443 lines
16 KiB
C++
443 lines
16 KiB
C++
/**
|
|
* Marlin 3D Printer Firmware
|
|
* Copyright (c) 2020 MarlinFirmware [https://github.com/MarlinFirmware/Marlin]
|
|
*
|
|
* Based on Sprinter and grbl.
|
|
* Copyright (c) 2011 Camiel Gubbels / Erik van der Zalm
|
|
*
|
|
* This program is free software: you can redistribute it and/or modify
|
|
* it under the terms of the GNU General Public License as published by
|
|
* the Free Software Foundation, either version 3 of the License, or
|
|
* (at your option) any later version.
|
|
*
|
|
* This program is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with this program. If not, see <https://www.gnu.org/licenses/>.
|
|
*
|
|
*/
|
|
#pragma once
|
|
|
|
/**
|
|
* parser.h - Parser for a GCode line, providing a parameter interface.
|
|
* Codes like M149 control the way the GCode parser behaves,
|
|
* so settings for these codes are located in this class.
|
|
*/
|
|
|
|
#include "../inc/MarlinConfig.h"
|
|
|
|
//#define DEBUG_GCODE_PARSER
|
|
#if ENABLED(DEBUG_GCODE_PARSER)
|
|
#include "../libs/hex_print.h"
|
|
#endif
|
|
|
|
#if ENABLED(TEMPERATURE_UNITS_SUPPORT)
|
|
typedef enum : uint8_t { TEMPUNIT_C, TEMPUNIT_K, TEMPUNIT_F } TempUnit;
|
|
#endif
|
|
|
|
#if ENABLED(INCH_MODE_SUPPORT)
|
|
typedef enum : uint8_t { LINEARUNIT_MM, LINEARUNIT_INCH } LinearUnit;
|
|
#endif
|
|
|
|
/**
|
|
* GCode parser
|
|
*
|
|
* - Parse a single gcode line for its letter, code, subcode, and parameters
|
|
* - FASTER_GCODE_PARSER:
|
|
* - Flags existing params (1 bit each)
|
|
* - Stores value offsets (1 byte each)
|
|
* - Provide accessors for parameters:
|
|
* - Parameter exists
|
|
* - Parameter has value
|
|
* - Parameter value in different units and types
|
|
*/
|
|
class GCodeParser {
|
|
|
|
private:
|
|
static char *value_ptr; // Set by seen, used to fetch the value
|
|
|
|
#if ENABLED(FASTER_GCODE_PARSER)
|
|
static uint32_t codebits; // Parameters pre-scanned
|
|
static uint8_t param[26]; // For A-Z, offsets into command args
|
|
#else
|
|
static char *command_args; // Args start here, for slow scan
|
|
#endif
|
|
|
|
public:
|
|
|
|
// Global states for GCode-level units features
|
|
|
|
static bool volumetric_enabled;
|
|
|
|
#if ENABLED(INCH_MODE_SUPPORT)
|
|
static float linear_unit_factor, volumetric_unit_factor;
|
|
#endif
|
|
|
|
#if ENABLED(TEMPERATURE_UNITS_SUPPORT)
|
|
static TempUnit input_temp_units;
|
|
#endif
|
|
|
|
// Command line state
|
|
static char *command_ptr, // The command, so it can be echoed
|
|
*string_arg, // string of command line
|
|
command_letter; // G, M, or T
|
|
static uint16_t codenum; // 123
|
|
#if USE_GCODE_SUBCODES
|
|
static uint8_t subcode; // .1
|
|
#endif
|
|
|
|
#if ENABLED(GCODE_MOTION_MODES)
|
|
static int16_t motion_mode_codenum;
|
|
#if USE_GCODE_SUBCODES
|
|
static uint8_t motion_mode_subcode;
|
|
#endif
|
|
FORCE_INLINE static void cancel_motion_mode() { motion_mode_codenum = -1; }
|
|
#endif
|
|
|
|
#if ENABLED(DEBUG_GCODE_PARSER)
|
|
static void debug();
|
|
#endif
|
|
|
|
// Reset is done before parsing
|
|
static void reset();
|
|
|
|
#define LETTER_BIT(N) ((N) - 'A')
|
|
|
|
FORCE_INLINE static bool valid_signless(const char * const p) {
|
|
return NUMERIC(p[0]) || (p[0] == '.' && NUMERIC(p[1])); // .?[0-9]
|
|
}
|
|
|
|
FORCE_INLINE static bool valid_float(const char * const p) {
|
|
return valid_signless(p) || ((p[0] == '-' || p[0] == '+') && valid_signless(&p[1])); // [-+]?.?[0-9]
|
|
}
|
|
|
|
FORCE_INLINE static bool valid_number(const char * const p) {
|
|
// TODO: With MARLIN_DEV_MODE allow HEX values starting with "x"
|
|
return valid_float(p);
|
|
}
|
|
|
|
#if ENABLED(FASTER_GCODE_PARSER)
|
|
|
|
FORCE_INLINE static bool valid_int(const char * const p) {
|
|
return NUMERIC(p[0]) || ((p[0] == '-' || p[0] == '+') && NUMERIC(p[1])); // [-+]?[0-9]
|
|
}
|
|
|
|
// Set the flag and pointer for a parameter
|
|
static inline void set(const char c, char * const ptr) {
|
|
const uint8_t ind = LETTER_BIT(c);
|
|
if (ind >= COUNT(param)) return; // Only A-Z
|
|
SBI32(codebits, ind); // parameter exists
|
|
param[ind] = ptr ? ptr - command_ptr : 0; // parameter offset or 0
|
|
#if ENABLED(DEBUG_GCODE_PARSER)
|
|
if (codenum == 800) {
|
|
SERIAL_ECHOPGM("Set bit ", ind, " of codebits (", hex_address((void*)(codebits >> 16)));
|
|
print_hex_word((uint16_t)(codebits & 0xFFFF));
|
|
SERIAL_ECHOLNPGM(") | param = ", param[ind]);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
// Code seen bit was set. If not found, value_ptr is unchanged.
|
|
// This allows "if (seen('A')||seen('B'))" to use the last-found value.
|
|
static inline bool seen(const char c) {
|
|
const uint8_t ind = LETTER_BIT(c);
|
|
if (ind >= COUNT(param)) return false; // Only A-Z
|
|
const bool b = TEST32(codebits, ind);
|
|
if (b) {
|
|
if (param[ind]) {
|
|
char * const ptr = command_ptr + param[ind];
|
|
value_ptr = valid_number(ptr) ? ptr : nullptr;
|
|
}
|
|
else
|
|
value_ptr = nullptr;
|
|
}
|
|
return b;
|
|
}
|
|
|
|
FORCE_INLINE static constexpr uint32_t letter_bits(const char * const str) {
|
|
return (str[0] ? _BV32(LETTER_BIT(str[0])) |
|
|
(str[1] ? _BV32(LETTER_BIT(str[1])) |
|
|
(str[2] ? _BV32(LETTER_BIT(str[2])) |
|
|
(str[3] ? _BV32(LETTER_BIT(str[3])) |
|
|
(str[4] ? _BV32(LETTER_BIT(str[4])) |
|
|
(str[5] ? _BV32(LETTER_BIT(str[5])) |
|
|
(str[6] ? _BV32(LETTER_BIT(str[6])) |
|
|
(str[7] ? _BV32(LETTER_BIT(str[7])) |
|
|
(str[8] ? _BV32(LETTER_BIT(str[8])) |
|
|
(str[9] ? _BV32(LETTER_BIT(str[9]))
|
|
: 0) : 0) : 0) : 0) : 0) : 0) : 0) : 0) : 0) : 0);
|
|
}
|
|
|
|
// At least one of a list of code letters was seen
|
|
#ifdef CPU_32_BIT
|
|
FORCE_INLINE static bool seen(const char * const str) { return !!(codebits & letter_bits(str)); }
|
|
#else
|
|
FORCE_INLINE static bool seen(const char * const str) {
|
|
const uint32_t letrbits = letter_bits(str);
|
|
const uint8_t * const cb = (uint8_t*)&codebits;
|
|
const uint8_t * const lb = (uint8_t*)&letrbits;
|
|
return (cb[0] & lb[0]) || (cb[1] & lb[1]) || (cb[2] & lb[2]) || (cb[3] & lb[3]);
|
|
}
|
|
#endif
|
|
|
|
static inline bool seen_any() { return !!codebits; }
|
|
|
|
FORCE_INLINE static bool seen_test(const char c) { return TEST32(codebits, LETTER_BIT(c)); }
|
|
|
|
#else // !FASTER_GCODE_PARSER
|
|
|
|
#if ENABLED(GCODE_CASE_INSENSITIVE)
|
|
FORCE_INLINE static char* strgchr(char *p, char g) {
|
|
auto uppercase = [](char c) {
|
|
return c + (WITHIN(c, 'a', 'z') ? 'A' - 'a' : 0);
|
|
};
|
|
const char d = uppercase(g);
|
|
for (char cc; (cc = uppercase(*p)); p++) if (cc == d) return p;
|
|
return nullptr;
|
|
}
|
|
#else
|
|
#define strgchr strchr
|
|
#endif
|
|
|
|
// Code is found in the string. If not found, value_ptr is unchanged.
|
|
// This allows "if (seen('A')||seen('B'))" to use the last-found value.
|
|
static inline bool seen(const char c) {
|
|
char *p = strgchr(command_args, c);
|
|
const bool b = !!p;
|
|
if (b) value_ptr = valid_number(&p[1]) ? &p[1] : nullptr;
|
|
return b;
|
|
}
|
|
|
|
static inline bool seen_any() { return *command_args == '\0'; }
|
|
|
|
FORCE_INLINE static bool seen_test(const char c) { return (bool)strgchr(command_args, c); }
|
|
|
|
// At least one of a list of code letters was seen
|
|
static inline bool seen(const char * const str) {
|
|
for (uint8_t i = 0; const char c = str[i]; i++)
|
|
if (seen_test(c)) return true;
|
|
return false;
|
|
}
|
|
|
|
#endif // !FASTER_GCODE_PARSER
|
|
|
|
// Seen any axis parameter
|
|
static inline bool seen_axis() { return seen(LOGICAL_AXES_STRING); }
|
|
|
|
#if ENABLED(GCODE_QUOTED_STRINGS)
|
|
static char* unescape_string(char* &src);
|
|
#else
|
|
FORCE_INLINE static char* unescape_string(char* &src) { return src; }
|
|
#endif
|
|
|
|
// Populate all fields by parsing a single line of GCode
|
|
// This uses 54 bytes of SRAM to speed up seen/value
|
|
static void parse(char * p);
|
|
|
|
#if ENABLED(CNC_COORDINATE_SYSTEMS)
|
|
// Parse the next parameter as a new command
|
|
static bool chain();
|
|
#endif
|
|
|
|
// Test whether the parsed command matches the input
|
|
static inline bool is_command(const char ltr, const uint16_t num) { return command_letter == ltr && codenum == num; }
|
|
|
|
// The code value pointer was set
|
|
FORCE_INLINE static bool has_value() { return !!value_ptr; }
|
|
|
|
// Seen a parameter with a value
|
|
static inline bool seenval(const char c) { return seen(c) && has_value(); }
|
|
|
|
// The value as a string
|
|
static inline char* value_string() { return value_ptr; }
|
|
|
|
// Float removes 'E' to prevent scientific notation interpretation
|
|
static inline float value_float() {
|
|
if (value_ptr) {
|
|
char *e = value_ptr;
|
|
for (;;) {
|
|
const char c = *e;
|
|
if (c == '\0' || c == ' ') break;
|
|
if (c == 'E' || c == 'e') {
|
|
*e = '\0';
|
|
const float ret = strtof(value_ptr, nullptr);
|
|
*e = c;
|
|
return ret;
|
|
}
|
|
++e;
|
|
}
|
|
return strtof(value_ptr, nullptr);
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
// Code value as a long or ulong
|
|
static inline int32_t value_long() { return value_ptr ? strtol(value_ptr, nullptr, 10) : 0L; }
|
|
static inline uint32_t value_ulong() { return value_ptr ? strtoul(value_ptr, nullptr, 10) : 0UL; }
|
|
|
|
// Code value for use as time
|
|
static inline millis_t value_millis() { return value_ulong(); }
|
|
static inline millis_t value_millis_from_seconds() { return (millis_t)SEC_TO_MS(value_float()); }
|
|
|
|
// Reduce to fewer bits
|
|
static inline int16_t value_int() { return (int16_t)value_long(); }
|
|
static inline uint16_t value_ushort() { return (uint16_t)value_long(); }
|
|
static inline uint8_t value_byte() { return (uint8_t)constrain(value_long(), 0, 255); }
|
|
|
|
// Bool is true with no value or non-zero
|
|
static inline bool value_bool() { return !has_value() || !!value_byte(); }
|
|
|
|
// Units modes: Inches, Fahrenheit, Kelvin
|
|
|
|
#if ENABLED(INCH_MODE_SUPPORT)
|
|
static inline float mm_to_linear_unit(const_float_t mm) { return mm / linear_unit_factor; }
|
|
static inline float mm_to_volumetric_unit(const_float_t mm) { return mm / (volumetric_enabled ? volumetric_unit_factor : linear_unit_factor); }
|
|
|
|
// Init linear units by constructor
|
|
GCodeParser() { set_input_linear_units(LINEARUNIT_MM); }
|
|
|
|
static inline void set_input_linear_units(const LinearUnit units) {
|
|
switch (units) {
|
|
default:
|
|
case LINEARUNIT_MM: linear_unit_factor = 1.0f; break;
|
|
case LINEARUNIT_INCH: linear_unit_factor = 25.4f; break;
|
|
}
|
|
volumetric_unit_factor = POW(linear_unit_factor, 3);
|
|
}
|
|
|
|
static inline float axis_unit_factor(const AxisEnum axis) {
|
|
return (
|
|
#if HAS_EXTRUDERS
|
|
axis >= E_AXIS && volumetric_enabled ? volumetric_unit_factor : linear_unit_factor
|
|
#else
|
|
linear_unit_factor
|
|
#endif
|
|
);
|
|
}
|
|
|
|
static inline float linear_value_to_mm(const_float_t v) { return v * linear_unit_factor; }
|
|
static inline float axis_value_to_mm(const AxisEnum axis, const float v) { return v * axis_unit_factor(axis); }
|
|
static inline float per_axis_value(const AxisEnum axis, const float v) { return v / axis_unit_factor(axis); }
|
|
|
|
#else
|
|
|
|
static inline float mm_to_linear_unit(const_float_t mm) { return mm; }
|
|
static inline float mm_to_volumetric_unit(const_float_t mm) { return mm; }
|
|
|
|
static inline float linear_value_to_mm(const_float_t v) { return v; }
|
|
static inline float axis_value_to_mm(const AxisEnum, const float v) { return v; }
|
|
static inline float per_axis_value(const AxisEnum, const float v) { return v; }
|
|
|
|
#endif
|
|
|
|
static inline bool using_inch_units() { return mm_to_linear_unit(1.0f) != 1.0f; }
|
|
|
|
#define IN_TO_MM(I) ((I) * 25.4f)
|
|
#define MM_TO_IN(M) ((M) / 25.4f)
|
|
#define LINEAR_UNIT(V) parser.mm_to_linear_unit(V)
|
|
#define VOLUMETRIC_UNIT(V) parser.mm_to_volumetric_unit(V)
|
|
|
|
static inline float value_linear_units() { return linear_value_to_mm(value_float()); }
|
|
static inline float value_axis_units(const AxisEnum axis) { return axis_value_to_mm(axis, value_float()); }
|
|
static inline float value_per_axis_units(const AxisEnum axis) { return per_axis_value(axis, value_float()); }
|
|
|
|
#if ENABLED(TEMPERATURE_UNITS_SUPPORT)
|
|
|
|
static inline void set_input_temp_units(const TempUnit units) { input_temp_units = units; }
|
|
|
|
static inline char temp_units_code() {
|
|
return input_temp_units == TEMPUNIT_K ? 'K' : input_temp_units == TEMPUNIT_F ? 'F' : 'C';
|
|
}
|
|
static inline FSTR_P temp_units_name() {
|
|
return input_temp_units == TEMPUNIT_K ? F("Kelvin") : input_temp_units == TEMPUNIT_F ? F("Fahrenheit") : F("Celsius");
|
|
}
|
|
|
|
#if HAS_LCD_MENU && DISABLED(DISABLE_M503)
|
|
|
|
static inline float to_temp_units(celsius_t c) {
|
|
switch (input_temp_units) {
|
|
default:
|
|
case TEMPUNIT_C: return c;
|
|
case TEMPUNIT_K: return c + 273.15f;
|
|
case TEMPUNIT_F: return c * 0.5555555556f + 32;
|
|
}
|
|
}
|
|
|
|
#endif // HAS_LCD_MENU && !DISABLE_M503
|
|
|
|
static inline celsius_t value_celsius() {
|
|
float f = value_float();
|
|
switch (input_temp_units) {
|
|
default:
|
|
case TEMPUNIT_C: break;
|
|
case TEMPUNIT_K: f -= 273.15f;
|
|
case TEMPUNIT_F: f = (f - 32) * 0.5555555556f;
|
|
}
|
|
return LROUND(f);
|
|
}
|
|
|
|
static inline celsius_t value_celsius_diff() {
|
|
float f = value_float();
|
|
switch (input_temp_units) {
|
|
default:
|
|
case TEMPUNIT_C:
|
|
case TEMPUNIT_K: break;
|
|
case TEMPUNIT_F: f *= 0.5555555556f;
|
|
}
|
|
return LROUND(f);
|
|
}
|
|
|
|
#else // !TEMPERATURE_UNITS_SUPPORT
|
|
|
|
static inline float to_temp_units(int16_t c) { return (float)c; }
|
|
|
|
static inline celsius_t value_celsius() { return value_int(); }
|
|
static inline celsius_t value_celsius_diff() { return value_int(); }
|
|
|
|
#endif // !TEMPERATURE_UNITS_SUPPORT
|
|
|
|
static inline feedRate_t value_feedrate() { return MMM_TO_MMS(value_linear_units()); }
|
|
|
|
void unknown_command_warning();
|
|
|
|
// Provide simple value accessors with default option
|
|
static inline char* stringval(const char c, char * const dval=nullptr) { return seenval(c) ? value_string() : dval; }
|
|
static inline float floatval(const char c, const float dval=0.0) { return seenval(c) ? value_float() : dval; }
|
|
static inline bool boolval(const char c, const bool dval=false) { return seenval(c) ? value_bool() : (seen(c) ? true : dval); }
|
|
static inline uint8_t byteval(const char c, const uint8_t dval=0) { return seenval(c) ? value_byte() : dval; }
|
|
static inline int16_t intval(const char c, const int16_t dval=0) { return seenval(c) ? value_int() : dval; }
|
|
static inline uint16_t ushortval(const char c, const uint16_t dval=0) { return seenval(c) ? value_ushort() : dval; }
|
|
static inline int32_t longval(const char c, const int32_t dval=0) { return seenval(c) ? value_long() : dval; }
|
|
static inline uint32_t ulongval(const char c, const uint32_t dval=0) { return seenval(c) ? value_ulong() : dval; }
|
|
static inline float linearval(const char c, const float dval=0) { return seenval(c) ? value_linear_units() : dval; }
|
|
static inline float axisunitsval(const char c, const AxisEnum a, const float dval=0)
|
|
{ return seenval(c) ? value_axis_units(a) : dval; }
|
|
static inline celsius_t celsiusval(const char c, const celsius_t dval=0) { return seenval(c) ? value_celsius() : dval; }
|
|
static inline feedRate_t feedrateval(const char c, const feedRate_t dval=0) { return seenval(c) ? value_feedrate() : dval; }
|
|
|
|
#if ENABLED(MARLIN_DEV_MODE)
|
|
|
|
static inline uint8_t* hex_adr_val(const char c, uint8_t * const dval=nullptr) {
|
|
if (!seen(c) || *value_ptr != 'x') return dval;
|
|
uint8_t *out = nullptr;
|
|
for (char *vp = value_ptr + 1; HEXCHR(*vp) >= 0; vp++)
|
|
out = (uint8_t*)((uintptr_t(out) << 8) | HEXCHR(*vp));
|
|
return out;
|
|
}
|
|
|
|
static inline uint16_t hex_val(const char c, uint16_t const dval=0) {
|
|
if (!seen(c) || *value_ptr != 'x') return dval;
|
|
uint16_t out = 0;
|
|
for (char *vp = value_ptr + 1; HEXCHR(*vp) >= 0; vp++)
|
|
out = ((out) << 8) | HEXCHR(*vp);
|
|
return out;
|
|
}
|
|
|
|
#endif
|
|
};
|
|
|
|
extern GCodeParser parser;
|