goddard: read DynList int operands via .ptr to fix 64-bit big-endian crash (#426)

[Scottcjn]

Summary

Fixes #426 — on 64-bit big-endian builds (e.g. ppc64) the game crashes/hangs at the Goddard ("Mario head") intro and never reaches the title screen.

Root cause

DynList integer operands are written through the union's pointer member — the construction macros in dynlist_macros.h cast the int argument to void *:

#define SetColourNum(w2) { 13, {0}, {(void *)(w2)}, {0.0,0.0,0.0} }
#define SetType(w2)      { 19, {0}, {(void *)(w2)}, {0.0,0.0,0.0} }

…but Dyn1AsInt / Dyn2AsInt read them back through the narrower .word member:

#define Dyn2AsInt(dyn) ((dyn)->w2.word)

On little-endian the value's low bytes line up with .word, so it works. On 64-bit big-endian the value sits in the high 4 bytes of the 8-byte pointer slot, so .word reads 0 — a garbage object type that makes d_makeobj fail and the Goddard intro abort.

Fix

Read the operand back the same way it was written — through .ptr — then narrow to s32. This mirrors the sibling Dyn*AsID accessors right next to it, which already read .ptr:

#define Dyn1AsInt(dyn) ((s32)(intptr_t)((dyn)->w1.ptr))
#define Dyn2AsInt(dyn) ((s32)(intptr_t)((dyn)->w2.ptr))

No struct layout change.

Why not just widen word to s64?

That's the tempting one-liner, but it regresses 32-bit big-endian (e.g. 32-bit PowerPC), because there the pointer is 4 bytes and the widened read picks up the wrong half. Worked out across word size × endianness:

DynUnion.word	32-bit BE	64-bit BE
s32 (current)	✅ ok	❌ reads 0 (the bug)
s64	❌ reads 0 (regression)	✅ ok
.ptr reader (this PR)	✅ ok	✅ ok

Reading via .ptr is the only option correct everywhere, and it leaves DynList's layout untouched (so LE and 32-bit builds are entirely unaffected).

Testing

• Cross-built the full game for ppc64 (big-endian) from this branch — compiles clean.
• A standalone reproduction using the repo's actual dynlist_macros.h confirms SetColourNum(5) reads back 5 on both endians with this change, vs 0 on 64-bit BE before it; and that the s64 alternative reads 0 on 32-bit BE.
• 32-bit and 64-bit little-endian behavior and layout are unchanged.

Original diagnosis by @KungFuJesus in #426 — thanks!

---

Note: building on PowerPC also needs a small Makefile tweak (LDFLAGS hardcodes -march=$(TARGET_ARCH), which ppc GCC rejects — it wants -mcpu). I kept that out of this PR to stay focused on the #426 crash; happy to send it separately if useful.

[Scottcjn]

Friendly nudge. This fixes a real 64-bit read of Goddard's DynList operands via .ptr. Small and self-contained. Anything needed from my side to get a review?