Make G26 compatible with inches and thermal unit modes

This commit is contained in:
Scott Lahteine 2017-05-03 17:10:44 -05:00
parent 241bdffe65
commit 1b2c7ec20a
3 changed files with 29 additions and 23 deletions

View File

@ -126,6 +126,8 @@
void set_destination_to_current();
void set_current_to_destination();
float code_value_float();
float code_value_linear_units();
float code_value_axis_units(const AxisEnum axis);
bool code_value_bool();
bool code_has_value();
void lcd_init();
@ -164,10 +166,11 @@
filament_diameter = FILAMENT,
prime_length = PRIME_LENGTH,
x_pos, y_pos,
bed_temp = BED_TEMP,
hotend_temp = HOTEND_TEMP,
ooze_amount = OOZE_AMOUNT;
static int16_t bed_temp = BED_TEMP,
hotend_temp = HOTEND_TEMP;
static int8_t prime_flag = 0;
static bool keep_heaters_on = false;
@ -379,9 +382,9 @@
if (!keep_heaters_on) {
#if HAS_TEMP_BED
thermalManager.setTargetBed(0.0);
thermalManager.setTargetBed(0);
#endif
thermalManager.setTargetHotend(0.0, 0);
thermalManager.setTargetHotend(0, 0);
}
}
@ -640,8 +643,8 @@
keep_heaters_on = false;
if (code_seen('B')) {
bed_temp = code_value_float();
if (!WITHIN(bed_temp, 15.0, 140.0)) {
bed_temp = code_value_temp_abs();
if (!WITHIN(bed_temp, 15, 140)) {
SERIAL_PROTOCOLLNPGM("?Specified bed temperature not plausible.");
return UBL_ERR;
}
@ -650,7 +653,7 @@
if (code_seen('C')) continue_with_closest++;
if (code_seen('L')) {
layer_height = code_value_float();
layer_height = code_value_linear_units();
if (!WITHIN(layer_height, 0.0, 2.0)) {
SERIAL_PROTOCOLLNPGM("?Specified layer height not plausible.");
return UBL_ERR;
@ -682,14 +685,14 @@
if (code_seen('K')) keep_heaters_on++;
if (code_seen('O') && code_has_value())
ooze_amount = code_value_float();
ooze_amount = code_value_linear_units();
if (code_seen('P')) {
if (!code_has_value())
prime_flag = -1;
else {
prime_flag++;
prime_length = code_value_float();
prime_length = code_value_linear_units();
if (!WITHIN(prime_length, 0.0, 25.0)) {
SERIAL_PROTOCOLLNPGM("?Specified prime length not plausible.");
return UBL_ERR;
@ -698,7 +701,7 @@
}
if (code_seen('F')) {
filament_diameter = code_value_float();
filament_diameter = code_value_linear_units();
if (!WITHIN(filament_diameter, 1.0, 4.0)) {
SERIAL_PROTOCOLLNPGM("?Specified filament size not plausible.");
return UBL_ERR;
@ -711,8 +714,8 @@
extrusion_multiplier *= filament_diameter * sq(nozzle) / sq(0.3); // Scale up by nozzle size
if (code_seen('H')) {
hotend_temp = code_value_float();
if (!WITHIN(hotend_temp, 165.0, 280.0)) {
hotend_temp = code_value_temp_abs();
if (!WITHIN(hotend_temp, 165, 280)) {
SERIAL_PROTOCOLLNPGM("?Specified nozzle temperature not plausible.");
return UBL_ERR;
}
@ -727,7 +730,7 @@
y_pos = current_position[Y_AXIS];
if (code_seen('X')) {
x_pos = code_value_float();
x_pos = code_value_axis_units(X_AXIS);
if (!WITHIN(x_pos, X_MIN_POS, X_MAX_POS)) {
SERIAL_PROTOCOLLNPGM("?Specified X coordinate not plausible.");
return UBL_ERR;
@ -736,7 +739,7 @@
else
if (code_seen('Y')) {
y_pos = code_value_float();
y_pos = code_value_axis_units(Y_AXIS);
if (!WITHIN(y_pos, Y_MIN_POS, Y_MAX_POS)) {
SERIAL_PROTOCOLLNPGM("?Specified Y coordinate not plausible.");
return UBL_ERR;

View File

@ -290,8 +290,18 @@ extern float soft_endstop_min[XYZ], soft_endstop_max[XYZ];
// GCode support for external objects
bool code_seen(char);
int code_value_int();
float code_value_temp_abs();
float code_value_temp_diff();
int16_t code_value_temp_abs();
int16_t code_value_temp_diff();
#if ENABLED(INCH_MODE_SUPPORT)
float code_value_linear_units();
float code_value_axis_units(const AxisEnum axis);
float code_value_per_axis_unit(const AxisEnum axis);
#else
#define code_value_linear_units() code_value_float()
#define code_value_axis_units(A) code_value_float()
#define code_value_per_axis_unit(A) code_value_float()
#endif
#if IS_KINEMATIC
extern float delta[ABC];

View File

@ -1292,13 +1292,6 @@ inline bool code_value_bool() { return !code_has_value() || code_value_byte() >
inline float code_value_linear_units() { return code_value_float() * linear_unit_factor; }
inline float code_value_axis_units(const AxisEnum axis) { return code_value_float() * axis_unit_factor(axis); }
inline float code_value_per_axis_unit(const AxisEnum axis) { return code_value_float() / axis_unit_factor(axis); }
#else
#define code_value_linear_units() code_value_float()
#define code_value_axis_units(A) code_value_float()
#define code_value_per_axis_unit(A) code_value_float()
#endif
#if ENABLED(TEMPERATURE_UNITS_SUPPORT)