Optimize stpcpy's size and speed

libsrc/common/stpcpy.c

@@ -1,7 +0,0 @@
-#include <string.h>
-
-char * __fastcall__ stpcpy (char * dst, const char * src)
-{
-    strcpy (dst, src);
-    return dst + strlen (src);
-}

libsrc/common/stpcpy.s

@@ -0,0 +1,22 @@
+;
+; Colin Leroy-Mira, 4 Sept. 2024
+;
+; char* stpcpy (char* dest, const char* src);
+;
+
+        .export         _stpcpy
+        .import         _strcpy
+
+        .importzp       tmp1, ptr2
+
+_stpcpy:
+        jsr     _strcpy
+
+        ldx     ptr2+1        ; Load dest pointer's last high byte
+        tya                   ; Get the last offset strcpy wrote to
+
+        clc
+        adc     ptr2          ; Add to low byte value
+        bcc     :+
+        inx
+:       rts                   ; Return pointer to dest's terminator

libsrc/common/strcpy.s

@@ -25,6 +25,9 @@ L1:     lda     (ptr1),y
         inc     ptr2+1
         bne     L1
 
-L9:     lda     ptr2            ; X still contains high byte
-        rts
+L9:     lda     ptr2            ; X still contains dest's original high byte
 
+        ; On exit, we want AX to be dest (as this is what strcpy returns).
+        ; We also want (ptr2),y to still point to dest's terminator, as this
+        ; is used by stpcpy().
+        rts