GENERAL_GET_RANDOM_INT

[oisee]

A uniformly distributed integer between zero and RANGE, both included.

The bounds are measured, not inferred

The parameter documentation says "The random int will be <= range" and says nothing about the lower bound. The obvious reading — one to range — is wrong. Three hundred calls per range on an AS ABAP 1909 sandbox:

range	values seen	zero in
1	0..1	163 of 300 (a half)
2	0..2	105 of 300 (a third)
6	0..6	53 of 300 (a seventh)
100	0..100	3 of 300 (a hundred-and-first)

So the span is RANGE + 1 values, and the zero frequency matches 1/(range+1) at every point. Edges, same system: range = 0 answered 0; range = -5 answered -2, so a negative range is not an error and the span runs towards RANGE on whichever side of zero it lies; 2000 draws at range = 1 gave 1002 zeroes. The module declares no exceptions and the real one raises none.

The numbers come from calling the module, from a throwaway report and test class in $TMP, not from reading its implementation — an observation can be contributed, an adaptation cannot.

Why not cl_abap_random_int

That was the first attempt and it does not fit. cl_abap_random=>intinrange opens with ASSERT high > low and ASSERT low >= 0, so a range of zero dumps and every negative range dumps. Reaching the measured contract through abs( ) and a negation works and reads as cleverness hiding the intent.

It is the same Math.random either way — that is what cl_abap_random uses — and generate_sec_random in this same function group is the precedent for the plain @KERNEL form.

Tests

Four, in kernel_fugr_test: the zero range, the bounds at 6 and at -5, and that both 0 and 1 are reachable at range 1. The last is the one that tells this apart from a generator over 1..RANGE; with the naive implementation in place it fails with Expected '0', got '1', which is how it was checked.

Where it came from

An ABAP Z-machine interpreter. MiniZork plays twenty-five commands and then dies on the troll, because the random opcode calls this module and gets CX_SY_DYN_CALL_ILLEGAL_FUNC. With it the whole walkthrough plays.

Worth noting for anyone porting a Z-machine: the Z standard wants 1..range, and this module gives 0..range, so the caller has to map. That is the caller's job and not this module's — matching the real system is.

🤖 Generated with Claude Code

https://claude.ai/code/session_01J7JpA3TGT48zBoix3Ut88s

[larshp]

hmm, might also need to go into deprecated?