Fix reused integers from being optimized out

[justwheel]

Summary

Fix segfault when reused integers are optimized out by compiler

Background

This patch was contributed by @Jan200101 to fix a reported segfault in the Fedora package from compiler optimizations. We can carry a custom patch for the Fedora-specific package, but we generally try to work with an upstream so everyone may benefit from a change, not only Fedora.

This change was tested with a local scratch build on at least two Fedora machines. It is likely other distributions could face similar issues when upgrading their compilers.

Details

It is a simple change, and honestly I am not a C developer. If you really want an explanation, @Jan200101 could probably provide one if you ask.

If this change is accepted, it would be helpful to cut a new release with a git tag. This will trigger the first step in the Fedora package distribution process for shipping an update. 🙂

Outcome

Support optimizations from newer versions of compilers without segfaulting after opening the mailbox

[Jan200101]

The problem was with how gcc optimized the program
(gcc versions before 10.2.1 might not have this problem)
compiling zork with -O2, as one does for Fedora packages, made the compiler try to optimize i__2 out
(gdb output copied from the previously mentioned report)

#0  sparse_ (lbuf=0x7fffffffd54c, lbuf@entry=0x7fffffffd550, llnt=3, vbflag=vbflag@entry=1) at np1.c:128
        ret_val = -1
        i__1 = 3
        i__2 = <optimized out>
        i = 3
        j = 44617
        obj = <optimized out>
        prep = 0
        lbuf1 = 3258
        lbuf2 = 0

This being the code that triggers this sigsegv

zork/np1.c

Lines 126 to 131 in 95b1fd1

	L200:
	i__2 = prplnt;
	for (j = 1; j <= i__2; j += 3) {
	/* !LOOK FOR PREPOSITION. */
	if (lbuf1 == prpvoc_1.pvoc[j - 1] && lbuf2 == prpvoc_1.pvoc[j]) {
	goto L4000;

if we look at prplnt

zork/np1.c

Line 58 in 95b1fd1

prplnt = 48;

we can tell something went horribly wrong in that gdb backtrace since j is far larger than prplnt
the way to solve this is to tell the compiler not to optimize i__2 out, and in the case of my patch I made all system generated locals that are reused over the file volatile

[justwheel]

Thanks for clarifying the change @Jan200101.

@devshane Do you have a moment to review this patch? It would be great to add this here in the upstream project instead of maintaining a downstream patch only available in the Fedora package.

[devshane]

Thanks for the fix!

https://github.com/devshane/zork/releases/tag/v1.0.3