powerpc aix/darwin build errors in 3.6.0

[Renegade334]

#951 appears to have been half-implemented in powerpc/ffi_darwin.c, and left in an unbuildable state.

• The PPC_LD_* definitions are absent, as these are defined in ffi_powerpc.h, which ffi_darwin.c does not include.
• The return type of ffi_closure_helper_common() has been left as ffi_type*, which is inconsistent with the PPC_LD_* constants that are now returned.

[atgreen]

Thanks for the detailed report — both observations are spot on. #951 added a switch in ffi_closure_helper_common() that returns PPC_LD_* jump-table indexes, but never defined those constants in ffi_darwin.c (they live in ffi_powerpc.h, which this file deliberately does not include — it has its own, differing FLAG_* enum and ffi_dblfl), and left the function signatures returning ffi_type *.

The patch below:

• Defines the PPC_LD_* constants locally in ffi_darwin.c, mirroring ffi_powerpc.h with the same POWERPC64 conditional. The values match the jump tables in aix_closure.S for both the 64-bit path (PPC_LD_U32=10, PPC_LD_S32=11) and the 32-bit path (where U32/S32 collapse to PPC_LD_R3 and I64 to PPC_LD_R3R4).
• Changes the return type of ffi_closure_helper_common(), ffi_closure_helper_DARWIN(), and ffi_go_closure_helper_DARWIN() (and the two forward prototypes) from ffi_type * to int, since the value is an integer index consumed by the assembly (r3 × 16 → jump table), consistent with ffi_closure_helper_SYSV/ffi_closure_helper_LINUX64.

I don't have AIX/Darwin PowerPC hardware to build-test this, so it would be great if you could confirm it builds and runs on your end.

diff --git a/src/powerpc/ffi_darwin.c b/src/powerpc/ffi_darwin.c
index 126367e1..d62737f4 100644
--- a/src/powerpc/ffi_darwin.c
+++ b/src/powerpc/ffi_darwin.c
@@ -39,6 +39,33 @@ struct ffi_aix_trampoline_struct {
     void * static_chain;	/* Pointer to closure */
 };
 
+/* Closure jump table indexes returned by ffi_closure_helper_common and
+   consumed by the jump tables in aix_closure.S.  These mirror the
+   definitions in ffi_powerpc.h, which this file cannot include because
+   it defines its own (differing) FLAG_* and ffi_dblfl.  */
+#define PPC_LD_NONE		0
+#define PPC_LD_R3		1
+#define PPC_LD_R3R4		2
+#define PPC_LD_F32		3
+#define PPC_LD_F64		4
+#define PPC_LD_F128		5
+#define PPC_LD_U8		6
+#define PPC_LD_S8		7
+#define PPC_LD_U16		8
+#define PPC_LD_S16		9
+
+#ifndef POWERPC64
+# define PPC_LD_U32		PPC_LD_R3
+# define PPC_LD_S32		PPC_LD_R3
+# define PPC_LD_PTR		PPC_LD_R3
+# define PPC_LD_I64		PPC_LD_R3R4
+#else
+# define PPC_LD_U32		10
+# define PPC_LD_S32		11
+# define PPC_LD_PTR		PPC_LD_R3
+# define PPC_LD_I64		PPC_LD_R3
+#endif
+
 extern void ffi_closure_ASM (void);
 
 #if defined (FFI_GO_CLOSURES)
@@ -1192,12 +1219,12 @@ typedef union
   double d;
 } ffi_dblfl;
 
-ffi_type *
+int
 ffi_closure_helper_DARWIN (ffi_closure *, void *,
 			   unsigned long *, ffi_dblfl *);
 
 #if defined (FFI_GO_CLOSURES)
-ffi_type *
+int
 ffi_go_closure_helper_DARWIN (ffi_go_closure*, void *,
 			      unsigned long *, ffi_dblfl *);
 #endif
@@ -1209,7 +1236,7 @@ ffi_go_closure_helper_DARWIN (ffi_go_closure*, void *,
    up space for a return value, ffi_closure_ASM invokes the
    following helper function to do most of the work.  */
 
-static ffi_type *
+static int
 ffi_closure_helper_common (ffi_cif* cif,
 			   void (*fun)(ffi_cif*, void*, void**, void*),
 			   void *user_data, void *rvalue,
@@ -1485,7 +1512,7 @@ ffi_closure_helper_common (ffi_cif* cif,
     }
 }
 
-ffi_type *
+int
 ffi_closure_helper_DARWIN (ffi_closure *closure, void *rvalue,
 			   unsigned long *pgr, ffi_dblfl *pfr)
 {
@@ -1494,7 +1521,7 @@ ffi_closure_helper_DARWIN (ffi_closure *closure, void *rvalue,
 }
 
 #if defined (FFI_GO_CLOSURES)
-ffi_type *
+int
 ffi_go_closure_helper_DARWIN (ffi_go_closure *closure, void *rvalue,
 			      unsigned long *pgr, ffi_dblfl *pfr)
 {

[Renegade334]

Many thanks. With this patch, we get successful builds on aix-ppc64 with clang-20.

[barracuda156]

@atgreen I will try this on Darwin.

Upd. Builds fine.

[barracuda156]

Unfortunately, the library does not work correctly now: #1002