Resource error cannot be caught

[UWN]

(I tried to celebrate the safe, conforming implementation of integers, but meanwhile ran against memory limits)

| ?- length(_,I),N is 2^2^I, catch(length(L,N),Err,true).

I = 0
L = [_, _]
N = 2 ? ;

I = 1
L = [_, _, _, _]
N = 4 ? ;

I = 2
L = [_, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _]
N = 16 ? ;

I = 3
L = [_, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _, _]
N = 256 ? ;

I = 4
L = _ % omitted
N = 65536 ? ;

Fatal Error: global stack overflow (size: 100000 Kb, reached: 99997 Kb, environment variable used: GLOBALSZ)

Expected:

I = 5
N = 4294967296
Err = error(resource_error(...),...)

The error handling mechanism is designed to catch such errors as well. Strictly speaking, an implementation does not have to handle such cases, but it does help a lot.

[didoudiaz]

I know, but it was a design choice: don't lose time to check stack overflow by software while the hadware (MMU+virtual memory) can do it for free. Unfortunately, it is then very difficult to continue/resume the execution after such hard error detection. Some cases could be (maybe) handled but I never find a stable and general solution working on every plateform.

[UWN]

But on the other hand you permit interrupts which should be similarly difficult to handle...

[UWN]

Just out of curiosity: How do you know that an overflow happened (and not a real bug)? In above case, it seems you do know that the heap overflowed. But if you just rely on the MMU, you cannot be sure, can you?