mirror of
https://github.com/Keychron/qmk_firmware.git
synced 2024-11-24 09:27:36 +06:00
1a3064afb1
Co-authored-by: Drashna Jaelre <drashna@live.com> Co-authored-by: Ryan <fauxpark@gmail.com>
262 lines
8.2 KiB
C
262 lines
8.2 KiB
C
/*
|
|
This is the modified version of [calculator by MWWorks](https://github.com/MWWorks/mw_calc_numpad/blob/master/calc.c). Below is the quote from [MWWorks](https://github.com/MWWorks).
|
|
|
|
Calculator for QMK-based keyboard by MWWorks, https://mwworks.uk
|
|
This is free, usual disclaimers, don't use it to calculate megaton yields, surgery plans, etc
|
|
|
|
I did not plan to reinvent the wheel for this - I figured surely somebody somewhere has working calculator code?
|
|
Found lots but none that actually work like you expect a calculator to, hence DIYing it
|
|
|
|
As such, this is probably a bit janky, especially as I am a bit of a hack at C
|
|
Seems to be working well, with occasional glitchs, solved by clearing it
|
|
And some occasional floating-point issues - eg get a long decimal rather than the whole number you were expecting
|
|
Feel free to fix it! I think it needs to detect the precision of the two operands and then figure out what the precision of the result should be
|
|
|
|
*/
|
|
#include "rubi.h"
|
|
|
|
static uint8_t calc_current_operand = 0;
|
|
static char calc_operand_0[CALC_DIGITS+1] = "";
|
|
static char calc_operand_1[CALC_DIGITS+1] = "";
|
|
char calc_result[CALC_DIGITS+1] = "";
|
|
static char calc_status[CALC_DIGITS+1] = "";
|
|
static char calc_operator = ' ';
|
|
static bool calc_reset = false;
|
|
|
|
|
|
void calcBegin(void){
|
|
}
|
|
|
|
//update display
|
|
void calcUpdate(void){
|
|
if (calc_display_lines == 2) {
|
|
if((calc_current_operand == 1) || (calc_reset)){
|
|
strcpy(calc_status, calc_operand_0);
|
|
if((strlen(calc_operand_0)>0) || (strlen(calc_operand_1)>0)){
|
|
uint8_t len = strlen(calc_status);
|
|
if (!(calc_operator == 's' || calc_operator == 'r' || calc_operator == 'n')) {
|
|
calc_status[len] = calc_operator;
|
|
}
|
|
calc_status[len+1] = 0;
|
|
if(calc_reset
|
|
&& !(calc_operator == 's' || calc_operator == 'r' || calc_operator == 'n')){
|
|
strncat(calc_status, calc_operand_1, CALC_DIGITS-strlen(calc_status));
|
|
calc_operator = ' ';
|
|
}
|
|
}
|
|
strcpy(calc_status_display, calc_status);
|
|
}
|
|
} else if (calc_display_lines == 1) {
|
|
if(calc_reset
|
|
&& !(calc_operator == 's' || calc_operator == 'r' || calc_operator == 'n')){
|
|
calc_operator = ' ';
|
|
}
|
|
}
|
|
calc_operator_display = calc_operator;
|
|
strcpy(calc_result_display, calc_result);
|
|
}
|
|
|
|
//perform calculation on the 2 operands
|
|
void calcOperands(void){
|
|
float result = 0;
|
|
switch (calc_operator){
|
|
|
|
//standard operators
|
|
case '+':
|
|
result = strtod(calc_operand_0, NULL) + strtod(calc_operand_1, NULL);
|
|
break;
|
|
|
|
case '-':
|
|
result = strtod(calc_operand_0, NULL) - strtod(calc_operand_1, NULL);
|
|
break;
|
|
|
|
case '/':
|
|
result = strtod(calc_operand_0, NULL) / strtod(calc_operand_1, NULL);
|
|
break;
|
|
|
|
case '*':
|
|
result = strtod(calc_operand_0, NULL) * strtod(calc_operand_1, NULL);
|
|
break;
|
|
|
|
//single operand operators - these are all in 2
|
|
case 's':
|
|
result = sqrt(strtod(calc_operand_0, NULL));
|
|
break;
|
|
|
|
case 'r':
|
|
result = 1/(strtod(calc_operand_0, NULL));
|
|
break;
|
|
|
|
}
|
|
|
|
//now convert the float result into a string
|
|
//we know the total string size but we need to find the size of the integer component to know how much we have for decimals
|
|
uint8_t magnitude = ceil(log10(result));
|
|
uint8_t max_decimals = CALC_DIGITS-magnitude-1;
|
|
//but max it at 7 because that seems the useful limit of our floats
|
|
if(max_decimals>7){
|
|
max_decimals = 7;
|
|
}
|
|
dtostrf(result, CALC_DIGITS, max_decimals, calc_result);
|
|
|
|
//now to clean up the result - we need it clean as it may be the input of next calculation
|
|
//this seems a lot of code to format this string :| note that this c doesn't support float in sprintf
|
|
uint8_t i;
|
|
|
|
//first find if theres a dot
|
|
uint8_t dotpos = CALC_DIGITS+1;
|
|
for(i=0; i<strlen(calc_result); i++){
|
|
if(calc_result[i] == '.'){
|
|
dotpos = i;
|
|
break;
|
|
}
|
|
}
|
|
|
|
//if there is, work back to it and remove trailing 0 or .
|
|
if(dotpos>=0){
|
|
for(i=strlen(calc_result)-1; i>=dotpos; i--){
|
|
if((calc_result[i] == '0') || (calc_result[i] == '.')){
|
|
calc_result[i] = 0;
|
|
}else{
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
//now find how many leading spaces
|
|
uint8_t spaces = 0;
|
|
for(i=0; i<strlen(calc_result); i++){
|
|
if(calc_result[i] == ' '){
|
|
spaces++;
|
|
}else{
|
|
break;
|
|
}
|
|
}
|
|
|
|
//and shift the string
|
|
for(i=0; i<strlen(calc_result)-spaces; i++){
|
|
calc_result[i] = calc_result[i+spaces];
|
|
}
|
|
calc_result[strlen(calc_result)-spaces] = 0;
|
|
|
|
calcUpdate();
|
|
//the result is available as the first operand for another calculation
|
|
strcpy(calc_operand_0, calc_result);
|
|
calc_operand_1[0] = 0;
|
|
|
|
}
|
|
|
|
void calcInput(char input){
|
|
char *operand = calc_operand_0;
|
|
if(calc_current_operand == 1){
|
|
operand = calc_operand_1;
|
|
}
|
|
uint8_t len = strlen(operand);
|
|
|
|
if(
|
|
((input >= 48) && (input <= 57)) ||
|
|
(input == '.')
|
|
){
|
|
//if this is following an equals, then we start from scratch as if new calculation
|
|
if(calc_reset == true){
|
|
calc_reset = false;
|
|
calc_current_operand = 0;
|
|
calc_operand_0[0] = 0;
|
|
calc_operand_1[0] = 0;
|
|
operand = calc_operand_0;
|
|
len = 0;
|
|
}
|
|
|
|
if(len<CALC_DIGITS){
|
|
operand[len] = input;
|
|
operand[len+1] = 0;
|
|
strcpy(calc_result, operand);
|
|
calcUpdate();
|
|
}
|
|
|
|
//special input to backspace
|
|
}else if(input == 'x'){
|
|
operand[len-1] = 0;
|
|
strcpy(calc_result, operand);
|
|
calcUpdate();
|
|
|
|
//clear
|
|
}else if(input == 'c'){
|
|
operand[0] = 0;
|
|
calc_operand_0[0] = 0;
|
|
calc_operand_1[0] = 0;
|
|
calc_operator = ' ';
|
|
calc_reset = true;
|
|
strcpy(calc_result, operand);
|
|
calcUpdate();
|
|
|
|
//special input switch neg/pos
|
|
}else if((input == 'n') && (len>0)){
|
|
uint8_t i;
|
|
|
|
if(operand[0] == '-'){
|
|
for(i=1; i<=len; i++){
|
|
operand[i-1] = operand[i];
|
|
}
|
|
}else if(len<CALC_DIGITS){
|
|
for(i=0; i<=len; i++){
|
|
operand[len-i+1] = operand[len-i];
|
|
}
|
|
operand[0] = '-';
|
|
}
|
|
calc_operator = input;
|
|
strcpy(calc_result, operand);
|
|
calcUpdate();
|
|
|
|
|
|
//standard 2 operand operators
|
|
}else if((input == '+') || (input == '-') || (input == '*') || (input == '/')){
|
|
|
|
//get ready for second operand
|
|
if(calc_current_operand == 0){
|
|
calc_operator = input;
|
|
calc_current_operand = 1;
|
|
calcUpdate();
|
|
|
|
//we pressed = we now expect a new second operand
|
|
}else if(calc_reset){
|
|
calc_operator = input;
|
|
calc_reset = false;
|
|
calc_operand_1[0] = 0;
|
|
calcUpdate();
|
|
|
|
}else {
|
|
//if we use this on the second operand, calculate first, then ready for a second operand again
|
|
if (strlen(calc_operand_1)>0){
|
|
calcOperands();
|
|
}
|
|
calc_operand_1[0] = 0;
|
|
calc_operator = input;
|
|
calcUpdate();
|
|
}
|
|
|
|
|
|
}else if(input == '='){
|
|
//only accept = if we are on the second operand
|
|
if(calc_current_operand == 1){
|
|
//keep the second operand for a subsequent press of =; but flag to reset if start entry of new operand
|
|
calc_reset = true;
|
|
calcOperands();
|
|
}
|
|
|
|
//single operands - square root and reciprocal - needs to operate on 0 so it works after a previous = result
|
|
}else if((input == 's') || (input == 'r')){
|
|
//but maybe we started entering 1
|
|
if(calc_current_operand == 1 && !calc_reset){
|
|
strcpy(calc_operand_0, calc_operand_1);
|
|
}
|
|
calc_current_operand = 1;
|
|
calc_operand_1[0] = 0;
|
|
calc_operator = input;
|
|
calc_reset = true; //simulate another =
|
|
calcOperands();
|
|
|
|
}
|
|
|
|
}
|