {
      data->fallback_mode = true;
      data->number_font = fonts_get_system_font(FONT_KEY_GOTHIC_24_BOLD);
      size = graphics_text_layout_get_content_size(widget->number, data->number_font, GRect(0, 0, inset_bounds.size.w, bounds.size.h + 100), GTextOverflowModeWordWrap, GTextAlignmentLeft);
      data->number_height = size.h;
    }
  }
}

My best guess is that loading all four of these fonts (and using them to do measurements) in sequence causes the kernel to exhaust some resource (memory?), resulting in the watch resetting. In the situation where there is no dictation session, though, we don't crash, perhaps because the session is not consuming any resources.

In theory, I don't think there is anything wrong with this code. Certainly not sufficiently wrong to merit the watch rebooting. But alas, the Pebble firmware is quite buggy.

Unfortunately, since the crash doesn't happen without dictation, I can't reproduce it with a debugger attached.

[Noremacam]

Thanks for looking into it!