As currently supplied, using just one VLA in a program requires about 32K of heap. On space limited systems, that's a lot. Somehow change the implementation so that less heap memory is required. See the related forum thread for more of the use case background.

One idea to consider ... In vla_alloc.c there is this preprocessor setting ...

/*
Relatively small VLAs are allocated within blocks of NORMAL_BLOCK_SIZE bytes.
It is usually a good idea to choose this number to be a little smaller than
a power of 2 since many malloc implementations handle those values best.
*/
#define NORMAL_BLOCK_SIZE       32700

Consider making NORMAL_BLOCK_SIZE something that is defined at link time, similar to how the size of the stack and heap are handled.