bpo-24565: f->f_lineno is now -1 when tracing is not set

[xdegaye]

This is a follow up on PR #6233 that has been closed after becoming from unknown repository.

The Python/ceval.c conflict in PR #6233 has been resolved and master merged into branch [bpo-24565](https://bugs.python.org/issue24565).

https://bugs.python.org/issue24565

[blueyed]

This has a (trivial?) conflict now due to 09532fe.

/cc @vstinner: maybe you can help get this merged? It fixes issues with wrong stack traces, especially when pdb / settrace is being used, and I've hoped for it to get into 3.7 already - let's not miss it for 3.8 also, please.

[blueyed]

@xdegaye
Please fix conflicts again (I've done so manually for py38):

diff --cc Python/ceval.c
index 9f46b56fdd,3306fb9728..853b8b9881
--- a/Python/ceval.c
+++ b/Python/ceval.c
@@@ -4467,8 -4596,7 +4597,8 @@@ call_trace_protected(Py_tracefunc func
  {
      PyObject *type, *value, *traceback;
      int err;
 +    frame->f_lineno = PyFrame_GetLineNumber(frame);
-     PyErr_Fetch(&type, &value, &traceback);
+     _PyErr_Fetch(tstate, &type, &value, &traceback);
      err = call_trace(func, obj, tstate, frame, what, arg);
      if (err == 0)
      {
@@@ -4574,7 -4706,10 +4708,11 @@@ PyEval_SetProfile(Py_tracefunc func, Py
  void
  PyEval_SetTrace(Py_tracefunc func, PyObject *arg)
  {
+     if (PySys_Audit("sys.settrace", NULL) < 0) {
+         return;
+     }
+
 +    PyFrameObject *f;
      _PyRuntimeState *runtime = &_PyRuntime;
      PyThreadState *tstate = _PyRuntimeState_GetThreadState(runtime);
      PyObject *temp = tstate->c_traceobj;

[blueyed]

@xdegaye
Are you giving up? (which I could understand, but would be a pity really)

[blueyed]

Related: https://www.python.org/dev/peps/pep-0626/
Implemented for Python 3.10 (in 877df85).