# This Source Code Form is subject to the terms of the Mozilla Public # License, v. 2.0. If a copy of the MPL was not distributed with this # file, You can obtain one at http://mozilla.org/MPL/2.0/. import sys import ReservedWordReader def line(opt, s): opt["output"].write("{}{}\n".format(" " * opt["indent_level"], s)) def indent(opt): opt["indent_level"] += 1 def dedent(opt): opt["indent_level"] -= 1 def span_and_count_at(reserved_word_list, column): assert len(reserved_word_list) != 0 chars_dict = {} for index, word in reserved_word_list: chars_dict[ord(word[column])] = True chars = sorted(chars_dict.keys()) return chars[-1] - chars[0] + 1, len(chars) def optimal_switch_column(opt, reserved_word_list, columns, unprocessed_columns): assert len(reserved_word_list) != 0 assert unprocessed_columns != 0 min_count = 0 min_span = 0 min_count_index = 0 min_span_index = 0 for index in range(0, unprocessed_columns): span, count = span_and_count_at(reserved_word_list, columns[index]) assert span != 0 if span == 1: assert count == 1 return 1, True assert count != 1 if index == 0 or min_span > span: min_span = span min_span_index = index if index == 0 or min_count > count: min_count = count min_count_index = index if min_count <= opt["use_if_threshold"]: return min_count_index, True return min_span_index, False def split_list_per_column(reserved_word_list, column): assert len(reserved_word_list) != 0 column_dict = {} for item in reserved_word_list: index, word = item per_column = column_dict.setdefault(word[column], []) per_column.append(item) return sorted(column_dict.items()) def generate_letter_switch(opt, unprocessed_columns, reserved_word_list, columns=None): assert len(reserved_word_list) != 0 if not columns: columns = range(0, unprocessed_columns) if len(reserved_word_list) == 1: index, word = reserved_word_list[0] if unprocessed_columns == 0: line(opt, f"JSRW_GOT_MATCH({index}) /* {word} */") return if unprocessed_columns > opt["char_tail_test_threshold"]: line(opt, f"JSRW_TEST_GUESS({index}) /* {word} */") return conds = [] for column in columns[0:unprocessed_columns]: quoted = repr(word[column]) conds.append(f"JSRW_AT({column})=={quoted}") line(opt, "if ({}) {{".format(" && ".join(conds))) indent(opt) line(opt, f"JSRW_GOT_MATCH({index}) /* {word} */") dedent(opt) line(opt, "}") line(opt, "JSRW_NO_MATCH()") return assert unprocessed_columns != 0 optimal_column_index, use_if = optimal_switch_column( opt, reserved_word_list, columns, unprocessed_columns ) optimal_column = columns[optimal_column_index] # Make a copy to avoid breaking passed list. columns = list(columns) columns[optimal_column_index] = columns[unprocessed_columns - 1] list_per_column = split_list_per_column(reserved_word_list, optimal_column) if not use_if: line(opt, f"switch (JSRW_AT({optimal_column})) {{") for char, reserved_word_list_per_column in list_per_column: quoted = repr(char) if use_if: line(opt, f"if (JSRW_AT({optimal_column}) == {quoted}) {{") else: line(opt, f" case {quoted}:") indent(opt) generate_letter_switch( opt, unprocessed_columns - 1, reserved_word_list_per_column, columns ) dedent(opt) if use_if: line(opt, "}") if not use_if: line(opt, "}") line(opt, "JSRW_NO_MATCH()") def split_list_per_length(reserved_word_list): assert len(reserved_word_list) != 0 length_dict = {} for item in reserved_word_list: index, word = item per_length = length_dict.setdefault(len(word), []) per_length.append(item) return sorted(length_dict.items()) def generate_switch(opt, reserved_word_list): assert len(reserved_word_list) != 0 line(opt, "/*") line( opt, f" * Generating switch for the list of {len(reserved_word_list)} entries:", ) for index, word in reserved_word_list: line(opt, f" * {word}") line(opt, " */") list_per_length = split_list_per_length(reserved_word_list) use_if = False if len(list_per_length) < opt["use_if_threshold"]: use_if = True if not use_if: line(opt, "switch (JSRW_LENGTH()) {") for length, reserved_word_list_per_length in list_per_length: if use_if: line(opt, f"if (JSRW_LENGTH() == {length}) {{") else: line(opt, f" case {length}:") indent(opt) generate_letter_switch(opt, length, reserved_word_list_per_length) dedent(opt) if use_if: line(opt, "}") if not use_if: line(opt, "}") line(opt, "JSRW_NO_MATCH()") def main(output, reserved_words_h, *args): reserved_word_list = ReservedWordReader.read_reserved_word_list( reserved_words_h, *args ) opt = { "indent_level": 1, "use_if_threshold": 3, "char_tail_test_threshold": 4, "output": output, } generate_switch(opt, reserved_word_list) if __name__ == "__main__": main(sys.stdout, *sys.argv[1:])