circuitpython/py/parse.c

1192 lines
46 KiB
C
Raw Normal View History

/*
* This file is part of the MicroPython project, http://micropython.org/
*
* The MIT License (MIT)
*
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
* Copyright (c) 2013-2017 Damien P. George
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
* THE SOFTWARE.
*/
#include <stdbool.h>
2013-10-04 14:53:11 -04:00
#include <stdint.h>
#include <stdio.h>
#include <unistd.h> // for ssize_t
2013-10-04 14:53:11 -04:00
#include <assert.h>
#include <string.h>
2013-10-04 14:53:11 -04:00
#include "py/lexer.h"
#include "py/parse.h"
#include "py/parsenum.h"
#include "py/runtime.h"
#include "py/objint.h"
#include "py/objstr.h"
#include "py/builtin.h"
2013-10-04 14:53:11 -04:00
#if MICROPY_ENABLE_COMPILER
2013-10-04 14:53:11 -04:00
#define RULE_ACT_ARG_MASK (0x0f)
#define RULE_ACT_KIND_MASK (0x30)
#define RULE_ACT_ALLOW_IDENT (0x40)
#define RULE_ACT_ADD_BLANK (0x80)
2013-10-04 14:53:11 -04:00
#define RULE_ACT_OR (0x10)
#define RULE_ACT_AND (0x20)
#define RULE_ACT_LIST (0x30)
#define RULE_ARG_KIND_MASK (0xf000)
#define RULE_ARG_ARG_MASK (0x0fff)
#define RULE_ARG_TOK (0x1000)
#define RULE_ARG_RULE (0x2000)
#define RULE_ARG_OPT_RULE (0x3000)
2013-10-04 14:53:11 -04:00
// (un)comment to use rule names; for debugging
// #define USE_RULE_NAME (1)
2013-10-04 14:53:11 -04:00
// *FORMAT-OFF*
2013-10-04 14:53:11 -04:00
enum {
// define rules with a compile function
#define DEF_RULE(rule, comp, kind, ...) RULE_##rule,
#define DEF_RULE_NC(rule, kind, ...)
#include "py/grammar.h"
2013-10-04 14:53:11 -04:00
#undef DEF_RULE
#undef DEF_RULE_NC
RULE_const_object, // special node for a constant, generic Python object
// define rules without a compile function
#define DEF_RULE(rule, comp, kind, ...)
#define DEF_RULE_NC(rule, kind, ...) RULE_##rule,
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
2013-10-04 14:53:11 -04:00
};
// Define an array of actions corresponding to each rule
STATIC const uint8_t rule_act_table[] = {
2013-10-04 14:53:11 -04:00
#define or(n) (RULE_ACT_OR | n)
#define and(n) (RULE_ACT_AND | n)
#define and_ident(n) (RULE_ACT_AND | n | RULE_ACT_ALLOW_IDENT)
#define and_blank(n) (RULE_ACT_AND | n | RULE_ACT_ADD_BLANK)
2013-10-04 14:53:11 -04:00
#define one_or_more (RULE_ACT_LIST | 2)
#define list (RULE_ACT_LIST | 1)
#define list_with_end (RULE_ACT_LIST | 3)
#define DEF_RULE(rule, comp, kind, ...) kind,
#define DEF_RULE_NC(rule, kind, ...)
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
0, // RULE_const_object
#define DEF_RULE(rule, comp, kind, ...)
#define DEF_RULE_NC(rule, kind, ...) kind,
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
2013-10-04 14:53:11 -04:00
#undef or
#undef and
#undef and_ident
#undef and_blank
#undef one_or_more
2013-10-04 14:53:11 -04:00
#undef list
#undef list_with_end
};
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
// Define the argument data for each rule, as a combined array
STATIC const uint16_t rule_arg_combined_table[] = {
#define tok(t) (RULE_ARG_TOK | MP_TOKEN_##t)
#define rule(r) (RULE_ARG_RULE | RULE_##r)
#define opt_rule(r) (RULE_ARG_OPT_RULE | RULE_##r)
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
#define DEF_RULE(rule, comp, kind, ...) __VA_ARGS__,
#define DEF_RULE_NC(rule, kind, ...)
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
#define DEF_RULE(rule, comp, kind, ...)
#define DEF_RULE_NC(rule, kind, ...) __VA_ARGS__,
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
2013-10-04 14:53:11 -04:00
#undef tok
#undef rule
#undef opt_rule
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
};
// Macro to create a list of N identifiers where N is the number of variable arguments to the macro
#define RULE_EXPAND(x) x
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
#define RULE_PADDING(rule, ...) RULE_PADDING2(rule, __VA_ARGS__, RULE_PADDING_IDS(rule))
#define RULE_PADDING2(rule, ...) RULE_EXPAND(RULE_PADDING3(rule, __VA_ARGS__))
#define RULE_PADDING3(rule, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, ...) __VA_ARGS__
#define RULE_PADDING_IDS(r) PAD13_##r, PAD12_##r, PAD11_##r, PAD10_##r, PAD9_##r, PAD8_##r, PAD7_##r, PAD6_##r, PAD5_##r, PAD4_##r, PAD3_##r, PAD2_##r, PAD1_##r,
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
// Use an enum to create constants specifying how much room a rule takes in rule_arg_combined_table
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
enum {
#define DEF_RULE(rule, comp, kind, ...) RULE_PADDING(rule, __VA_ARGS__)
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
#define DEF_RULE_NC(rule, kind, ...)
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
#define DEF_RULE(rule, comp, kind, ...)
#define DEF_RULE_NC(rule, kind, ...) RULE_PADDING(rule, __VA_ARGS__)
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
};
2013-10-04 14:53:11 -04:00
// Macro to compute the start of a rule in rule_arg_combined_table
#define RULE_ARG_OFFSET(rule, ...) RULE_ARG_OFFSET2(rule, __VA_ARGS__, RULE_ARG_OFFSET_IDS(rule))
#define RULE_ARG_OFFSET2(rule, ...) RULE_EXPAND(RULE_ARG_OFFSET3(rule, __VA_ARGS__))
#define RULE_ARG_OFFSET3(rule, _1, _2, _3, _4, _5, _6, _7, _8, _9, _10, _11, _12, _13, _14, ...) _14
#define RULE_ARG_OFFSET_IDS(r) PAD13_##r, PAD12_##r, PAD11_##r, PAD10_##r, PAD9_##r, PAD8_##r, PAD7_##r, PAD6_##r, PAD5_##r, PAD4_##r, PAD3_##r, PAD2_##r, PAD1_##r, PAD0_##r,
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
// Use the above enum values to create a table of offsets for each rule's arg
// data, which indexes rule_arg_combined_table. The offsets require 9 bits of
// storage but only the lower 8 bits are stored here. The 9th bit is computed
// in get_rule_arg using the FIRST_RULE_WITH_OFFSET_ABOVE_255 constant.
STATIC const uint8_t rule_arg_offset_table[] = {
#define DEF_RULE(rule, comp, kind, ...) RULE_ARG_OFFSET(rule, __VA_ARGS__) & 0xff,
#define DEF_RULE_NC(rule, kind, ...)
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
0, // RULE_const_object
#define DEF_RULE(rule, comp, kind, ...)
#define DEF_RULE_NC(rule, kind, ...) RULE_ARG_OFFSET(rule, __VA_ARGS__) & 0xff,
#include "py/grammar.h"
2013-10-04 14:53:11 -04:00
#undef DEF_RULE
#undef DEF_RULE_NC
2013-10-04 14:53:11 -04:00
};
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
// Define a constant that's used to determine the 9th bit of the values in rule_arg_offset_table
static const size_t FIRST_RULE_WITH_OFFSET_ABOVE_255 =
#define DEF_RULE(rule, comp, kind, ...) RULE_ARG_OFFSET(rule, __VA_ARGS__) >= 0x100 ? RULE_##rule :
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
#define DEF_RULE_NC(rule, kind, ...)
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
#define DEF_RULE(rule, comp, kind, ...)
#define DEF_RULE_NC(rule, kind, ...) RULE_ARG_OFFSET(rule, __VA_ARGS__) >= 0x100 ? RULE_##rule :
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
0;
#if USE_RULE_NAME
// Define an array of rule names corresponding to each rule
STATIC const char *const rule_name_table[] = {
#define DEF_RULE(rule, comp, kind, ...) #rule,
#define DEF_RULE_NC(rule, kind, ...)
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
"", // RULE_const_object
#define DEF_RULE(rule, comp, kind, ...)
#define DEF_RULE_NC(rule, kind, ...) #rule,
#include "py/grammar.h"
#undef DEF_RULE
#undef DEF_RULE_NC
};
#endif
// *FORMAT-ON*
2013-10-04 14:53:11 -04:00
typedef struct _rule_stack_t {
size_t src_line : (8 * sizeof(size_t) - 8); // maximum bits storing source line number
size_t rule_id : 8; // this must be large enough to fit largest rule number
size_t arg_i; // this dictates the maximum nodes in a "list" of things
2013-10-04 14:53:11 -04:00
} rule_stack_t;
typedef struct _mp_parse_chunk_t {
size_t alloc;
union {
size_t used;
struct _mp_parse_chunk_t *next;
} union_;
byte data[];
} mp_parse_chunk_t;
2013-10-04 14:53:11 -04:00
typedef struct _parser_t {
size_t rule_stack_alloc;
size_t rule_stack_top;
2013-10-04 14:53:11 -04:00
rule_stack_t *rule_stack;
size_t result_stack_alloc;
size_t result_stack_top;
mp_parse_node_t *result_stack;
mp_lexer_t *lexer;
mp_parse_tree_t tree;
mp_parse_chunk_t *cur_chunk;
2013-10-04 14:53:11 -04:00
#if MICROPY_COMP_CONST
mp_map_t consts;
#endif
} parser_t;
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
STATIC const uint16_t *get_rule_arg(uint8_t r_id) {
size_t off = rule_arg_offset_table[r_id];
if (r_id >= FIRST_RULE_WITH_OFFSET_ABOVE_255) {
off |= 0x100;
}
return &rule_arg_combined_table[off];
}
STATIC void *parser_alloc(parser_t *parser, size_t num_bytes) {
// use a custom memory allocator to store parse nodes sequentially in large chunks
mp_parse_chunk_t *chunk = parser->cur_chunk;
if (chunk != NULL && chunk->union_.used + num_bytes > chunk->alloc) {
// not enough room at end of previously allocated chunk so try to grow
mp_parse_chunk_t *new_data = (mp_parse_chunk_t *)m_renew_maybe(byte, chunk,
sizeof(mp_parse_chunk_t) + chunk->alloc,
sizeof(mp_parse_chunk_t) + chunk->alloc + num_bytes, false);
if (new_data == NULL) {
// could not grow existing memory; shrink it to fit previous
(void)m_renew_maybe(byte, chunk, sizeof(mp_parse_chunk_t) + chunk->alloc,
sizeof(mp_parse_chunk_t) + chunk->union_.used, false);
chunk->alloc = chunk->union_.used;
chunk->union_.next = parser->tree.chunk;
parser->tree.chunk = chunk;
chunk = NULL;
} else {
// could grow existing memory
chunk->alloc += num_bytes;
}
}
if (chunk == NULL) {
// no previous chunk, allocate a new chunk
size_t alloc = MICROPY_ALLOC_PARSE_CHUNK_INIT;
if (alloc < num_bytes) {
alloc = num_bytes;
}
chunk = (mp_parse_chunk_t *)m_new(byte, sizeof(mp_parse_chunk_t) + alloc);
chunk->alloc = alloc;
chunk->union_.used = 0;
parser->cur_chunk = chunk;
}
byte *ret = chunk->data + chunk->union_.used;
chunk->union_.used += num_bytes;
return ret;
}
STATIC void push_rule(parser_t *parser, size_t src_line, uint8_t rule_id, size_t arg_i) {
2013-10-04 14:53:11 -04:00
if (parser->rule_stack_top >= parser->rule_stack_alloc) {
rule_stack_t *rs = m_renew(rule_stack_t, parser->rule_stack, parser->rule_stack_alloc, parser->rule_stack_alloc + MICROPY_ALLOC_PARSE_RULE_INC);
parser->rule_stack = rs;
parser->rule_stack_alloc += MICROPY_ALLOC_PARSE_RULE_INC;
2013-10-04 14:53:11 -04:00
}
rule_stack_t *rs = &parser->rule_stack[parser->rule_stack_top++];
rs->src_line = src_line;
rs->rule_id = rule_id;
rs->arg_i = arg_i;
2013-10-04 14:53:11 -04:00
}
STATIC void push_rule_from_arg(parser_t *parser, size_t arg) {
2013-10-04 14:53:11 -04:00
assert((arg & RULE_ARG_KIND_MASK) == RULE_ARG_RULE || (arg & RULE_ARG_KIND_MASK) == RULE_ARG_OPT_RULE);
size_t rule_id = arg & RULE_ARG_ARG_MASK;
push_rule(parser, parser->lexer->tok_line, rule_id, 0);
2013-10-04 14:53:11 -04:00
}
STATIC uint8_t pop_rule(parser_t *parser, size_t *arg_i, size_t *src_line) {
2013-10-04 14:53:11 -04:00
parser->rule_stack_top -= 1;
uint8_t rule_id = parser->rule_stack[parser->rule_stack_top].rule_id;
2013-10-04 14:53:11 -04:00
*arg_i = parser->rule_stack[parser->rule_stack_top].arg_i;
*src_line = parser->rule_stack[parser->rule_stack_top].src_line;
return rule_id;
2013-10-04 14:53:11 -04:00
}
bool mp_parse_node_is_const_false(mp_parse_node_t pn) {
return MP_PARSE_NODE_IS_TOKEN_KIND(pn, MP_TOKEN_KW_FALSE)
|| (MP_PARSE_NODE_IS_SMALL_INT(pn) && MP_PARSE_NODE_LEAF_SMALL_INT(pn) == 0);
}
bool mp_parse_node_is_const_true(mp_parse_node_t pn) {
return MP_PARSE_NODE_IS_TOKEN_KIND(pn, MP_TOKEN_KW_TRUE)
|| (MP_PARSE_NODE_IS_SMALL_INT(pn) && MP_PARSE_NODE_LEAF_SMALL_INT(pn) != 0);
}
bool mp_parse_node_get_int_maybe(mp_parse_node_t pn, mp_obj_t *o) {
if (MP_PARSE_NODE_IS_SMALL_INT(pn)) {
*o = MP_OBJ_NEW_SMALL_INT(MP_PARSE_NODE_LEAF_SMALL_INT(pn));
return true;
} else if (MP_PARSE_NODE_IS_STRUCT_KIND(pn, RULE_const_object)) {
mp_parse_node_struct_t *pns = (mp_parse_node_struct_t *)pn;
#if MICROPY_OBJ_REPR == MICROPY_OBJ_REPR_D
// nodes are 32-bit pointers, but need to extract 64-bit object
*o = (uint64_t)pns->nodes[0] | ((uint64_t)pns->nodes[1] << 32);
#else
*o = (mp_obj_t)pns->nodes[0];
#endif
return mp_obj_is_int(*o);
} else {
return false;
}
}
size_t mp_parse_node_extract_list(mp_parse_node_t *pn, size_t pn_kind, mp_parse_node_t **nodes) {
if (MP_PARSE_NODE_IS_NULL(*pn)) {
*nodes = NULL;
return 0;
} else if (MP_PARSE_NODE_IS_LEAF(*pn)) {
*nodes = pn;
return 1;
} else {
mp_parse_node_struct_t *pns = (mp_parse_node_struct_t *)(*pn);
if (MP_PARSE_NODE_STRUCT_KIND(pns) != pn_kind) {
*nodes = pn;
return 1;
} else {
*nodes = pns->nodes;
return MP_PARSE_NODE_STRUCT_NUM_NODES(pns);
}
}
}
#if MICROPY_DEBUG_PRINTERS
void mp_parse_node_print(const mp_print_t *print, mp_parse_node_t pn, size_t indent) {
if (MP_PARSE_NODE_IS_STRUCT(pn)) {
mp_printf(print, "[% 4d] ", (int)((mp_parse_node_struct_t *)pn)->source_line);
} else {
mp_printf(print, " ");
}
for (size_t i = 0; i < indent; i++) {
mp_printf(print, " ");
2013-10-04 14:53:11 -04:00
}
if (MP_PARSE_NODE_IS_NULL(pn)) {
mp_printf(print, "NULL\n");
} else if (MP_PARSE_NODE_IS_SMALL_INT(pn)) {
mp_int_t arg = MP_PARSE_NODE_LEAF_SMALL_INT(pn);
mp_printf(print, "int(" INT_FMT ")\n", arg);
} else if (MP_PARSE_NODE_IS_LEAF(pn)) {
uintptr_t arg = MP_PARSE_NODE_LEAF_ARG(pn);
switch (MP_PARSE_NODE_LEAF_KIND(pn)) {
case MP_PARSE_NODE_ID:
mp_printf(print, "id(%s)\n", qstr_str(arg));
break;
case MP_PARSE_NODE_STRING:
mp_printf(print, "str(%s)\n", qstr_str(arg));
break;
case MP_PARSE_NODE_BYTES:
mp_printf(print, "bytes(%s)\n", qstr_str(arg));
break;
default:
assert(MP_PARSE_NODE_LEAF_KIND(pn) == MP_PARSE_NODE_TOKEN);
mp_printf(print, "tok(%u)\n", (uint)arg);
break;
2013-10-04 14:53:11 -04:00
}
} else {
// node must be a mp_parse_node_struct_t
mp_parse_node_struct_t *pns = (mp_parse_node_struct_t *)pn;
if (MP_PARSE_NODE_STRUCT_KIND(pns) == RULE_const_object) {
#if MICROPY_OBJ_REPR == MICROPY_OBJ_REPR_D
mp_printf(print, "literal const(%016llx)\n", (uint64_t)pns->nodes[0] | ((uint64_t)pns->nodes[1] << 32));
#else
mp_printf(print, "literal const(%p)\n", (mp_obj_t)pns->nodes[0]);
#endif
} else {
size_t n = MP_PARSE_NODE_STRUCT_NUM_NODES(pns);
#if USE_RULE_NAME
mp_printf(print, "%s(%u) (n=%u)\n", rule_name_table[MP_PARSE_NODE_STRUCT_KIND(pns)], (uint)MP_PARSE_NODE_STRUCT_KIND(pns), (uint)n);
#else
mp_printf(print, "rule(%u) (n=%u)\n", (uint)MP_PARSE_NODE_STRUCT_KIND(pns), (uint)n);
#endif
for (size_t i = 0; i < n; i++) {
mp_parse_node_print(print, pns->nodes[i], indent + 2);
}
2013-10-04 14:53:11 -04:00
}
}
}
#endif // MICROPY_DEBUG_PRINTERS
2013-10-04 14:53:11 -04:00
/*
STATIC void result_stack_show(const mp_print_t *print, parser_t *parser) {
mp_printf(print, "result stack, most recent first\n");
for (ssize_t i = parser->result_stack_top - 1; i >= 0; i--) {
mp_parse_node_print(print, parser->result_stack[i], 0);
2013-10-04 14:53:11 -04:00
}
}
*/
STATIC mp_parse_node_t pop_result(parser_t *parser) {
2013-10-04 14:53:11 -04:00
assert(parser->result_stack_top > 0);
return parser->result_stack[--parser->result_stack_top];
}
STATIC mp_parse_node_t peek_result(parser_t *parser, size_t pos) {
2013-10-04 14:53:11 -04:00
assert(parser->result_stack_top > pos);
return parser->result_stack[parser->result_stack_top - 1 - pos];
}
STATIC void push_result_node(parser_t *parser, mp_parse_node_t pn) {
if (parser->result_stack_top >= parser->result_stack_alloc) {
mp_parse_node_t *stack = m_renew(mp_parse_node_t, parser->result_stack, parser->result_stack_alloc, parser->result_stack_alloc + MICROPY_ALLOC_PARSE_RESULT_INC);
parser->result_stack = stack;
parser->result_stack_alloc += MICROPY_ALLOC_PARSE_RESULT_INC;
}
2013-10-04 14:53:11 -04:00
parser->result_stack[parser->result_stack_top++] = pn;
}
STATIC mp_parse_node_t make_node_const_object(parser_t *parser, size_t src_line, mp_obj_t obj) {
mp_parse_node_struct_t *pn = parser_alloc(parser, sizeof(mp_parse_node_struct_t) + sizeof(mp_obj_t));
pn->source_line = src_line;
#if MICROPY_OBJ_REPR == MICROPY_OBJ_REPR_D
// nodes are 32-bit pointers, but need to store 64-bit object
pn->kind_num_nodes = RULE_const_object | (2 << 8);
pn->nodes[0] = (uint64_t)obj;
pn->nodes[1] = (uint64_t)obj >> 32;
#else
pn->kind_num_nodes = RULE_const_object | (1 << 8);
pn->nodes[0] = (uintptr_t)obj;
#endif
return (mp_parse_node_t)pn;
}
STATIC mp_parse_node_t mp_parse_node_new_small_int_checked(parser_t *parser, mp_obj_t o_val) {
(void)parser;
mp_int_t val = MP_OBJ_SMALL_INT_VALUE(o_val);
#if MICROPY_OBJ_REPR == MICROPY_OBJ_REPR_D
// A parse node is only 32-bits and the small-int value must fit in 31-bits
if (((val ^ (val << 1)) & 0xffffffff80000000) != 0) {
return make_node_const_object(parser, 0, o_val);
}
#endif
return mp_parse_node_new_small_int(val);
}
STATIC void push_result_token(parser_t *parser, uint8_t rule_id) {
mp_parse_node_t pn;
mp_lexer_t *lex = parser->lexer;
if (lex->tok_kind == MP_TOKEN_NAME) {
qstr id = qstr_from_strn(lex->vstr.buf, lex->vstr.len);
#if MICROPY_COMP_CONST
// if name is a standalone identifier, look it up in the table of dynamic constants
mp_map_elem_t *elem;
if (rule_id == RULE_atom
&& (elem = mp_map_lookup(&parser->consts, MP_OBJ_NEW_QSTR(id), MP_MAP_LOOKUP)) != NULL) {
if (mp_obj_is_small_int(elem->value)) {
pn = mp_parse_node_new_small_int_checked(parser, elem->value);
} else {
pn = make_node_const_object(parser, lex->tok_line, elem->value);
}
} else {
pn = mp_parse_node_new_leaf(MP_PARSE_NODE_ID, id);
}
#else
(void)rule_id;
pn = mp_parse_node_new_leaf(MP_PARSE_NODE_ID, id);
#endif
} else if (lex->tok_kind == MP_TOKEN_INTEGER) {
mp_obj_t o = mp_parse_num_integer(lex->vstr.buf, lex->vstr.len, 0, lex);
if (mp_obj_is_small_int(o)) {
pn = mp_parse_node_new_small_int_checked(parser, o);
2013-10-04 14:53:11 -04:00
} else {
pn = make_node_const_object(parser, lex->tok_line, o);
2013-10-04 14:53:11 -04:00
}
} else if (lex->tok_kind == MP_TOKEN_FLOAT_OR_IMAG) {
mp_obj_t o = mp_parse_num_decimal(lex->vstr.buf, lex->vstr.len, true, false, lex);
pn = make_node_const_object(parser, lex->tok_line, o);
} else if (lex->tok_kind == MP_TOKEN_STRING || lex->tok_kind == MP_TOKEN_BYTES) {
// Don't automatically intern all strings/bytes. doc strings (which are usually large)
// will be discarded by the compiler, and so we shouldn't intern them.
qstr qst = MP_QSTRnull;
if (lex->vstr.len <= MICROPY_ALLOC_PARSE_INTERN_STRING_LEN) {
// intern short strings
qst = qstr_from_strn(lex->vstr.buf, lex->vstr.len);
} else {
// check if this string is already interned
qst = qstr_find_strn(lex->vstr.buf, lex->vstr.len);
}
if (qst != MP_QSTRnull) {
// qstr exists, make a leaf node
pn = mp_parse_node_new_leaf(lex->tok_kind == MP_TOKEN_STRING ? MP_PARSE_NODE_STRING : MP_PARSE_NODE_BYTES, qst);
} else {
// not interned, make a node holding a pointer to the string/bytes object
mp_obj_t o = mp_obj_new_str_copy(
lex->tok_kind == MP_TOKEN_STRING ? &mp_type_str : &mp_type_bytes,
(const byte *)lex->vstr.buf, lex->vstr.len);
pn = make_node_const_object(parser, lex->tok_line, o);
}
2013-10-04 14:53:11 -04:00
} else {
pn = mp_parse_node_new_leaf(MP_PARSE_NODE_TOKEN, lex->tok_kind);
2013-10-04 14:53:11 -04:00
}
push_result_node(parser, pn);
}
#if MICROPY_COMP_MODULE_CONST
STATIC const mp_rom_map_elem_t mp_constants_table[] = {
#if MICROPY_PY_UERRNO
{ MP_ROM_QSTR(MP_QSTR_errno), MP_ROM_PTR(&mp_module_uerrno) },
#endif
#if MICROPY_PY_UCTYPES
{ MP_ROM_QSTR(MP_QSTR_uctypes), MP_ROM_PTR(&mp_module_uctypes) },
#endif
// Extra constants as defined by a port
MICROPY_PORT_CONSTANTS
};
STATIC MP_DEFINE_CONST_MAP(mp_constants_map, mp_constants_table);
#endif
STATIC void push_result_rule(parser_t *parser, size_t src_line, uint8_t rule_id, size_t num_args);
#if MICROPY_COMP_CONST_FOLDING
STATIC bool fold_logical_constants(parser_t *parser, uint8_t rule_id, size_t *num_args) {
if (rule_id == RULE_or_test
|| rule_id == RULE_and_test) {
// folding for binary logical ops: or and
size_t copy_to = *num_args;
for (size_t i = copy_to; i > 0;) {
mp_parse_node_t pn = peek_result(parser, --i);
parser->result_stack[parser->result_stack_top - copy_to] = pn;
if (i == 0) {
// always need to keep the last value
break;
}
if (rule_id == RULE_or_test) {
if (mp_parse_node_is_const_true(pn)) {
//
break;
} else if (!mp_parse_node_is_const_false(pn)) {
copy_to -= 1;
}
} else {
// RULE_and_test
if (mp_parse_node_is_const_false(pn)) {
break;
} else if (!mp_parse_node_is_const_true(pn)) {
copy_to -= 1;
}
}
}
copy_to -= 1; // copy_to now contains number of args to pop
// pop and discard all the short-circuited expressions
for (size_t i = 0; i < copy_to; ++i) {
pop_result(parser);
}
*num_args -= copy_to;
// we did a complete folding if there's only 1 arg left
return *num_args == 1;
} else if (rule_id == RULE_not_test_2) {
// folding for unary logical op: not
mp_parse_node_t pn = peek_result(parser, 0);
if (mp_parse_node_is_const_false(pn)) {
pn = mp_parse_node_new_leaf(MP_PARSE_NODE_TOKEN, MP_TOKEN_KW_TRUE);
} else if (mp_parse_node_is_const_true(pn)) {
pn = mp_parse_node_new_leaf(MP_PARSE_NODE_TOKEN, MP_TOKEN_KW_FALSE);
} else {
return false;
}
pop_result(parser);
push_result_node(parser, pn);
return true;
}
return false;
}
STATIC bool fold_constants(parser_t *parser, uint8_t rule_id, size_t num_args) {
// this code does folding of arbitrary integer expressions, eg 1 + 2 * 3 + 4
// it does not do partial folding, eg 1 + 2 + x -> 3 + x
mp_obj_t arg0;
if (rule_id == RULE_expr
|| rule_id == RULE_xor_expr
|| rule_id == RULE_and_expr
|| rule_id == RULE_power) {
// folding for binary ops: | ^ & **
mp_parse_node_t pn = peek_result(parser, num_args - 1);
if (!mp_parse_node_get_int_maybe(pn, &arg0)) {
return false;
}
mp_binary_op_t op;
if (rule_id == RULE_expr) {
op = MP_BINARY_OP_OR;
} else if (rule_id == RULE_xor_expr) {
op = MP_BINARY_OP_XOR;
} else if (rule_id == RULE_and_expr) {
op = MP_BINARY_OP_AND;
} else {
op = MP_BINARY_OP_POWER;
}
for (ssize_t i = num_args - 2; i >= 0; --i) {
pn = peek_result(parser, i);
mp_obj_t arg1;
if (!mp_parse_node_get_int_maybe(pn, &arg1)) {
return false;
}
if (op == MP_BINARY_OP_POWER && mp_obj_int_sign(arg1) < 0) {
// ** can't have negative rhs
return false;
}
arg0 = mp_binary_op(op, arg0, arg1);
}
} else if (rule_id == RULE_shift_expr
|| rule_id == RULE_arith_expr
|| rule_id == RULE_term) {
// folding for binary ops: << >> + - * @ / % //
mp_parse_node_t pn = peek_result(parser, num_args - 1);
if (!mp_parse_node_get_int_maybe(pn, &arg0)) {
return false;
}
for (ssize_t i = num_args - 2; i >= 1; i -= 2) {
pn = peek_result(parser, i - 1);
mp_obj_t arg1;
if (!mp_parse_node_get_int_maybe(pn, &arg1)) {
return false;
}
mp_token_kind_t tok = MP_PARSE_NODE_LEAF_ARG(peek_result(parser, i));
if (tok == MP_TOKEN_OP_AT || tok == MP_TOKEN_OP_SLASH) {
// Can't fold @ or /
return false;
}
mp_binary_op_t op = MP_BINARY_OP_LSHIFT + (tok - MP_TOKEN_OP_DBL_LESS);
int rhs_sign = mp_obj_int_sign(arg1);
if (op <= MP_BINARY_OP_RSHIFT) {
// << and >> can't have negative rhs
if (rhs_sign < 0) {
return false;
}
} else if (op >= MP_BINARY_OP_FLOOR_DIVIDE) {
// % and // can't have zero rhs
if (rhs_sign == 0) {
return false;
}
}
arg0 = mp_binary_op(op, arg0, arg1);
}
} else if (rule_id == RULE_factor_2) {
// folding for unary ops: + - ~
mp_parse_node_t pn = peek_result(parser, 0);
if (!mp_parse_node_get_int_maybe(pn, &arg0)) {
return false;
}
mp_token_kind_t tok = MP_PARSE_NODE_LEAF_ARG(peek_result(parser, 1));
mp_unary_op_t op;
if (tok == MP_TOKEN_OP_TILDE) {
op = MP_UNARY_OP_INVERT;
} else {
assert(tok == MP_TOKEN_OP_PLUS || tok == MP_TOKEN_OP_MINUS); // should be
op = MP_UNARY_OP_POSITIVE + (tok - MP_TOKEN_OP_PLUS);
}
arg0 = mp_unary_op(op, arg0);
#if MICROPY_COMP_CONST
} else if (rule_id == RULE_expr_stmt) {
mp_parse_node_t pn1 = peek_result(parser, 0);
if (!MP_PARSE_NODE_IS_NULL(pn1)
&& !(MP_PARSE_NODE_IS_STRUCT_KIND(pn1, RULE_expr_stmt_augassign)
|| MP_PARSE_NODE_IS_STRUCT_KIND(pn1, RULE_expr_stmt_assign_list))) {
// this node is of the form <x> = <y>
mp_parse_node_t pn0 = peek_result(parser, 1);
if (MP_PARSE_NODE_IS_ID(pn0)
&& MP_PARSE_NODE_IS_STRUCT_KIND(pn1, RULE_atom_expr_normal)
&& MP_PARSE_NODE_IS_ID(((mp_parse_node_struct_t *)pn1)->nodes[0])
&& MP_PARSE_NODE_LEAF_ARG(((mp_parse_node_struct_t *)pn1)->nodes[0]) == MP_QSTR_const
&& MP_PARSE_NODE_IS_STRUCT_KIND(((mp_parse_node_struct_t *)pn1)->nodes[1], RULE_trailer_paren)
) {
// code to assign dynamic constants: id = const(value)
// get the id
qstr id = MP_PARSE_NODE_LEAF_ARG(pn0);
// get the value
mp_parse_node_t pn_value = ((mp_parse_node_struct_t *)((mp_parse_node_struct_t *)pn1)->nodes[1])->nodes[0];
mp_obj_t value;
if (!mp_parse_node_get_int_maybe(pn_value, &value)) {
mp_obj_t exc = mp_obj_new_exception_msg(&mp_type_SyntaxError,
MP_ERROR_TEXT("constant must be an integer"));
mp_obj_exception_add_traceback(exc, parser->lexer->source_name,
((mp_parse_node_struct_t *)pn1)->source_line, MP_QSTRnull);
nlr_raise(exc);
}
// store the value in the table of dynamic constants
mp_map_elem_t *elem = mp_map_lookup(&parser->consts, MP_OBJ_NEW_QSTR(id), MP_MAP_LOOKUP_ADD_IF_NOT_FOUND);
assert(elem->value == MP_OBJ_NULL);
elem->value = value;
// If the constant starts with an underscore then treat it as a private
// variable and don't emit any code to store the value to the id.
if (qstr_str(id)[0] == '_') {
pop_result(parser); // pop const(value)
pop_result(parser); // pop id
push_result_rule(parser, 0, RULE_pass_stmt, 0); // replace with "pass"
return true;
}
// replace const(value) with value
pop_result(parser);
push_result_node(parser, pn_value);
// finished folding this assignment, but we still want it to be part of the tree
return false;
}
}
return false;
#endif
#if MICROPY_COMP_MODULE_CONST
} else if (rule_id == RULE_atom_expr_normal) {
mp_parse_node_t pn0 = peek_result(parser, 1);
mp_parse_node_t pn1 = peek_result(parser, 0);
if (!(MP_PARSE_NODE_IS_ID(pn0)
&& MP_PARSE_NODE_IS_STRUCT_KIND(pn1, RULE_trailer_period))) {
return false;
}
// id1.id2
// look it up in constant table, see if it can be replaced with an integer
mp_parse_node_struct_t *pns1 = (mp_parse_node_struct_t *)pn1;
assert(MP_PARSE_NODE_IS_ID(pns1->nodes[0]));
qstr q_base = MP_PARSE_NODE_LEAF_ARG(pn0);
qstr q_attr = MP_PARSE_NODE_LEAF_ARG(pns1->nodes[0]);
mp_map_elem_t *elem = mp_map_lookup((mp_map_t *)&mp_constants_map, MP_OBJ_NEW_QSTR(q_base), MP_MAP_LOOKUP);
if (elem == NULL) {
return false;
}
mp_obj_t dest[2];
mp_load_method_maybe(elem->value, q_attr, dest);
if (!(dest[0] != MP_OBJ_NULL && mp_obj_is_int(dest[0]) && dest[1] == MP_OBJ_NULL)) {
return false;
}
arg0 = dest[0];
#endif
} else {
return false;
}
// success folding this rule
for (size_t i = num_args; i > 0; i--) {
pop_result(parser);
}
if (mp_obj_is_small_int(arg0)) {
push_result_node(parser, mp_parse_node_new_small_int_checked(parser, arg0));
} else {
// TODO reuse memory for parse node struct?
push_result_node(parser, make_node_const_object(parser, 0, arg0));
}
return true;
}
#endif
STATIC void push_result_rule(parser_t *parser, size_t src_line, uint8_t rule_id, size_t num_args) {
// optimise away parenthesis around an expression if possible
if (rule_id == RULE_atom_paren) {
// there should be just 1 arg for this rule
mp_parse_node_t pn = peek_result(parser, 0);
if (MP_PARSE_NODE_IS_NULL(pn)) {
// need to keep parenthesis for ()
} else if (MP_PARSE_NODE_IS_STRUCT_KIND(pn, RULE_testlist_comp)) {
// need to keep parenthesis for (a, b, ...)
} else {
// parenthesis around a single expression, so it's just the expression
return;
}
}
#if MICROPY_COMP_CONST_FOLDING
if (fold_logical_constants(parser, rule_id, &num_args)) {
// we folded this rule so return straight away
return;
}
if (fold_constants(parser, rule_id, num_args)) {
// we folded this rule so return straight away
return;
}
#endif
mp_parse_node_struct_t *pn = parser_alloc(parser, sizeof(mp_parse_node_struct_t) + sizeof(mp_parse_node_t) * num_args);
pn->source_line = src_line;
pn->kind_num_nodes = (rule_id & 0xff) | (num_args << 8);
for (size_t i = num_args; i > 0; i--) {
2013-10-04 14:53:11 -04:00
pn->nodes[i - 1] = pop_result(parser);
}
push_result_node(parser, (mp_parse_node_t)pn);
2013-10-04 14:53:11 -04:00
}
mp_parse_tree_t mp_parse(mp_lexer_t *lex, mp_parse_input_kind_t input_kind) {
// initialise parser and allocate memory for its stacks
parser_t parser;
parser.rule_stack_alloc = MICROPY_ALLOC_PARSE_RULE_INIT;
parser.rule_stack_top = 0;
parser.rule_stack = m_new(rule_stack_t, parser.rule_stack_alloc);
2013-10-04 14:53:11 -04:00
parser.result_stack_alloc = MICROPY_ALLOC_PARSE_RESULT_INIT;
parser.result_stack_top = 0;
parser.result_stack = m_new(mp_parse_node_t, parser.result_stack_alloc);
2013-10-04 14:53:11 -04:00
parser.lexer = lex;
parser.tree.chunk = NULL;
parser.cur_chunk = NULL;
#if MICROPY_COMP_CONST
mp_map_init(&parser.consts, 0);
#endif
// work out the top-level rule to use, and push it on the stack
size_t top_level_rule;
2013-10-18 14:58:12 -04:00
switch (input_kind) {
case MP_PARSE_SINGLE_INPUT:
top_level_rule = RULE_single_input;
break;
case MP_PARSE_EVAL_INPUT:
top_level_rule = RULE_eval_input;
break;
default:
top_level_rule = RULE_file_input;
2013-10-18 14:58:12 -04:00
}
push_rule(&parser, lex->tok_line, top_level_rule, 0);
2013-10-04 14:53:11 -04:00
// parse!
2013-10-04 14:53:11 -04:00
bool backtrack = false;
for (;;) {
next_rule:
if (parser.rule_stack_top == 0) {
2013-10-04 14:53:11 -04:00
break;
}
// Pop the next rule to process it
size_t i; // state for the current rule
size_t rule_src_line; // source line for the first token matched by the current rule
uint8_t rule_id = pop_rule(&parser, &i, &rule_src_line);
uint8_t rule_act = rule_act_table[rule_id];
py/parse: Compress rule pointer table to table of offsets. This is the sixth and final patch in a series of patches to the parser that aims to reduce code size by compressing the data corresponding to the rules of the grammar. Prior to this set of patches the rules were stored as rule_t structs with rule_id, act and arg members. And then there was a big table of pointers which allowed to lookup the address of a rule_t struct given the id of that rule. The changes that have been made are: - Breaking up of the rule_t struct into individual components, with each component in a separate array. - Removal of the rule_id part of the struct because it's not needed. - Put all the rule arg data in a big array. - Change the table of pointers to rules to a table of offsets within the array of rule arg data. The last point is what is done in this patch here and brings about the biggest decreases in code size, because an array of pointers is now an array of bytes. Code size changes for the six patches combined is: bare-arm: -644 minimal x86: -1856 unix x64: -5408 unix nanbox: -2080 stm32: -720 esp8266: -812 cc3200: -712 For the change in parser performance: it was measured on pyboard that these six patches combined gave an increase in script parse time of about 0.4%. This is due to the slightly more complicated way of looking up the data for a rule (since the 9th bit of the offset into the rule arg data table is calculated with an if statement). This is an acceptable increase in parse time considering that parsing is only done once per script (if compiled on the target).
2017-12-25 21:39:02 -05:00
const uint16_t *rule_arg = get_rule_arg(rule_id);
size_t n = rule_act & RULE_ACT_ARG_MASK;
2013-10-04 14:53:11 -04:00
#if 0
2013-10-04 14:53:11 -04:00
// debugging
printf("depth=" UINT_FMT " ", parser.rule_stack_top);
for (int j = 0; j < parser.rule_stack_top; ++j) {
2013-10-04 14:53:11 -04:00
printf(" ");
}
printf("%s n=" UINT_FMT " i=" UINT_FMT " bt=%d\n", rule_name_table[rule_id], n, i, backtrack);
#endif
2013-10-04 14:53:11 -04:00
switch (rule_act & RULE_ACT_KIND_MASK) {
2013-10-04 14:53:11 -04:00
case RULE_ACT_OR:
if (i > 0 && !backtrack) {
goto next_rule;
} else {
backtrack = false;
}
for (; i < n; ++i) {
uint16_t kind = rule_arg[i] & RULE_ARG_KIND_MASK;
if (kind == RULE_ARG_TOK) {
if (lex->tok_kind == (rule_arg[i] & RULE_ARG_ARG_MASK)) {
push_result_token(&parser, rule_id);
mp_lexer_to_next(lex);
2013-10-04 14:53:11 -04:00
goto next_rule;
}
2013-10-04 14:53:11 -04:00
} else {
assert(kind == RULE_ARG_RULE);
if (i + 1 < n) {
push_rule(&parser, rule_src_line, rule_id, i + 1); // save this or-rule
}
push_rule_from_arg(&parser, rule_arg[i]); // push child of or-rule
2013-10-04 14:53:11 -04:00
goto next_rule;
}
}
backtrack = true;
2013-10-04 14:53:11 -04:00
break;
case RULE_ACT_AND: {
2013-10-04 14:53:11 -04:00
// failed, backtrack if we can, else syntax error
if (backtrack) {
assert(i > 0);
if ((rule_arg[i - 1] & RULE_ARG_KIND_MASK) == RULE_ARG_OPT_RULE) {
2013-10-04 14:53:11 -04:00
// an optional rule that failed, so continue with next arg
push_result_node(&parser, MP_PARSE_NODE_NULL);
2013-10-04 14:53:11 -04:00
backtrack = false;
} else {
// a mandatory rule that failed, so propagate backtrack
if (i > 1) {
// already eaten tokens so can't backtrack
goto syntax_error;
} else {
goto next_rule;
}
}
}
// progress through the rule
for (; i < n; ++i) {
if ((rule_arg[i] & RULE_ARG_KIND_MASK) == RULE_ARG_TOK) {
// need to match a token
mp_token_kind_t tok_kind = rule_arg[i] & RULE_ARG_ARG_MASK;
if (lex->tok_kind == tok_kind) {
// matched token
if (tok_kind == MP_TOKEN_NAME) {
push_result_token(&parser, rule_id);
}
mp_lexer_to_next(lex);
} else {
// failed to match token
if (i > 0) {
// already eaten tokens so can't backtrack
goto syntax_error;
2013-10-04 14:53:11 -04:00
} else {
// this rule failed, so backtrack
backtrack = true;
goto next_rule;
2013-10-04 14:53:11 -04:00
}
}
} else {
push_rule(&parser, rule_src_line, rule_id, i + 1); // save this and-rule
push_rule_from_arg(&parser, rule_arg[i]); // push child of and-rule
goto next_rule;
2013-10-04 14:53:11 -04:00
}
}
assert(i == n);
// matched the rule, so now build the corresponding parse_node
unix-cpy: Remove unix-cpy. It's no longer needed. unix-cpy was originally written to get semantic equivalent with CPython without writing functional tests. When writing the initial implementation of uPy it was a long way between lexer and functional tests, so the half-way test was to make sure that the bytecode was correct. The idea was that if the uPy bytecode matched CPython 1-1 then uPy would be proper Python if the bytecodes acted correctly. And having matching bytecode meant that it was less likely to miss some deep subtlety in the Python semantics that would require an architectural change later on. But that is all history and it no longer makes sense to retain the ability to output CPython bytecode, because: 1. It outputs CPython 3.3 compatible bytecode. CPython's bytecode changes from version to version, and seems to have changed quite a bit in 3.5. There's no point in changing the bytecode output to match CPython anymore. 2. uPy and CPy do different optimisations to the bytecode which makes it harder to match. 3. The bytecode tests are not run. They were never part of Travis and are not run locally anymore. 4. The EMIT_CPYTHON option needs a lot of extra source code which adds heaps of noise, especially in compile.c. 5. Now that there is an extensive test suite (which tests functionality) there is no need to match the bytecode. Some very subtle behaviour is tested with the test suite and passing these tests is a much better way to stay Python-language compliant, rather than trying to match CPy bytecode.
2015-08-14 07:24:11 -04:00
#if !MICROPY_ENABLE_DOC_STRING
// this code discards lonely statements, such as doc strings
if (input_kind != MP_PARSE_SINGLE_INPUT && rule_id == RULE_expr_stmt && peek_result(&parser, 0) == MP_PARSE_NODE_NULL) {
mp_parse_node_t p = peek_result(&parser, 1);
if ((MP_PARSE_NODE_IS_LEAF(p) && !MP_PARSE_NODE_IS_ID(p))
|| MP_PARSE_NODE_IS_STRUCT_KIND(p, RULE_const_object)) {
pop_result(&parser); // MP_PARSE_NODE_NULL
pop_result(&parser); // const expression (leaf or RULE_const_object)
// Pushing the "pass" rule here will overwrite any RULE_const_object
// entry that was on the result stack, allowing the GC to reclaim
// the memory from the const object when needed.
push_result_rule(&parser, rule_src_line, RULE_pass_stmt, 0);
break;
}
}
unix-cpy: Remove unix-cpy. It's no longer needed. unix-cpy was originally written to get semantic equivalent with CPython without writing functional tests. When writing the initial implementation of uPy it was a long way between lexer and functional tests, so the half-way test was to make sure that the bytecode was correct. The idea was that if the uPy bytecode matched CPython 1-1 then uPy would be proper Python if the bytecodes acted correctly. And having matching bytecode meant that it was less likely to miss some deep subtlety in the Python semantics that would require an architectural change later on. But that is all history and it no longer makes sense to retain the ability to output CPython bytecode, because: 1. It outputs CPython 3.3 compatible bytecode. CPython's bytecode changes from version to version, and seems to have changed quite a bit in 3.5. There's no point in changing the bytecode output to match CPython anymore. 2. uPy and CPy do different optimisations to the bytecode which makes it harder to match. 3. The bytecode tests are not run. They were never part of Travis and are not run locally anymore. 4. The EMIT_CPYTHON option needs a lot of extra source code which adds heaps of noise, especially in compile.c. 5. Now that there is an extensive test suite (which tests functionality) there is no need to match the bytecode. Some very subtle behaviour is tested with the test suite and passing these tests is a much better way to stay Python-language compliant, rather than trying to match CPy bytecode.
2015-08-14 07:24:11 -04:00
#endif
// count number of arguments for the parse node
i = 0;
size_t num_not_nil = 0;
for (size_t x = n; x > 0;) {
--x;
if ((rule_arg[x] & RULE_ARG_KIND_MASK) == RULE_ARG_TOK) {
mp_token_kind_t tok_kind = rule_arg[x] & RULE_ARG_ARG_MASK;
if (tok_kind == MP_TOKEN_NAME) {
// only tokens which were names are pushed to stack
i += 1;
num_not_nil += 1;
}
} else {
// rules are always pushed
if (peek_result(&parser, i) != MP_PARSE_NODE_NULL) {
num_not_nil += 1;
}
i += 1;
2013-10-04 14:53:11 -04:00
}
}
if (num_not_nil == 1 && (rule_act & RULE_ACT_ALLOW_IDENT)) {
// this rule has only 1 argument and should not be emitted
mp_parse_node_t pn = MP_PARSE_NODE_NULL;
for (size_t x = 0; x < i; ++x) {
mp_parse_node_t pn2 = pop_result(&parser);
if (pn2 != MP_PARSE_NODE_NULL) {
2013-10-04 14:53:11 -04:00
pn = pn2;
}
}
push_result_node(&parser, pn);
} else {
// this rule must be emitted
if (rule_act & RULE_ACT_ADD_BLANK) {
// and add an extra blank node at the end (used by the compiler to store data)
push_result_node(&parser, MP_PARSE_NODE_NULL);
i += 1;
}
push_result_rule(&parser, rule_src_line, rule_id, i);
2013-10-04 14:53:11 -04:00
}
break;
}
2013-10-04 14:53:11 -04:00
default: {
assert((rule_act & RULE_ACT_KIND_MASK) == RULE_ACT_LIST);
2013-10-04 14:53:11 -04:00
// n=2 is: item item*
// n=1 is: item (sep item)*
// n=3 is: item (sep item)* [sep]
bool had_trailing_sep;
2013-10-04 14:53:11 -04:00
if (backtrack) {
list_backtrack:
2013-10-04 14:53:11 -04:00
had_trailing_sep = false;
if (n == 2) {
if (i == 1) {
// fail on item, first time round; propagate backtrack
goto next_rule;
} else {
// fail on item, in later rounds; finish with this rule
backtrack = false;
}
} else {
if (i == 1) {
// fail on item, first time round; propagate backtrack
goto next_rule;
} else if ((i & 1) == 1) {
// fail on item, in later rounds; have eaten tokens so can't backtrack
if (n == 3) {
// list allows trailing separator; finish parsing list
had_trailing_sep = true;
backtrack = false;
} else {
// list doesn't allowing trailing separator; fail
goto syntax_error;
}
} else {
// fail on separator; finish parsing list
backtrack = false;
}
}
} else {
for (;;) {
size_t arg = rule_arg[i & 1 & n];
if ((arg & RULE_ARG_KIND_MASK) == RULE_ARG_TOK) {
if (lex->tok_kind == (arg & RULE_ARG_ARG_MASK)) {
if (i & 1 & n) {
// separators which are tokens are not pushed to result stack
2013-10-04 14:53:11 -04:00
} else {
push_result_token(&parser, rule_id);
2013-10-04 14:53:11 -04:00
}
mp_lexer_to_next(lex);
// got element of list, so continue parsing list
i += 1;
} else {
// couldn't get element of list
i += 1;
backtrack = true;
goto list_backtrack;
}
} else {
assert((arg & RULE_ARG_KIND_MASK) == RULE_ARG_RULE);
push_rule(&parser, rule_src_line, rule_id, i + 1); // save this list-rule
push_rule_from_arg(&parser, arg); // push child of list-rule
goto next_rule;
2013-10-04 14:53:11 -04:00
}
}
}
assert(i >= 1);
// compute number of elements in list, result in i
i -= 1;
if ((n & 1) && (rule_arg[1] & RULE_ARG_KIND_MASK) == RULE_ARG_TOK) {
2013-10-04 14:53:11 -04:00
// don't count separators when they are tokens
i = (i + 1) / 2;
}
if (i == 1) {
// list matched single item
if (had_trailing_sep) {
// if there was a trailing separator, make a list of a single item
push_result_rule(&parser, rule_src_line, rule_id, i);
2013-10-04 14:53:11 -04:00
} else {
// just leave single item on stack (ie don't wrap in a list)
}
} else {
push_result_rule(&parser, rule_src_line, rule_id, i);
2013-10-04 14:53:11 -04:00
}
break;
}
2013-10-04 14:53:11 -04:00
}
}
#if MICROPY_COMP_CONST
mp_map_deinit(&parser.consts);
#endif
// truncate final chunk and link into chain of chunks
if (parser.cur_chunk != NULL) {
(void)m_renew_maybe(byte, parser.cur_chunk,
sizeof(mp_parse_chunk_t) + parser.cur_chunk->alloc,
sizeof(mp_parse_chunk_t) + parser.cur_chunk->union_.used,
false);
parser.cur_chunk->alloc = parser.cur_chunk->union_.used;
parser.cur_chunk->union_.next = parser.tree.chunk;
parser.tree.chunk = parser.cur_chunk;
}
if (
lex->tok_kind != MP_TOKEN_END // check we are at the end of the token stream
|| parser.result_stack_top == 0 // check that we got a node (can fail on empty input)
) {
syntax_error:;
mp_obj_t exc;
if (lex->tok_kind == MP_TOKEN_INDENT) {
exc = mp_obj_new_exception_msg(&mp_type_IndentationError,
MP_ERROR_TEXT("unexpected indent"));
} else if (lex->tok_kind == MP_TOKEN_DEDENT_MISMATCH) {
exc = mp_obj_new_exception_msg(&mp_type_IndentationError,
MP_ERROR_TEXT("unindent doesn't match any outer indent level"));
} else {
exc = mp_obj_new_exception_msg(&mp_type_SyntaxError,
MP_ERROR_TEXT("invalid syntax"));
}
// add traceback to give info about file name and location
// we don't have a 'block' name, so just pass the NULL qstr to indicate this
mp_obj_exception_add_traceback(exc, lex->source_name, lex->tok_line, MP_QSTRnull);
nlr_raise(exc);
}
// get the root parse node that we created
assert(parser.result_stack_top == 1);
parser.tree.root = parser.result_stack[0];
// free the memory that we don't need anymore
m_del(rule_stack_t, parser.rule_stack, parser.rule_stack_alloc);
m_del(mp_parse_node_t, parser.result_stack, parser.result_stack_alloc);
// we also free the lexer on behalf of the caller
mp_lexer_free(lex);
return parser.tree;
2013-10-04 14:53:11 -04:00
}
void mp_parse_tree_clear(mp_parse_tree_t *tree) {
mp_parse_chunk_t *chunk = tree->chunk;
while (chunk != NULL) {
mp_parse_chunk_t *next = chunk->union_.next;
m_del(byte, chunk, sizeof(mp_parse_chunk_t) + chunk->alloc);
chunk = next;
}
}
#endif // MICROPY_ENABLE_COMPILER