lvgl: Regression in v9, lodepng fails if cache is too small

LVGL version

v9.0.0

What happened?

In Tasmota, we have the same build for devices with PSRAM or without PSRAM. With PSRAM we can set define LV_CACHE_DEF_SIZE to a decently high value. But without PSRAM, the cache size is very small, or potentially set to 0.

The change is done at runtime with lv_cache_set_max_size()

This was not a problem with v8, but in v9 if the cache is too small to fit the image, the decoding fails.

However if the firmware is compiled with #define LV_CACHE_DEF_SIZE 0 decoding works fine although there is no cache.

My ask: if the cache is too small to fit the new image, decode it without caching instead of failing and not displaying the image.

How to reproduce?

No response