In configure.ac, if the user passes --tail-call-interp in the negative, we set Py_TAIL_CALL_INTERP to 0:
|
if test "$withval" = no |
|
then |
|
AC_DEFINE([Py_TAIL_CALL_INTERP], [0], |
|
[Define if you want to use tail-calling interpreters in CPython.]) |
|
AC_MSG_RESULT([no]) |
|
fi |
However, in ceval_macros.h and bytecodes.c, we check for #ifdef Py_TAIL_CALL_INTERP and #if defined(Py_TAIL_CALL_INTERP), for instance:
|
#ifdef Py_TAIL_CALL_INTERP |
)
Thus, defining the macro -- even with a value of 0 -- actually enables the feature! I think the consistent behavior (consistent with USE_COMPUTED_GOTOS) is probably to switch the preprocessor guards to #if, instead of #ifdef.
Linked PRs
In
configure.ac, if the user passes--tail-call-interpin the negative, we setPy_TAIL_CALL_INTERPto0:cpython/configure.ac
Lines 7027 to 7032 in 1feaecc
However, in
ceval_macros.handbytecodes.c, we check for#ifdef Py_TAIL_CALL_INTERPand#if defined(Py_TAIL_CALL_INTERP), for instance:cpython/Python/ceval_macros.h
Line 76 in 1feaecc
Thus, defining the macro -- even with a value of
0-- actually enables the feature! I think the consistent behavior (consistent withUSE_COMPUTED_GOTOS) is probably to switch the preprocessor guards to#if, instead of#ifdef.Linked PRs