My patch #597221. Use f_lasti more consistently.

diff --git a/Python/ceval.c b/Python/ceval.c
index 00237c04d81b3662fcc378f11b029a35190d107a..3494a3854db9e196b240c5a37f883213d61d88b5 100644 (file)
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -599,12 +599,15 @@ eval_frame(PyFrameObject *f)
        fastlocals = f->f_localsplus;
        freevars = f->f_localsplus + f->f_nlocals;
        _PyCode_GETCODEPTR(co, &first_instr);
-       if (f->f_lasti < 0) {
-               next_instr = first_instr;
-       }
-       else {
-               next_instr = first_instr + f->f_lasti;
-       }
+       /* An explanation is in order for the next line.
+
+          f->f_lasti now refers to the index of the last instruction
+          executed.  You might think this was obvious from the name, but
+          this wasn't always true before 2.3!  PyFrame_New now sets
+          f->f_lasti to -1 (i.e. the index *before* the first instruction
+          and YIELD_VALUE doesn't fiddle with f_lasti any more.  So this
+          does work.  Promise. */
+       next_instr = first_instr + f->f_lasti + 1;
        stack_pointer = f->f_stacktop;
        assert(stack_pointer != NULL);
        f->f_stacktop = NULL;   /* remains NULL unless yield suspends frame */
@@ -1521,9 +1524,6 @@ eval_frame(PyFrameObject *f)
                case YIELD_VALUE:
                        retval = POP();
                        f->f_stacktop = stack_pointer;
-                       /* abuse the lasti field: here it points to 
-                          the *next* instruction */
-                       f->f_lasti = INSTR_OFFSET();
                        why = WHY_YIELD;
                        break;