Fix out-of-bounds read of exitreversaltable in exit1()

[ApolloniaUK]

Hi David,

I’ve recently got back to working on my Mac port with a view to incorporating version 5.2 changes and seriously modernizing the code - basically a complete rewrite moving from old Mac Toolbox to AppKit or SwiftUi.

As part of my preparation I’ve been using Claude Code (Fable 5) to give Andreas’s test harness a very thorough workout so I could be sure my test games were all exactly matching Andreas’s. In the process Fable found a tiny bug in the interpreter with the test version of The Price of Magik. Here's Fable's report:

The Exit opcode reads two game variables, one of which is a direction code, and when the forward search finds no matching exit, exit1() looks up the reverse direction with d6 = exitreversaltable[d6] (level9.c:2935). That table is 20 bytes. The v5.2 changelog notes fixing "an array overrun when handling the commands forwards, backwards, left or right" — that fix extended the table with 0xff sentinel entries for direction codes 16–19. But Price of Magik (during the test script, right after CAST FIN turns you into a fish and a WAIT command) passes a direction code of 20 or more, which still indexes past the end of the table.

In C this reads whatever byte happens to sit after the table in the data segment — undefined behaviour that's deterministic for a given build and, by luck, harmless: the garbage value doesn't match any exit nibble, so the game just reports you can't go that way. Harmless at the moment but a compiler upgrade or layout change could someday turn into a real misbehavior.

I had it patch level9.c and I’ve merged that fix from my dev branch back into master. Here it is in this pull request if you are interested. I haven’t bumped version number or changed copyright date - didn’t know whether you would think either necessary for such a tiny tweak.

HTH.

Alan.
(Alan Staniforth)

[DavidKinder]

Hi Alan, thanks for this. I wasn't entirely happy with the commit message Claude created, and I'm not entirely thrilled at having Claude as a contributor to any of my personal repos, so I changed the comment and committed it as 67fc925.