From 608fa5154c01420ff8f0946655ef16c99dec56a4 Mon Sep 17 00:00:00 2001 From: Joel Challis Date: Sun, 15 May 2022 22:39:29 +0100 Subject: Data driven `g_led_config` (#16728) --- lib/python/qmk/c_parse.py | 118 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 118 insertions(+) (limited to 'lib/python/qmk/c_parse.py') diff --git a/lib/python/qmk/c_parse.py b/lib/python/qmk/c_parse.py index 72be690019..359aaccbbc 100644 --- a/lib/python/qmk/c_parse.py +++ b/lib/python/qmk/c_parse.py @@ -1,5 +1,9 @@ """Functions for working with config.h files. """ +from pygments.lexers.c_cpp import CLexer +from pygments.token import Token +from pygments import lex +from itertools import islice from pathlib import Path import re @@ -13,6 +17,13 @@ multi_comment_regex = re.compile(r'/\*(.|\n)*?\*/', re.MULTILINE) layout_macro_define_regex = re.compile(r'^#\s*define') +def _get_chunks(it, size): + """Break down a collection into smaller parts + """ + it = iter(it) + return iter(lambda: tuple(islice(it, size)), ()) + + def strip_line_comment(string): """Removes comments from a single line string. """ @@ -170,3 +181,110 @@ def _parse_matrix_locations(matrix, file, macro_name): matrix_locations[identifier] = [row_num, col_num] return matrix_locations + + +def _coerce_led_token(_type, value): + """ Convert token to valid info.json content + """ + value_map = { + 'NO_LED': None, + 'LED_FLAG_ALL': 0xFF, + 'LED_FLAG_NONE': 0x00, + 'LED_FLAG_MODIFIER': 0x01, + 'LED_FLAG_UNDERGLOW': 0x02, + 'LED_FLAG_KEYLIGHT': 0x04, + 'LED_FLAG_INDICATOR': 0x08, + } + if _type is Token.Literal.Number.Integer: + return int(value) + if _type is Token.Literal.Number.Float: + return float(value) + if _type is Token.Literal.Number.Hex: + return int(value, 0) + if _type is Token.Name and value in value_map.keys(): + return value_map[value] + + +def _parse_led_config(file, matrix_cols, matrix_rows): + """Return any 'raw' led/rgb matrix config + """ + file_contents = file.read_text(encoding='utf-8') + file_contents = comment_remover(file_contents) + file_contents = file_contents.replace('\\\n', '') + + matrix_raw = [] + position_raw = [] + flags = [] + + found_led_config = False + bracket_count = 0 + section = 0 + for _type, value in lex(file_contents, CLexer()): + # Assume g_led_config..stuff..; + if value == 'g_led_config': + found_led_config = True + elif value == ';': + found_led_config = False + elif found_led_config: + # Assume bracket count hints to section of config we are within + if value == '{': + bracket_count += 1 + if bracket_count == 2: + section += 1 + elif value == '}': + bracket_count -= 1 + else: + # Assume any non whitespace value here is important enough to stash + if _type in [Token.Literal.Number.Integer, Token.Literal.Number.Float, Token.Literal.Number.Hex, Token.Name]: + if section == 1 and bracket_count == 3: + matrix_raw.append(_coerce_led_token(_type, value)) + if section == 2 and bracket_count == 3: + position_raw.append(_coerce_led_token(_type, value)) + if section == 3 and bracket_count == 2: + flags.append(_coerce_led_token(_type, value)) + + # Slightly better intrim format + matrix = list(_get_chunks(matrix_raw, matrix_cols)) + position = list(_get_chunks(position_raw, 2)) + matrix_indexes = list(filter(lambda x: x is not None, matrix_raw)) + + # If we have not found anything - bail + if not section: + return None + + # TODO: Improve crude parsing/validation + if len(matrix) != matrix_rows and len(matrix) != (matrix_rows / 2): + raise ValueError("Unable to parse g_led_config matrix data") + if len(position) != len(flags): + raise ValueError("Unable to parse g_led_config position data") + if len(matrix_indexes) and (max(matrix_indexes) >= len(flags)): + raise ValueError("OOB within g_led_config matrix data") + + return (matrix, position, flags) + + +def find_led_config(file, matrix_cols, matrix_rows): + """Search file for led/rgb matrix config + """ + found = _parse_led_config(file, matrix_cols, matrix_rows) + if not found: + return None + + # Expand collected content + (matrix, position, flags) = found + + # Align to output format + led_config = [] + for index, item in enumerate(position, start=0): + led_config.append({ + 'x': item[0], + 'y': item[1], + 'flags': flags[index], + }) + for r in range(len(matrix)): + for c in range(len(matrix[r])): + index = matrix[r][c] + if index is not None: + led_config[index]['matrix'] = [r, c] + + return led_config -- cgit v1.2.3