Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit 7c1f2a6

Browse filesBrowse files
markshannonmarkshannon
committed
Simplify and speed up interpreter for f-strings. Split FORMAT_VALUE opcode into CONVERT_VALUE, FORMAT_SIMPLE and FORMAT_WITH_SPEC.
Compiler can then emit optimal sequence for each format expression.
1 parent 7ecd342 commit 7c1f2a6
Copy full SHA for 7c1f2a6

File tree

Expand file treeCollapse file tree

13 files changed

+5022
-5006
lines changed
Filter options
Expand file treeCollapse file tree

13 files changed

+5022
-5006
lines changed

‎Include/opcode.h

Copy file name to clipboardExpand all lines: Include/opcode.h
+20-18Lines changed: 20 additions & 18 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

‎Lib/dis.py

Copy file name to clipboardExpand all lines: Lib/dis.py
+4-8Lines changed: 4 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -16,7 +16,7 @@
1616
_have_code = (types.MethodType, types.FunctionType, types.CodeType,
1717
classmethod, staticmethod, type)
1818

19-
FORMAT_VALUE = opmap['FORMAT_VALUE']
19+
CONVERT_VALUE = opmap['CONVERT_VALUE']
2020
FORMAT_VALUE_CONVERTERS = (
2121
(None, ''),
2222
(str, 'str'),
@@ -415,13 +415,9 @@ def _get_instructions_bytes(code, varname_from_oparg=None,
415415
elif op in hascompare:
416416
argval = cmp_op[arg]
417417
argrepr = argval
418-
elif op == FORMAT_VALUE:
419-
argval, argrepr = FORMAT_VALUE_CONVERTERS[arg & 0x3]
420-
argval = (argval, bool(arg & 0x4))
421-
if argval[1]:
422-
if argrepr:
423-
argrepr += ', '
424-
argrepr += 'with format'
418+
elif op == CONVERT_VALUE:
419+
argval = (None, str, repr, ascii)[arg]
420+
argrepr = ('', 'str', 'repr', 'ascii')[arg]
425421
elif op == MAKE_FUNCTION:
426422
argrepr = ', '.join(s for i, s in enumerate(MAKE_FUNCTION_FLAGS)
427423
if arg & (1<<i))

‎Lib/importlib/_bootstrap_external.py

Copy file name to clipboardExpand all lines: Lib/importlib/_bootstrap_external.py
+1Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -330,6 +330,7 @@ def _write_atomic(path, data, mode=0o666):
330330
# this might affected the first line number #32911)
331331
# Python 3.8a1 3400 (move frame block handling to compiler #17611)
332332
# Python 3.8a1 3401 (add END_ASYNC_FOR #33041)
333+
# Python 3.8a1 3402 (more efficient bytecodes for f-strings #33092)
333334
# Python 3.8a1 3410 (PEP570 Python Positional-Only Parameters #36540)
334335
# Python 3.8b2 3411 (Reverse evaluation order of key: value in dict
335336
# comprehensions #35224)

‎Lib/opcode.py

Copy file name to clipboardExpand all lines: Lib/opcode.py
+6-1Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -90,7 +90,11 @@ def jabs_op(name, op):
9090

9191
def_op('POP_EXCEPT_AND_RERAISE', 37)
9292

93+
def_op('FORMAT_SIMPLE', 40)
94+
def_op('FORMAT_WITH_SPEC', 41)
95+
9396
def_op('WITH_EXCEPT_START', 49)
97+
9498
def_op('GET_AITER', 50)
9599
def_op('GET_ANEXT', 51)
96100
def_op('BEFORE_ASYNC_WITH', 52)
@@ -202,10 +206,11 @@ def jabs_op(name, op):
202206
def_op('MAP_ADD', 147)
203207
def_op('LOAD_CLASSDEREF', 148)
204208
hasfree.append(148)
209+
def_op('CONVERT_VALUE', 149)
205210

206211
def_op('MATCH_CLASS', 152)
207212

208-
def_op('FORMAT_VALUE', 155)
213+
209214
def_op('BUILD_CONST_KEY_MAP', 156)
210215
def_op('BUILD_STRING', 157)
211216

‎Lib/test/test_dis.py

Copy file name to clipboardExpand all lines: Lib/test/test_dis.py
+11-9Lines changed: 11 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -334,20 +334,22 @@ def _fstring(a, b, c, d):
334334

335335
dis_fstring = """\
336336
%3d 0 LOAD_FAST 0 (a)
337-
2 FORMAT_VALUE 0
337+
2 FORMAT_SIMPLE
338338
4 LOAD_CONST 1 (' ')
339339
6 LOAD_FAST 1 (b)
340340
8 LOAD_CONST 2 ('4')
341-
10 FORMAT_VALUE 4 (with format)
341+
10 FORMAT_WITH_SPEC
342342
12 LOAD_CONST 1 (' ')
343343
14 LOAD_FAST 2 (c)
344-
16 FORMAT_VALUE 2 (repr)
345-
18 LOAD_CONST 1 (' ')
346-
20 LOAD_FAST 3 (d)
347-
22 LOAD_CONST 2 ('4')
348-
24 FORMAT_VALUE 6 (repr, with format)
349-
26 BUILD_STRING 7
350-
28 RETURN_VALUE
344+
16 CONVERT_VALUE 2 (repr)
345+
18 FORMAT_SIMPLE
346+
20 LOAD_CONST 1 (' ')
347+
22 LOAD_FAST 3 (d)
348+
24 CONVERT_VALUE 2 (repr)
349+
26 LOAD_CONST 2 ('4')
350+
28 FORMAT_WITH_SPEC
351+
30 BUILD_STRING 7
352+
32 RETURN_VALUE
351353
""" % (_fstring.__code__.co_firstlineno + 1,)
352354

353355
def _tryfinally(a, b):

‎Misc/NEWS.d/next/Core and Builtins/2018-03-17-05-36-19.bpo-33092.jZfBZd.rst

Copy file name to clipboard
+3Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,3 @@
1+
Simplify and speed up interpreter for f-strings. Removes ``FORMAT_VALUE``
2+
opcode. Add ``CONVERT_VALUE``, ``FORMAT_SIMPLE`` and ``FORMAT_WITH_SPEC`` opcode.
3+
Compiler emits more efficient sequence for each format expression.

‎Programs/test_frozenmain.h

Copy file name to clipboardExpand all lines: Programs/test_frozenmain.h
+2-2Lines changed: 2 additions & 2 deletions
Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

‎Python/ceval.c

Copy file name to clipboardExpand all lines: Python/ceval.c
+41-48Lines changed: 41 additions & 48 deletions
Original file line numberDiff line numberDiff line change
@@ -99,6 +99,14 @@ static void format_kwargs_error(PyThreadState *, PyObject *func, PyObject *kwarg
9999
static void format_awaitable_error(PyThreadState *, PyTypeObject *, int, int);
100100
static int get_exception_handler(PyCodeObject *, int, int*, int*, int*);
101101

102+
typedef PyObject *(*convertion_func_ptr)(PyObject *);
103+
104+
static const convertion_func_ptr CONVERSION_FUNCTIONS[4] = {
105+
[FVC_STR] = PyObject_Str,
106+
[FVC_REPR] = PyObject_Repr,
107+
[FVC_ASCII] = PyObject_ASCII
108+
};
109+
102110
#define NAME_ERROR_MSG \
103111
"name '%.200s' is not defined"
104112
#define UNBOUNDLOCAL_ERROR_MSG \
@@ -4346,7 +4354,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
43464354
oparg = cache->adaptive.original_oparg;
43474355
STAT_DEC(LOAD_METHOD, unquickened);
43484356
JUMP_TO_INSTRUCTION(LOAD_METHOD);
4349-
}
4357+
}
43504358
}
43514359

43524360
TARGET(LOAD_METHOD_CACHED): {
@@ -4364,7 +4372,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
43644372
assert(cache1->tp_version != 0);
43654373
assert(self_cls->tp_dictoffset >= 0);
43664374
assert(Py_TYPE(self_cls)->tp_dictoffset > 0);
4367-
4375+
43684376
// inline version of _PyObject_GetDictPtr for offset >= 0
43694377
PyObject *dict = self_cls->tp_dictoffset != 0 ?
43704378
*(PyObject **) ((char *)self + self_cls->tp_dictoffset) : NULL;
@@ -4627,62 +4635,47 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, InterpreterFrame *frame, int thr
46274635
DISPATCH();
46284636
}
46294637

4630-
TARGET(FORMAT_VALUE): {
4631-
/* Handles f-string value formatting. */
4638+
TARGET(CONVERT_VALUE): {
46324639
PyObject *result;
4633-
PyObject *fmt_spec;
4634-
PyObject *value;
4635-
PyObject *(*conv_fn)(PyObject *);
4636-
int which_conversion = oparg & FVC_MASK;
4637-
int have_fmt_spec = (oparg & FVS_MASK) == FVS_HAVE_SPEC;
4638-
4639-
fmt_spec = have_fmt_spec ? POP() : NULL;
4640-
value = POP();
4641-
4642-
/* See if any conversion is specified. */
4643-
switch (which_conversion) {
4644-
case FVC_NONE: conv_fn = NULL; break;
4645-
case FVC_STR: conv_fn = PyObject_Str; break;
4646-
case FVC_REPR: conv_fn = PyObject_Repr; break;
4647-
case FVC_ASCII: conv_fn = PyObject_ASCII; break;
4648-
default:
4649-
_PyErr_Format(tstate, PyExc_SystemError,
4650-
"unexpected conversion flag %d",
4651-
which_conversion);
4640+
PyObject *value = POP();
4641+
convertion_func_ptr conv_fn;
4642+
assert(oparg >= FVC_STR && oparg <= FVC_ASCII);
4643+
conv_fn = CONVERSION_FUNCTIONS[oparg];
4644+
result = conv_fn(value);
4645+
Py_DECREF(value);
4646+
if (result == NULL) {
46524647
goto error;
46534648
}
4649+
PUSH(result);
4650+
DISPATCH();
4651+
}
46544652

4655-
/* If there's a conversion function, call it and replace
4656-
value with that result. Otherwise, just use value,
4657-
without conversion. */
4658-
if (conv_fn != NULL) {
4659-
result = conv_fn(value);
4660-
Py_DECREF(value);
4653+
TARGET(FORMAT_SIMPLE): {
4654+
PyObject *value = TOP();
4655+
/* If value is a unicode object, then we know the result
4656+
* of format(value) is value itself. */
4657+
if (!PyUnicode_CheckExact(value)) {
4658+
PyObject *result = PyObject_Format(value, NULL);
46614659
if (result == NULL) {
4662-
Py_XDECREF(fmt_spec);
46634660
goto error;
46644661
}
4665-
value = result;
4666-
}
4667-
4668-
/* If value is a unicode object, and there's no fmt_spec,
4669-
then we know the result of format(value) is value
4670-
itself. In that case, skip calling format(). I plan to
4671-
move this optimization in to PyObject_Format()
4672-
itself. */
4673-
if (PyUnicode_CheckExact(value) && fmt_spec == NULL) {
4674-
/* Do nothing, just transfer ownership to result. */
4675-
result = value;
4676-
} else {
4677-
/* Actually call format(). */
4678-
result = PyObject_Format(value, fmt_spec);
4662+
SET_TOP(result);
46794663
Py_DECREF(value);
4680-
Py_XDECREF(fmt_spec);
4681-
if (result == NULL) {
4682-
goto error;
4683-
}
46844664
}
4665+
DISPATCH();
4666+
}
46854667

4668+
TARGET(FORMAT_WITH_SPEC): {
4669+
PyObject *fmt_spec = POP();
4670+
PyObject *value = POP();
4671+
PyObject *result;
4672+
/* Call format(). */
4673+
result = PyObject_Format(value, fmt_spec);
4674+
Py_DECREF(value);
4675+
Py_DECREF(fmt_spec);
4676+
if (result == NULL) {
4677+
goto error;
4678+
}
46864679
PUSH(result);
46874680
DISPATCH();
46884681
}

‎Python/compile.c

Copy file name to clipboardExpand all lines: Python/compile.c
+19-17Lines changed: 19 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -1231,10 +1231,12 @@ stack_effect(int opcode, int oparg, int jump)
12311231
return 0;
12321232
case END_ASYNC_FOR:
12331233
return -4;
1234-
case FORMAT_VALUE:
1235-
/* If there's a fmt_spec on the stack, we go from 2->1,
1236-
else 1->1. */
1237-
return (oparg & FVS_MASK) == FVS_HAVE_SPEC ? -1 : 0;
1234+
case CONVERT_VALUE:
1235+
return 0;
1236+
case FORMAT_WITH_SPEC:
1237+
return -1;
1238+
case FORMAT_SIMPLE:
1239+
return 0;
12381240
case LOAD_METHOD:
12391241
return 1;
12401242
case LOAD_ASSERTION_ERROR:
@@ -4486,26 +4488,26 @@ compiler_formatted_value(struct compiler *c, expr_ty e)
44864488

44874489
/* The expression to be formatted. */
44884490
VISIT(c, expr, e->v.FormattedValue.value);
4489-
4490-
switch (conversion) {
4491-
case 's': oparg = FVC_STR; break;
4492-
case 'r': oparg = FVC_REPR; break;
4493-
case 'a': oparg = FVC_ASCII; break;
4494-
case -1: oparg = FVC_NONE; break;
4495-
default:
4491+
if (e->v.FormattedValue.conversion != -1) {
4492+
switch (e->v.FormattedValue.conversion) {
4493+
case 's': oparg = FVC_STR; break;
4494+
case 'r': oparg = FVC_REPR; break;
4495+
case 'a': oparg = FVC_ASCII; break;
4496+
default:
44964497
PyErr_Format(PyExc_SystemError,
44974498
"Unrecognized conversion character %d", conversion);
4498-
return 0;
4499+
return 0;
4500+
}
4501+
ADDOP_I(c, CONVERT_VALUE, oparg);
44994502
}
45004503
if (e->v.FormattedValue.format_spec) {
4501-
/* Evaluate the format spec, and update our opcode arg. */
4504+
/* Evaluate the format spec, and emit format opcode. */
45024505
VISIT(c, expr, e->v.FormattedValue.format_spec);
4503-
oparg |= FVS_HAVE_SPEC;
4506+
ADDOP(c, FORMAT_WITH_SPEC);
4507+
} else {
4508+
ADDOP(c, FORMAT_SIMPLE);
45044509
}
45054510

4506-
/* And push our opcode and oparg */
4507-
ADDOP_I(c, FORMAT_VALUE, oparg);
4508-
45094511
return 1;
45104512
}
45114513

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.