There's a huge difference in stack vs. heap allocations. Stack space might be effectively infinite in the basic user-space world, but not in any environment using green threads and certainly not in the kernel. In the Linux kernel this is made far worse by the fact that each task's stack sits right above the task structure itself, so a huge allocation might smash the task structure or the next task's stack. The latter is exactly what I had to debug a couple of jobs ago, and it was one of the more difficult debugging journeys of my long career (it was quite sporadic and manifested as hard hangs). VLAs make that kind of scenario much more likely. Being able to check the length before a potentially disastrous allocation is cumbersome, but safer.

There's no way to return an error from a VLA allocation, and also within C you have no standard way of knowing how much is safe to allocate on the stack (and in fact it may not be possible to know).