bpo-39794: Add --without-decimal-contextvar (#18702)

11 files changed, 248 insertions, 36 deletions

diff --git a/Doc/library/decimal.rst b/Doc/library/decimal.rst
index 4e640cc..9e31781 100644
--- a/Doc/library/decimal.rst
+++ b/Doc/library/decimal.rst
@@ -1475,9 +1475,18 @@ are also included in the pure Python version for compatibility.
 
 .. data:: HAVE_THREADS
 
-   The default value is ``True``. If Python is compiled without threads, the
-   C version automatically disables the expensive thread local context
-   machinery. In this case, the value is ``False``.
+   The value is ``True``.  Deprecated, because Python now always has threads.
+
+.. deprecated:: 3.9
+
+.. data:: HAVE_CONTEXTVAR
+
+   The default value is ``True``. If Python is compiled ``--without-decimal-contextvar``,
+   the C version uses a thread-local rather than a coroutine-local context and the value
+   is ``False``.  This is slightly faster in some nested context scenarios.
+
+.. versionadded:: 3.9
+
 
 Rounding modes
 --------------
diff --git a/Lib/_pydecimal.py b/Lib/_pydecimal.py
index c14d8ca..ab989e5 100644
--- a/Lib/_pydecimal.py
+++ b/Lib/_pydecimal.py
@@ -140,8 +140,11 @@ __all__ = [
     # Limits for the C version for compatibility
     'MAX_PREC',  'MAX_EMAX', 'MIN_EMIN', 'MIN_ETINY',
 
-    # C version: compile time choice that enables the thread local context
-    'HAVE_THREADS'
+    # C version: compile time choice that enables the thread local context (deprecated, now always true)
+    'HAVE_THREADS',
+
+    # C version: compile time choice that enables the coroutine local context
+    'HAVE_CONTEXTVAR'
 ]
 
 __xname__ = __name__    # sys.modules lookup (--without-threads)
@@ -172,6 +175,7 @@ ROUND_05UP = 'ROUND_05UP'
 
 # Compatibility with the C version
 HAVE_THREADS = True
+HAVE_CONTEXTVAR = True
 if sys.maxsize == 2**63-1:
     MAX_PREC = 999999999999999999
     MAX_EMAX = 999999999999999999
diff --git a/Lib/test/test_asyncio/test_context.py b/Lib/test/test_asyncio/test_context.py
index c309faa..63b1eb3 100644
--- a/Lib/test/test_asyncio/test_context.py
+++ b/Lib/test/test_asyncio/test_context.py
@@ -7,6 +7,7 @@ def tearDownModule():
     asyncio.set_event_loop_policy(None)
 
 
+@unittest.skipUnless(decimal.HAVE_CONTEXTVAR, "decimal is built with a thread-local context")
 class DecimalContextTest(unittest.TestCase):
 
     def test_asyncio_task_decimal_context(self):
diff --git a/Misc/NEWS.d/next/Library/2020-02-29-19-17-39.bpo-39794.7VjatS.rst b/Misc/NEWS.d/next/Library/2020-02-29-19-17-39.bpo-39794.7VjatS.rst
new file mode 100644
index 0000000..b2a4726
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2020-02-29-19-17-39.bpo-39794.7VjatS.rst
@@ -0,0 +1,2 @@
+Add --without-decimal-contextvar build option.  This enables a thread-local
+rather than a coroutine local context.
diff --git a/Modules/_decimal/_decimal.c b/Modules/_decimal/_decimal.c
index e2f6ea1..617941b 100644
--- a/Modules/_decimal/_decimal.c
+++ b/Modules/_decimal/_decimal.c
@@ -122,7 +122,14 @@ incr_false(void)
 }
 
 
+#ifndef WITH_DECIMAL_CONTEXTVAR
+/* Key for thread state dictionary */
+static PyObject *tls_context_key = NULL;
+/* Invariant: NULL or the most recently accessed thread local context */
+static PyDecContextObject *cached_context = NULL;
+#else
 static PyObject *current_context_var;
+#endif
 
 /* Template for creating new thread contexts, calling Context() without
  * arguments and initializing the module_context on first access. */
@@ -1217,6 +1224,12 @@ context_new(PyTypeObject *type, PyObject *args UNUSED, PyObject *kwds UNUSED)
 static void
 context_dealloc(PyDecContextObject *self)
 {
+#ifndef WITH_DECIMAL_CONTEXTVAR
+    if (self == cached_context) {
+        cached_context = NULL;
+    }
+#endif
+
     Py_XDECREF(self->traps);
     Py_XDECREF(self->flags);
     Py_TYPE(self)->tp_free(self);
@@ -1491,6 +1504,134 @@ static PyGetSetDef context_getsets [] =
  * operation.
  */
 
+#ifndef WITH_DECIMAL_CONTEXTVAR
+/* Get the context from the thread state dictionary. */
+static PyObject *
+current_context_from_dict(void)
+{
+    PyObject *dict;
+    PyObject *tl_context;
+    PyThreadState *tstate;
+
+    dict = PyThreadState_GetDict();
+    if (dict == NULL) {
+        PyErr_SetString(PyExc_RuntimeError,
+            "cannot get thread state");
+        return NULL;
+    }
+
+    tl_context = PyDict_GetItemWithError(dict, tls_context_key);
+    if (tl_context != NULL) {
+        /* We already have a thread local context. */
+        CONTEXT_CHECK(tl_context);
+    }
+    else {
+        if (PyErr_Occurred()) {
+            return NULL;
+        }
+
+        /* Set up a new thread local context. */
+        tl_context = context_copy(default_context_template, NULL);
+        if (tl_context == NULL) {
+            return NULL;
+        }
+        CTX(tl_context)->status = 0;
+
+        if (PyDict_SetItem(dict, tls_context_key, tl_context) < 0) {
+            Py_DECREF(tl_context);
+            return NULL;
+        }
+        Py_DECREF(tl_context);
+    }
+
+    /* Cache the context of the current thread, assuming that it
+     * will be accessed several times before a thread switch. */
+    tstate = PyThreadState_GET();
+    if (tstate) {
+        cached_context = (PyDecContextObject *)tl_context;
+        cached_context->tstate = tstate;
+    }
+
+    /* Borrowed reference with refcount==1 */
+    return tl_context;
+}
+
+/* Return borrowed reference to thread local context. */
+static PyObject *
+current_context(void)
+{
+    PyThreadState *tstate;
+
+    tstate = PyThreadState_GET();
+    if (cached_context && cached_context->tstate == tstate) {
+        return (PyObject *)cached_context;
+    }
+
+    return current_context_from_dict();
+}
+
+/* ctxobj := borrowed reference to the current context */
+#define CURRENT_CONTEXT(ctxobj) \
+    ctxobj = current_context(); \
+    if (ctxobj == NULL) {       \
+        return NULL;            \
+    }
+
+/* Return a new reference to the current context */
+static PyObject *
+PyDec_GetCurrentContext(PyObject *self UNUSED, PyObject *args UNUSED)
+{
+    PyObject *context;
+
+    context = current_context();
+    if (context == NULL) {
+        return NULL;
+    }
+
+    Py_INCREF(context);
+    return context;
+}
+
+/* Set the thread local context to a new context, decrement old reference */
+static PyObject *
+PyDec_SetCurrentContext(PyObject *self UNUSED, PyObject *v)
+{
+    PyObject *dict;
+
+    CONTEXT_CHECK(v);
+
+    dict = PyThreadState_GetDict();
+    if (dict == NULL) {
+        PyErr_SetString(PyExc_RuntimeError,
+            "cannot get thread state");
+        return NULL;
+    }
+
+    /* If the new context is one of the templates, make a copy.
+     * This is the current behavior of decimal.py. */
+    if (v == default_context_template ||
+        v == basic_context_template ||
+        v == extended_context_template) {
+        v = context_copy(v, NULL);
+        if (v == NULL) {
+            return NULL;
+        }
+        CTX(v)->status = 0;
+    }
+    else {
+        Py_INCREF(v);
+    }
+
+    cached_context = NULL;
+    if (PyDict_SetItem(dict, tls_context_key, v) < 0) {
+        Py_DECREF(v);
+        return NULL;
+    }
+
+    Py_DECREF(v);
+    Py_RETURN_NONE;
+}
+#else
 static PyObject *
 init_current_context(void)
 {
@@ -1570,6 +1711,7 @@ PyDec_SetCurrentContext(PyObject *self UNUSED, PyObject *v)
 
     Py_RETURN_NONE;
 }
+#endif
 
 /* Context manager object for the 'with' statement. The manager
  * owns one reference to the global (outer) context and one
@@ -4388,15 +4530,8 @@ _dec_hash(PyDecObject *v)
     mpd_ssize_t exp;
     uint32_t status = 0;
     mpd_context_t maxctx;
-    PyObject *context;
 
 
-    context = current_context();
-    if (context == NULL) {
-        return -1;
-    }
-    Py_DECREF(context);
-
     if (mpd_isspecial(MPD(v))) {
         if (mpd_issnan(MPD(v))) {
             PyErr_SetString(PyExc_TypeError,
@@ -5538,11 +5673,6 @@ PyInit__decimal(void)
     mpd_free = PyMem_Free;
     mpd_setminalloc(_Py_DEC_MINALLOC);
 
-    /* Init context variable */
-    current_context_var = PyContextVar_New("decimal_context", NULL);
-    if (current_context_var == NULL) {
-        goto error;
-    }
 
     /* Init external C-API functions */
     _py_long_multiply = PyLong_Type.tp_as_number->nb_multiply;
@@ -5714,6 +5844,15 @@ PyInit__decimal(void)
     CHECK_INT(PyModule_AddObject(m, "DefaultContext",
                                  default_context_template));
 
+#ifndef WITH_DECIMAL_CONTEXTVAR
+    ASSIGN_PTR(tls_context_key, PyUnicode_FromString("___DECIMAL_CTX__"));
+    Py_INCREF(Py_False);
+    CHECK_INT(PyModule_AddObject(m, "HAVE_CONTEXTVAR", Py_False));
+#else
+    ASSIGN_PTR(current_context_var, PyContextVar_New("decimal_context", NULL));
+    Py_INCREF(Py_True);
+    CHECK_INT(PyModule_AddObject(m, "HAVE_CONTEXTVAR", Py_True));
+#endif
     Py_INCREF(Py_True);
     CHECK_INT(PyModule_AddObject(m, "HAVE_THREADS", Py_True));
 
@@ -5773,9 +5912,13 @@ error:
     Py_CLEAR(SignalTuple); /* GCOV_NOT_REACHED */
     Py_CLEAR(DecimalTuple); /* GCOV_NOT_REACHED */
     Py_CLEAR(default_context_template); /* GCOV_NOT_REACHED */
+#ifndef WITH_DECIMAL_CONTEXTVAR
+    Py_CLEAR(tls_context_key); /* GCOV_NOT_REACHED */
+#else
+    Py_CLEAR(current_context_var); /* GCOV_NOT_REACHED */
+#endif
     Py_CLEAR(basic_context_template); /* GCOV_NOT_REACHED */
     Py_CLEAR(extended_context_template); /* GCOV_NOT_REACHED */
-    Py_CLEAR(current_context_var); /* GCOV_NOT_REACHED */
     Py_CLEAR(m); /* GCOV_NOT_REACHED */
 
     return NULL; /* GCOV_NOT_REACHED */
diff --git a/Modules/_decimal/tests/runall-memorydebugger.sh b/Modules/_decimal/tests/runall-memorydebugger.sh
index 29b7723..7f3e527 100755
--- a/Modules/_decimal/tests/runall-memorydebugger.sh
+++ b/Modules/_decimal/tests/runall-memorydebugger.sh
@@ -1,8 +1,8 @@
 #!/bin/sh
 
 #
-# Purpose: test all machine configurations, pydebug, refleaks, release build
-#          and release build with valgrind.
+# Purpose: test with and without contextvar, all machine configurations, pydebug,
+#          refleaks, release build and release build with valgrind.
 #
 # Synopsis: ./runall-memorydebugger.sh [--all-configs64 | --all-configs32]
 #
@@ -57,7 +57,8 @@ print_config ()
 cd ..
 
 # test_decimal: refleak, regular and Valgrind tests
-for config in $CONFIGS; do
+for args in "--without-decimal-contextvar" ""; do
+    for config in $CONFIGS; do
 
         unset PYTHON_DECIMAL_WITH_MACHINE
         libmpdec_config=$config
@@ -69,12 +70,12 @@ for config in $CONFIGS; do
         fi
 
         ############ refleak tests ###########
-        print_config "refleak tests: config=$config"
+        print_config "refleak tests: config=$config" $args
         printf "\nbuilding python ...\n\n"
 
         cd ../../
         $GMAKE distclean > /dev/null 2>&1
-        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --with-pydebug > /dev/null 2>&1
+        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --with-pydebug $args > /dev/null 2>&1
         $GMAKE | grep _decimal
 
         printf "\n\n# ======================== refleak tests ===========================\n\n"
@@ -82,11 +83,11 @@ for config in $CONFIGS; do
 
 
         ############ regular tests ###########
-        print_config "regular tests: config=$config"
+        print_config "regular tests: config=$config" $args
         printf "\nbuilding python ...\n\n"
 
         $GMAKE distclean > /dev/null 2>&1
-        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" > /dev/null 2>&1
+        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" $args > /dev/null 2>&1
         $GMAKE | grep _decimal
 
         printf "\n\n# ======================== regular tests ===========================\n\n"
@@ -103,21 +104,23 @@ for config in $CONFIGS; do
                   esac
         esac
 
-        print_config "valgrind tests: config=$config"
+        print_config "valgrind tests: config=$config" $args
         printf "\nbuilding python ...\n\n"
         $GMAKE distclean > /dev/null 2>&1
-        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --without-pymalloc > /dev/null 2>&1
+        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --without-pymalloc $args > /dev/null 2>&1
         $GMAKE | grep _decimal
 
         printf "\n\n# ======================== valgrind tests ===========================\n\n"
         $valgrind ./python -m test -uall test_decimal
 
         cd Modules/_decimal
+    done
 done
 
 # deccheck
 cd ../../
-for config in $CONFIGS; do
+for args in "--without-decimal-contextvar" ""; do
+    for config in $CONFIGS; do
 
         unset PYTHON_DECIMAL_WITH_MACHINE
         if [ X"$config" != X"auto" ]; then
@@ -126,22 +129,22 @@ for config in $CONFIGS; do
         fi
 
         ############ debug ############
-        print_config "deccheck: config=$config --with-pydebug"
+        print_config "deccheck: config=$config --with-pydebug" $args
         printf "\nbuilding python ...\n\n"
 
         $GMAKE distclean > /dev/null 2>&1
-        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --with-pydebug > /dev/null 2>&1
+        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --with-pydebug $args > /dev/null 2>&1
         $GMAKE | grep _decimal
 
         printf "\n\n# ========================== debug ===========================\n\n"
         ./python Modules/_decimal/tests/deccheck.py
 
         ########### regular ###########
-        print_config "deccheck: config=$config "
+        print_config "deccheck: config=$config" $args
         printf "\nbuilding python ...\n\n"
 
         $GMAKE distclean > /dev/null 2>&1
-        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" > /dev/null 2>&1
+        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" $args > /dev/null 2>&1
         $GMAKE | grep _decimal
 
         printf "\n\n# ======================== regular ===========================\n\n"
@@ -157,15 +160,16 @@ for config in $CONFIGS; do
                   esac
         esac
 
-        print_config "valgrind deccheck: config=$config "
+        print_config "valgrind deccheck: config=$config" $args
         printf "\nbuilding python ...\n\n"
 
         $GMAKE distclean > /dev/null 2>&1
-        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --without-pymalloc > /dev/null 2>&1
+        ./configure CFLAGS="$ADD_CFLAGS" LDFLAGS="$ADD_LDFLAGS" --without-pymalloc $args > /dev/null 2>&1
         $GMAKE | grep _decimal
 
         printf "\n\n# ======================== valgrind ==========================\n\n"
         $valgrind ./python Modules/_decimal/tests/deccheck.py
+    done
 done
 
 
diff --git a/PC/pyconfig.h b/PC/pyconfig.h
index 424c5be..0aa4f95 100644
--- a/PC/pyconfig.h
+++ b/PC/pyconfig.h
@@ -465,6 +465,10 @@ Py_NO_ENABLE_SHARED to find out.  Also support MS_NO_COREDLL for b/w compat */
    (which you can't on SCO ODT 3.0). */
 /* #undef SYS_SELECT_WITH_SYS_TIME */
 
+/* Define if you want build the _decimal module using a coroutine-local rather
+   than a thread-local context */
+#define WITH_DECIMAL_CONTEXTVAR 1
+
 /* Define if you want documentation strings in extension modules */
 #define WITH_DOC_STRINGS 1
 
diff --git a/aclocal.m4 b/aclocal.m4
index 85f00dd..f98db73 100644
--- a/aclocal.m4
+++ b/aclocal.m4
@@ -1,6 +1,6 @@
-# generated automatically by aclocal 1.16.1 -*- Autoconf -*-
+# generated automatically by aclocal 1.15 -*- Autoconf -*-
 
-# Copyright (C) 1996-2018 Free Software Foundation, Inc.
+# Copyright (C) 1996-2014 Free Software Foundation, Inc.
 
 # This file is free software; the Free Software Foundation
 # gives unlimited permission to copy and/or distribute it,
diff --git a/configure b/configure
index 846116e..0bffe8b 100755
--- a/configure
+++ b/configure
@@ -826,6 +826,7 @@ with_libs
 with_system_expat
 with_system_ffi
 with_system_libmpdec
+with_decimal_contextvar
 enable_loadable_sqlite_extensions
 with_tcltk_includes
 with_tcltk_libs
@@ -1548,6 +1549,9 @@ Optional Packages:
                           system-dependent)
   --with-system-libmpdec  build _decimal module using an installed libmpdec
                           library, see Doc/library/decimal.rst (default is no)
+  --with-decimal-contextvar
+                          build _decimal module using a coroutine-local rather
+                          than a thread-local context (default is yes)
   --with-tcltk-includes='-I...'
                           override search for Tcl and Tk include files
   --with-tcltk-libs='-L...'
@@ -10496,6 +10500,28 @@ fi
 { $as_echo "$as_me:${as_lineno-$LINENO}: result: $with_system_libmpdec" >&5
 $as_echo "$with_system_libmpdec" >&6; }
 
+# Check whether _decimal should use a coroutine-local or thread-local context
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for --with-decimal-contextvar" >&5
+$as_echo_n "checking for --with-decimal-contextvar... " >&6; }
+
+# Check whether --with-decimal_contextvar was given.
+if test "${with_decimal_contextvar+set}" = set; then :
+  withval=$with_decimal_contextvar;
+else
+  with_decimal_contextvar="yes"
+fi
+
+
+if test "$with_decimal_contextvar" != "no"
+then
+
+$as_echo "#define WITH_DECIMAL_CONTEXTVAR 1" >>confdefs.h
+
+fi
+
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $with_decimal_contextvar" >&5
+$as_echo "$with_decimal_contextvar" >&6; }
+
 # Check for support for loadable sqlite extensions
 { $as_echo "$as_me:${as_lineno-$LINENO}: checking for --enable-loadable-sqlite-extensions" >&5
 $as_echo_n "checking for --enable-loadable-sqlite-extensions... " >&6; }
diff --git a/configure.ac b/configure.ac
index 840caf3..bdd6073 100644
--- a/configure.ac
+++ b/configure.ac
@@ -3050,6 +3050,21 @@ AC_ARG_WITH(system_libmpdec,
 
 AC_MSG_RESULT($with_system_libmpdec)
 
+# Check whether _decimal should use a coroutine-local or thread-local context
+AC_MSG_CHECKING(for --with-decimal-contextvar)
+AC_ARG_WITH(decimal_contextvar,
+            AS_HELP_STRING([--with-decimal-contextvar], [build _decimal module using a coroutine-local rather than a thread-local context (default is yes)]),
+            [],
+            [with_decimal_contextvar="yes"])
+
+if test "$with_decimal_contextvar" != "no"
+then
+    AC_DEFINE(WITH_DECIMAL_CONTEXTVAR, 1,
+      [Define if you want build the _decimal module using a coroutine-local rather than a thread-local context])
+fi
+
+AC_MSG_RESULT($with_decimal_contextvar)
+
 # Check for support for loadable sqlite extensions
 AC_MSG_CHECKING(for --enable-loadable-sqlite-extensions)
 AC_ARG_ENABLE(loadable-sqlite-extensions,
diff --git a/pyconfig.h.in b/pyconfig.h.in
index b560213..2a72e9e 100644
--- a/pyconfig.h.in
+++ b/pyconfig.h.in
@@ -1512,6 +1512,10 @@
 /* Define if WINDOW in curses.h offers a field _flags. */
 #undef WINDOW_HAS_FLAGS
 
+/* Define if you want build the _decimal module using a coroutine-local rather
+   than a thread-local context */
+#undef WITH_DECIMAL_CONTEXTVAR
+
 /* Define if you want documentation strings in extension modules */
 #undef WITH_DOC_STRINGS