diff --git a/Makefile.pre.in b/Makefile.pre.in
index 4071a2a..e729ffb 100644
--- a/Makefile.pre.in
+++ b/Makefile.pre.in
@@ -879,7 +879,9 @@ Objects/setobject.o: $(srcdir)/Objects/stringlib/eq.h
 $(OPCODETARGETS_H): $(OPCODETARGETGEN_FILES)
 	$(OPCODETARGETGEN) $(OPCODETARGETS_H)
 
-Python/ceval.o: $(OPCODETARGETS_H) $(srcdir)/Python/ceval_gil.h
+Python/ceval.o: $(OPCODETARGETS_H) $(srcdir)/Python/ceval_gil.h $(srcdir)/Python/wordcode_helpers.h
+Python/compile.o: $(srcdir)/Python/wordcode_helpers.h
+Python/peephole.o: $(srcdir)/Python/wordcode_helpers.h
 
 Python/frozen.o: Python/importlib.h Python/importlib_external.h
 
diff --git a/Python/ceval.c b/Python/ceval.c
index 20e93cc..3096ad9 100644
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@ -17,6 +17,7 @@
 #include "opcode.h"
 #include "setobject.h"
 #include "structmember.h"
+#include "wordcode_helpers.h"
 
 #include <ctype.h>
 
@@ -136,7 +137,8 @@ static int call_trace_protected(Py_tracefunc, PyObject *,
 static void call_exc_trace(Py_tracefunc, PyObject *,
                            PyThreadState *, PyFrameObject *);
 static int maybe_call_line_trace(Py_tracefunc, PyObject *,
-                                 PyThreadState *, PyFrameObject *, int *, int *, int *);
+                                 PyThreadState *, PyFrameObject *,
+                                 int *, int *, int *, int, int);
 
 static PyObject * cmp_outcome(int, PyObject *, PyObject *);
 static PyObject * import_from(PyObject *, PyObject *);
@@ -1307,14 +1309,24 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
         if (_Py_TracingPossible &&
             tstate->c_tracefunc != NULL && !tstate->tracing) {
             int err;
+            int lasti;
+            if (OPCODE(*next_instr) == FOR_ITER) {
+                /* Trace FOR_ITER as if we're the corresponding FOR_BEGIN,
+                   as tracer expects loop logic to precede loop body */
+                lasti = (int)get_arg((unsigned char*)next_instr, 0) - 2;
+            }
+            else {
+                lasti = f->f_lasti;
+            }
             /* see maybe_call_line_trace
                for expository comments */
             f->f_stacktop = stack_pointer;
 
             err = maybe_call_line_trace(tstate->c_tracefunc,
-                                        tstate->c_traceobj,
-                                        tstate, f,
-                                        &instr_lb, &instr_ub, &instr_prev);
+                                        tstate->c_traceobj, tstate, f,
+                                        &instr_lb, &instr_ub, &instr_prev,
+                                        f->f_lasti, lasti);
+
             /* Reload possibly changed frame fields */
             JUMPTO(f->f_lasti);
             if (f->f_stacktop != NULL) {
@@ -1670,7 +1682,7 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
             Py_DECREF(v);
             if (err != 0)
                 goto error;
-            PREDICT(JUMP_ABSOLUTE);
+            PREDICT(FOR_ITER);
             DISPATCH();
         }
 
@@ -1682,7 +1694,7 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
             Py_DECREF(v);
             if (err != 0)
                 goto error;
-            PREDICT(JUMP_ABSOLUTE);
+            PREDICT(FOR_ITER);
             DISPATCH();
         }
 
@@ -2704,7 +2716,7 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
             Py_DECREF(key);
             if (err != 0)
                 goto error;
-            PREDICT(JUMP_ABSOLUTE);
+            PREDICT(FOR_ITER);
             DISPATCH();
         }
 
@@ -2920,7 +2932,6 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
             DISPATCH();
         }
 
-        PREDICTED(JUMP_ABSOLUTE);
         TARGET(JUMP_ABSOLUTE) {
             JUMPTO(oparg);
 #if FAST_LOOPS
@@ -2999,6 +3010,7 @@ PyEval_EvalFrameEx(PyFrameObject *f, int throwflag)
             DISPATCH();
         }
 
+        PREDICTED(FOR_ITER)
         TARGET(FOR_ITER) {
             /* before: [iter]; after: [iter, iter()] *or* [] */
             PyObject *iter = TOP();
@@ -4445,7 +4457,7 @@ _PyEval_CallTracing(PyObject *func, PyObject *args)
 static int
 maybe_call_line_trace(Py_tracefunc func, PyObject *obj,
                       PyThreadState *tstate, PyFrameObject *frame,
-                      int *instr_lb, int *instr_ub, int *instr_prev)
+                      int *instr_lb, int *instr_ub, int *instr_prev, int lasti, int nlasti)
 {
     int result = 0;
     int line = frame->f_lineno;
@@ -4453,9 +4465,10 @@ maybe_call_line_trace(Py_tracefunc func, PyObject *obj,
     /* If the last instruction executed isn't in the current
        instruction window, reset the window.
     */
-    if (frame->f_lasti < *instr_lb || frame->f_lasti >= *instr_ub) {
+redo:
+    if (lasti < *instr_lb || lasti >= *instr_ub) {
         PyAddrPair bounds;
-        line = _PyCode_CheckLineNumber(frame->f_code, frame->f_lasti,
+        line = _PyCode_CheckLineNumber(frame->f_code, lasti,
                                        &bounds);
         *instr_lb = bounds.ap_lower;
         *instr_ub = bounds.ap_upper;
@@ -4463,11 +4476,22 @@ maybe_call_line_trace(Py_tracefunc func, PyObject *obj,
     /* If the last instruction falls at the start of a line or if
        it represents a jump backwards, update the frame's line
        number and call the trace function. */
-    if (frame->f_lasti == *instr_lb || frame->f_lasti < *instr_prev) {
+    if (lasti == *instr_lb || lasti < *instr_prev) {
         frame->f_lineno = line;
         result = call_trace(func, obj, tstate, frame, PyTrace_LINE, Py_None);
     }
-    *instr_prev = frame->f_lasti;
+
+    if (!result && lasti != nlasti) {
+        /* FOR_ITER may generate 2 line traces if the body is 'pass'
+           We want to generate a trace as if the end of the loop is a mere
+           JUMP_ABSOLUTE so we manually trace FOR_ITER as if FOR_BEGIN is
+           being executed each loop in lieu of there being no better instruction
+           at the beginning to designate loop iteration logic */
+        lasti = nlasti;
+        goto redo;
+    }
+
+    *instr_prev = lasti;
     return result;
 }
 
diff --git a/Python/peephole.c b/Python/peephole.c
index 5795c62..94f5f47 100644
--- a/Python/peephole.c
+++ b/Python/peephole.c
@@ -104,25 +104,6 @@ find_op(unsigned char *codestr, Py_ssize_t i)
 }
 
 /* Given the index of the effective opcode,
-   scan back to construct the oparg with EXTENDED_ARG */
-static unsigned int
-get_arg(unsigned char *codestr, Py_ssize_t i)
-{
-    unsigned int oparg = codestr[i+1];
-    assert((i&1) == 0);
-    if (i >= 2 && codestr[i-2] == EXTENDED_ARG) {
-        oparg |= codestr[i-1] << 8;
-        if (i >= 4 && codestr[i-4] == EXTENDED_ARG) {
-            oparg |= codestr[i-3] << 16;
-            if (i >= 6 && codestr[i-6] == EXTENDED_ARG) {
-                oparg |= codestr[i-5] << 24;
-            }
-        }
-    }
-    return oparg;
-}
-
-/* Given the index of the effective opcode,
    attempt to replace the argument, taking into account EXTENDED_ARG.
    Returns -1 on failure, or the new op index on success */
 static Py_ssize_t
@@ -539,7 +520,7 @@ PyCode_Optimize(PyObject *code, PyObject* consts, PyObject *names,
                          ((opcode == BUILD_LIST || opcode == BUILD_SET) &&
                           ((nextop==COMPARE_OP &&
                           (codestr[nexti+1]==6 ||
-                           codestr[nexti+1]==7))
+                           codestr[nexti+1]==7)) ||
                           nextop == FOR_BEGIN) && ISBASICBLOCK(blocks, h, i + 2))) {
                         h = fold_tuple_on_constants(codestr, h, i+2, opcode,
                                                     consts, CONST_STACK_LASTN(j), j);
diff --git a/Python/wordcode_helpers.h b/Python/wordcode_helpers.h
index b61ba33..f6126e5 100644
--- a/Python/wordcode_helpers.h
+++ b/Python/wordcode_helpers.h
@@ -3,7 +3,7 @@
  */
 
 /* Minimum number of bytes necessary to encode instruction with EXTENDED_ARGs */
-static int
+Py_LOCAL_INLINE(int)
 instrsize(unsigned int oparg)
 {
     return oparg <= 0xff ? 2 :
@@ -14,7 +14,7 @@ instrsize(unsigned int oparg)
 
 /* Spits out op/oparg pair using ilen bytes. codestr should be pointed at the
    desired location of the first EXTENDED_ARG */
-static void
+Py_LOCAL_INLINE(void)
 write_op_arg(unsigned char *codestr, unsigned char opcode,
     unsigned int oparg, int ilen)
 {
@@ -36,3 +36,23 @@ write_op_arg(unsigned char *codestr, unsigned char opcode,
             assert(0);
     }
 }
+
+/* Given the index of the effective opcode,
+   scan back to construct the oparg with EXTENDED_ARG */
+Py_LOCAL_INLINE(unsigned int)
+get_arg(unsigned char *codestr, Py_ssize_t i)
+{
+    unsigned int oparg = codestr[i+1];
+    assert((i&1) == 0);
+    if (i >= 2 && codestr[i-2] == EXTENDED_ARG) {
+        oparg |= codestr[i-1] << 8;
+        if (i >= 4 && codestr[i-4] == EXTENDED_ARG) {
+            oparg |= codestr[i-3] << 16;
+            if (i >= 6 && codestr[i-6] == EXTENDED_ARG) {
+                oparg |= codestr[i-5] << 24;
+            }
+        }
+    }
+    return oparg;
+}
+