force-higher-level (TSA) crashes with free(): invalid next size (fast)

[dominiqueweber]

Hi all,

I ran a TSA analysis for 2025 using all three Sentinel-2 sensors (SEN2A, SEN2B, SEN2C) and it crashed with:

free(): invalid next size (fast)
srun: error: node12: task 0: Aborted
Exit-Code: 134

Environment: FORCE v3.10.04 (Docker image davidfrantz/force:3.10.04), Singularity container, single tile, Sentinel-2 10m BOA/QAI.

I used the same PRM file as in previous years, but I received an error as soon as I added SEN2C as a sensor.

After digging into it, I found it requires a specific combination of four conditions — removing any one of them makes it work. I tested all combinations below on the exact same input images (only three scenes from August 2025, single tile, one image per sensor):

SENSORS	SPECTRAL_ADJUST	SCREEN_QAI includes SUBZERO	Result
SEN2A, SEN2B, SEN2C	TRUE	yes	❌ Crash
SEN2A, SEN2B, SEN2C	TRUE	no	✅ OK
SEN2A, SEN2B, SEN2C	FALSE	yes	✅ OK
SEN2A, SEN2B	TRUE	yes	✅ OK
SEN2B, SEN2C	TRUE	yes	✅ OK

So it only crashes with all three sensors + SPECTRAL_ADJUST = TRUE + SUBZERO in SCREEN_QAI at the same time. Any single change (dropping one sensor, disabling SPECTRAL_ADJUST, or removing SUBZERO) avoids the crash.

I don't get it. Has anyone seen this before, or have an idea what could cause this? Find attached the prm file which leads to a crash.
sentinel_tsa_repex_minimal.zip

Best,
Dominique

[davidfrantz]

I will need to look into this. I elevated this question to an issue:
#439