mirror of
https://github.com/qmk/qmk_firmware.git
synced 2024-12-25 10:59:54 +00:00
231 lines
10 KiB
Python
231 lines
10 KiB
Python
|
#!/usr/bin/env python3
|
||
|
|
||
|
import json
|
||
|
from functools import reduce
|
||
|
from chord import *
|
||
|
import sys
|
||
|
|
||
|
comma_separator = (lambda x, y: str(x) + ", " + str(y))
|
||
|
string_sum = (lambda x, y: str(x) + " + " + str(y))
|
||
|
newline_separator = (lambda x, y: str(x) + "\n" + str(y))
|
||
|
|
||
|
def add_includes(data):
|
||
|
output_buffer = ""
|
||
|
if not ("do_not_include_QMK" in data["parameters"] and data["parameters"]["do_not_include_QMK"] == True):
|
||
|
output_buffer += "#include QMK_KEYBOARD_H\n"
|
||
|
if len(data["extra_dependencies"]) > 0:
|
||
|
for dependecy in data["extra_dependencies"]:
|
||
|
output_buffer += '#include "' + dependecy + '"\n'
|
||
|
|
||
|
return output_buffer + "\n"
|
||
|
|
||
|
def add_parameters(data):
|
||
|
output_buffer = ""
|
||
|
|
||
|
number_of_keys = len(data["keys"])
|
||
|
if number_of_keys <= 8:
|
||
|
hash_type = "uint8_t"
|
||
|
elif number_of_keys <= 16:
|
||
|
hash_type = "uint16_t"
|
||
|
elif number_of_keys <= 32:
|
||
|
hash_type = "uint32_t"
|
||
|
elif number_of_keys <= 64:
|
||
|
hash_type = "uint64_t"
|
||
|
else:
|
||
|
raise Exception("The engine currently supports only up to 64 keys.")
|
||
|
|
||
|
output_buffer += "#define CHORD_TIMEOUT " + str(data["parameters"]["chord_timeout"]) + "\n"
|
||
|
output_buffer += "#define DANCE_TIMEOUT " + str(data["parameters"]["dance_timeout"]) + "\n"
|
||
|
output_buffer += "#define LEADER_TIMEOUT " + str(data["parameters"]["leader_timeout"]) + "\n"
|
||
|
output_buffer += "#define TAP_TIMEOUT " + str(data["parameters"]["tap_timeout"]) + "\n"
|
||
|
output_buffer += "#define LONG_PRESS_MULTIPLIER " + str(data["parameters"]["long_press_multiplier"]) + "\n"
|
||
|
output_buffer += "#define DYNAMIC_MACRO_MAX_LENGTH " + str(data["parameters"]["dynamic_macro_max_length"]) + "\n"
|
||
|
output_buffer += "#define COMMAND_MAX_LENGTH " + str(data["parameters"]["command_max_length"]) + "\n"
|
||
|
output_buffer += "#define STRING_MAX_LENGTH " + str(data["parameters"]["string_max_length"]) + "\n"
|
||
|
output_buffer += "#define LEADER_MAX_LENGTH " + str(data["parameters"]["leader_max_length"]) + "\n"
|
||
|
output_buffer += "#define HASH_TYPE " + hash_type + "\n"
|
||
|
output_buffer += "#define NUMBER_OF_KEYS " + str(len(data["keys"])) + "\n"
|
||
|
output_buffer += "#define DEFAULT_PSEUDOLAYER " + data["parameters"]["default_pseudolayer"] + "\n"
|
||
|
|
||
|
return output_buffer + "\n"
|
||
|
|
||
|
def add_keycodes(data):
|
||
|
output_buffer = ""
|
||
|
|
||
|
if not len(data["keys"]) == len(set(data["keys"])):
|
||
|
raise Exception("The keys must have unique names")
|
||
|
|
||
|
for key, counter in zip(data["keys"], range(0, len(data["keys"]))):
|
||
|
output_buffer += "#define H_" + key + " ((HASH_TYPE) 1 << " + str(counter) + ")\n"
|
||
|
output_buffer += "\n"
|
||
|
|
||
|
output_buffer += "enum internal_keycodes {\n"
|
||
|
output_buffer += " " + data["keys"][0] + " = SAFE_RANGE,\n"
|
||
|
output_buffer += " " + reduce(comma_separator, [key for key in data["keys"][1:]]) + ",\n"
|
||
|
output_buffer += " FIRST_INTERNAL_KEYCODE = " + data["keys"][0] + ",\n"
|
||
|
output_buffer += " LAST_INTERNAL_KEYCODE = " + data["keys"][-1] + "\n"
|
||
|
output_buffer += "};\n"
|
||
|
|
||
|
return output_buffer + "\n"
|
||
|
|
||
|
def add_pseudolayers(data):
|
||
|
output_buffer = ""
|
||
|
|
||
|
if len(data["pseudolayers"]) == 0:
|
||
|
raise Exception("You didn't define any pseudolayers")
|
||
|
|
||
|
if not len([pseudolayer["name"] for pseudolayer in data["pseudolayers"]]) == len(set([pseudolayer["name"] for pseudolayer in data["pseudolayers"]])):
|
||
|
raise Exception("The pseudolayers must have unique names")
|
||
|
|
||
|
pseudolayers = data["pseudolayers"]
|
||
|
if not "ALWAYS_ON" in [layer["name"] for layer in pseudolayers]:
|
||
|
pseudolayers += [{"name": "ALWAYS_ON", "chords": []}] # the engine expects ALWAYS_ON to exist
|
||
|
|
||
|
output_buffer += "enum pseudolayers {\n"
|
||
|
output_buffer += " " + reduce(comma_separator, [layer["name"] for layer in pseudolayers]) + "\n"
|
||
|
output_buffer += "};\n"
|
||
|
|
||
|
return output_buffer + "\n"
|
||
|
|
||
|
def add_layers(data):
|
||
|
output_buffer = ""
|
||
|
|
||
|
output_buffer += "const uint16_t PROGMEM keymaps[][MATRIX_ROWS][MATRIX_COLS] = {\n"
|
||
|
for layer, counter in zip(data["layers"], range(0,len(data["layers"]))):
|
||
|
if layer["type"] == "auto":
|
||
|
output_buffer += " [" + str(counter) + "] = " + data["parameters"]["layout_function_name"] + "(" + reduce(comma_separator, [key for key in data["keys"]]) + "),\n"
|
||
|
else:
|
||
|
output_buffer += " [" + str(counter) + "] = " + data["parameters"]["layout_function_name"] + "(" + reduce(comma_separator, [key for key in layer["keycodes"]]) + "),\n"
|
||
|
output_buffer += "};\n"
|
||
|
output_buffer += "size_t keymapsCount = " + str(len(data["layers"])) + ";\n"
|
||
|
|
||
|
return output_buffer + "\n"
|
||
|
|
||
|
def prep_buffers(data):
|
||
|
output_buffer = ""
|
||
|
|
||
|
output_buffer += "uint8_t keycodes_buffer_array[] = {\n"
|
||
|
output_buffer += " " + reduce(comma_separator, ["0"] * len(data["keys"])) + "\n"
|
||
|
output_buffer += "};\n"
|
||
|
output_buffer += "\n"
|
||
|
|
||
|
output_buffer += "uint8_t command_buffer[] = {\n"
|
||
|
output_buffer += " " + reduce(comma_separator, ["0"] * data["parameters"]["command_max_length"]) + "\n"
|
||
|
output_buffer += "};\n"
|
||
|
output_buffer += "\n"
|
||
|
|
||
|
output_buffer += "uint16_t leader_buffer[] = {\n"
|
||
|
output_buffer += " " + reduce(comma_separator, ["0"] * data["parameters"]["leader_max_length"]) + "\n"
|
||
|
output_buffer += "};\n"
|
||
|
output_buffer += "\n"
|
||
|
|
||
|
output_buffer += "uint8_t dynamic_macro_buffer[] = {\n"
|
||
|
output_buffer += " " + reduce(comma_separator, ["0"] * data["parameters"]["dynamic_macro_max_length"]) + "\n"
|
||
|
output_buffer += "};"
|
||
|
|
||
|
return output_buffer + "\n"
|
||
|
|
||
|
def parse_keyboard_specifics(data):
|
||
|
keyboard_part_0 = add_includes(data)
|
||
|
keyboard_part_0 += add_keycodes(data)
|
||
|
keyboard_part_0 += add_pseudolayers(data)
|
||
|
keyboard_part_0 += add_parameters(data)
|
||
|
keyboard_part_0 += add_layers(data)
|
||
|
keyboard_part_0 += prep_buffers(data)
|
||
|
|
||
|
return keyboard_part_0 + '\n'
|
||
|
|
||
|
def parse_chords(data):
|
||
|
keyboard_part_2 = ""
|
||
|
strings = []
|
||
|
number_of_strings = 0
|
||
|
number_of_chords = 0
|
||
|
|
||
|
for pseudolayer in data["pseudolayers"]:
|
||
|
name = pseudolayer["name"]
|
||
|
for chord in pseudolayer["chords"]:
|
||
|
if chord["type"] == "chord_set":
|
||
|
keycodes = reduce(comma_separator, [word for word in chord["keycodes"]])
|
||
|
[keyboard_part_2, number_of_chords, number_of_strings, strings] = add_chord_set(name, keycodes, chord["set"], data, keyboard_part_2, number_of_chords, number_of_strings, strings)
|
||
|
if chord["type"] == "visual_array":
|
||
|
[keyboard_part_2, number_of_chords, number_of_strings, strings] = add_dictionary(name, chord["keys"], chord["dictionary"], keyboard_part_2, number_of_chords, number_of_strings, strings)
|
||
|
if chord["type"] == "visual":
|
||
|
keycodes = reduce(comma_separator, [word for word in chord["chord"]])
|
||
|
[keyboard_part_2, number_of_chords, number_of_strings, strings] = secret_chord(name, chord["keycode"], keycodes, data, keyboard_part_2, number_of_chords, number_of_strings, strings)
|
||
|
elif chord["type"] == "simple":
|
||
|
keycodes = reduce(string_sum, ["H_" + word for word in chord["chord"]])
|
||
|
[keyboard_part_2, number_of_chords, number_of_strings, strings] = add_key(name, keycodes, chord["keycode"], keyboard_part_2, number_of_chords, number_of_strings, strings)
|
||
|
keyboard_part_2 += "\n"
|
||
|
|
||
|
keyboard_part_2 += "const struct Chord* const list_of_chords[] PROGMEM = {\n"
|
||
|
keyboard_part_2 += " " + reduce(comma_separator, ["&chord_" + str(i) for i in range(0, number_of_chords)]) + "\n"
|
||
|
keyboard_part_2 += "};\n"
|
||
|
keyboard_part_2 += "\n"
|
||
|
|
||
|
if len(data["leader_sequences"]) > 0:
|
||
|
keyboard_part_2 += reduce(newline_separator, [sequence["function"] for sequence in data["leader_sequences"]]) + "\n\n"
|
||
|
keyboard_part_2 += "const uint16_t leader_triggers[][LEADER_MAX_LENGTH] PROGMEM = {\n"
|
||
|
for sequence in data["leader_sequences"]:
|
||
|
keyboard_part_2 += " {" + reduce(comma_separator, sequence["sequence"] + ["0"] * (data["parameters"]["leader_max_length"] - len(sequence["sequence"]))) + "},\n"
|
||
|
keyboard_part_2 += "};\n\n"
|
||
|
keyboard_part_2 += "void (*leader_functions[]) (void) = {\n"
|
||
|
keyboard_part_2 += " " + reduce(comma_separator, ["&" + sequence["name"] for sequence in data["leader_sequences"]]) + "\n"
|
||
|
keyboard_part_2 += "};\n"
|
||
|
else:
|
||
|
keyboard_part_2 += "const uint16_t** const leader_triggers PROGMEM = NULL;\n"
|
||
|
keyboard_part_2 += "void (*leader_functions[]) (void) = {};\n"
|
||
|
keyboard_part_2 += "\n"
|
||
|
|
||
|
keyboard_part_2 += "#define NUMBER_OF_CHORDS " + str(number_of_chords) + "\n"
|
||
|
keyboard_part_2 += "#define NUMBER_OF_LEADER_COMBOS " + str(len(data["leader_sequences"]))
|
||
|
|
||
|
return keyboard_part_2 + "\n\n"
|
||
|
|
||
|
def parse_strings_for_chords(data):
|
||
|
keyboard_part_1 = ""
|
||
|
|
||
|
for string, i in zip(strings, range(0, len(strings))):
|
||
|
keyboard_part_1 += "const char string_" + str(i) + " [] PROGMEM = \"" + string + "\";\n"
|
||
|
|
||
|
keyboard_part_1 += "\n"
|
||
|
keyboard_part_1 += "const char * const strings[] PROGMEM = {\n"
|
||
|
if len(strings) > 0:
|
||
|
keyboard_part_1 += " " + reduce(comma_separator, ["string_" + str(i) for i in range(0, len(strings))])
|
||
|
keyboard_part_1 += "\n};\n"
|
||
|
|
||
|
return keyboard_part_1
|
||
|
|
||
|
def main():
|
||
|
if len(sys.argv) != 3:
|
||
|
raise Exception("Wrong number of arguments.\n\nUsage: python parser.py keymap.json keymap.c")
|
||
|
|
||
|
input_filepath = sys.argv[1]
|
||
|
output_filepath = sys.argv[2]
|
||
|
|
||
|
with open(input_filepath, "r") as read_file:
|
||
|
data = json.load(read_file)
|
||
|
|
||
|
keyboard_part_0 = parse_keyboard_specifics(data)
|
||
|
keyboard_part_1 = parse_strings_for_chords(data)
|
||
|
keyboard_part_2 = parse_chords(data)
|
||
|
|
||
|
engine_part_1 = open("engine.part.1", "r").read()
|
||
|
engine_part_2 = open("engine.part.2", "r").read() + "\n"
|
||
|
engine_part_3 = open("engine.part.3", "r").read()
|
||
|
|
||
|
output_buffer = keyboard_part_0
|
||
|
output_buffer += engine_part_1
|
||
|
|
||
|
if len(data["extra_code"]) > 0:
|
||
|
output_buffer += data["extra_code"] + "\n"
|
||
|
|
||
|
output_buffer += keyboard_part_1
|
||
|
output_buffer += engine_part_2
|
||
|
output_buffer += keyboard_part_2
|
||
|
output_buffer += engine_part_3
|
||
|
|
||
|
with open(output_filepath, "w") as write_file:
|
||
|
write_file.write(output_buffer)
|
||
|
|
||
|
if __name__ == "__main__":
|
||
|
main()
|