#!/usr/bin/env python3 # This file is a part of toml++ and is subject to the the terms of the MIT license. # Copyright (c) Mark Gillard # See https://github.com/marzer/tomlplusplus/blob/master/LICENSE for the full license text. # SPDX-License-Identifier: MIT import sys import os.path as path import utils import re from io import StringIO class Preprocessor: __re_strip_blocks = re.compile(r'//[#!]\s*[{][{].*?//[#!]\s*[}][}]*?\n', re.I | re.S) __re_includes = re.compile(r'^\s*#\s*include\s+"(.+?)"', re.I | re.M) def __preprocess(self, match): raw_incl = match if isinstance(match, str) else match.group(1) incl = raw_incl.strip().lower() if incl in self.__processed_includes: return '' self.__processed_includes.append(incl) text = utils.read_all_text_from_file(path.join(utils.get_script_folder(), '..', 'include', 'toml++', incl)).strip() + '\n' text = text.replace('\r\n', '\n') # convert windows newlines text = self.__re_strip_blocks.sub('', text, 0) # strip {{ }} blocks self.__current_level += 1 text = self.__re_includes.sub(lambda m : self.__preprocess(m), text, 0) self.__current_level -= 1 if (self.__current_level == 1): header_text = '↓ ' + raw_incl lpad = 20 + ((25 * (self.__header_indent % 4)) - int((len(header_text) + 4) / 2)) self.__header_indent += 1 text = '#if 1 {}\n{}\n\n#endif {}\n'.format( utils.make_divider(header_text, lpad, line_length=113), text, utils.make_divider('↑ ' + raw_incl, lpad, line_length=113) ) return '\n\n' + text + '\n\n' # will get merged later def __init__(self, file): self.__processed_includes = [] self.__header_indent = 0 self.__current_level = 0 self.__string = self.__preprocess(file) def __str__(self): return self.__string def main(): # preprocess header(s) source_text = str(Preprocessor('toml.h')) # strip various things: # 'pragma once' source_text = re.sub(r'^\s*#\s*pragma\s+once\s*$', '', source_text, 0, re.I | re.M) # clang-format directives source_text = re.sub(r'^\s*//\s*clang-format\s+.+?$', '', source_text, 0, re.I | re.M) # spdx license identifiers source_text = re.sub(r'^\s*//\s*SPDX-License-Identifier:.+?$', '', source_text, 0, re.I | re.M) # 'magic' comment blocks (incl. doxygen) source_text = re.sub('(?:(?:\n|^)[ \t]*//[/#!<]+[^\n]*)+\n', '\n', source_text, 0, re.I | re.M) # 'magic' comments (incl. doxygen) source_text = re.sub('(?://[/#!<].*?)\n', '\n', source_text, 0, re.I | re.M) # remove trailing whitespace source_text = re.sub('([^ \t])[ \t]+\n', '\\1\n', source_text, 0, re.I | re.M) # bookended namespace blocks source_text = re.sub('}\n+TOML_NAMESPACE_END\n+TOML_NAMESPACE_START\n+{\n+', '\n', source_text, 0, re.I | re.M) source_text = re.sub('}\n+TOML_IMPL_NAMESPACE_END\n+TOML_IMPL_NAMESPACE_START\n+{\n+', '\n', source_text, 0, re.I | re.M) # blank lines before some preprocessor directives #source_text = re.sub('\n+\n(\s*)#\s*(elif|else|endif)(.*?)\n', '\n\\1#\\2\\3\n', source_text, 0, re.I | re.M) # blank lines after some preprocessor directives #source_text = re.sub('#\s*(if|ifn?def|elif|else)(.*?)\n\n+', '#\\1\\2\n', source_text, 0, re.I | re.M) # blank lines after opening braces source_text = re.sub('[{]\s*\n\s*\n+', '{\n', source_text, 0, re.I | re.M) # double newlines source_text = re.sub('\n(?:[ \t]*\n[ \t]*)+\n', '\n\n', source_text, 0, re.I | re.M) # source_text = re.sub( # blank lines between various preprocessor directives # '[#](endif(?:\s*//[^\n]*)?)\n{2,}[#](ifn?(?:def)?|define)', # '#\\1\n#\\2', # source_text, 0, re.I | re.M # ) return_type_pattern \ = r'(?:' \ + r'(?:\[\[nodiscard\]\]\s*)?' \ + r'(?:(?:friend|explicit|virtual|inline|const|operator)\s+)*' \ + r'(?:' \ + r'bool|int64_t|(?:const_)?iterator|double|void' \ + r'|node(?:_(?:view|of)<.+?>|)?|table|array|value(?:<.+?>)?' \ + r'|T|U|parse_(?:error|result)' \ + r')' \ + r'(?:\s*[&*]+)?' \ + r'(?:\s*[(]\s*[)])?' \ + r'\s+' \ + r')' blank_lines_between_returns_pattern = '({}[^\n]+)\n\n([ \t]*{})'.format(return_type_pattern, return_type_pattern) for i in range(0, 5): # remove blank lines between simple one-liner definitions source_text = re.sub('(using .+?;)\n\n([ \t]*using)', '\\1\n\\2', source_text, 0, re.I | re.M) source_text = re.sub( '([a-zA-Z_][a-zA-Z0-9_]*[ \t]+[a-zA-Z_][a-zA-Z0-9_]*[ \t]*;)' \ + '\n\n([ \t]*[a-zA-Z_][a-zA-Z0-9_]*[ \t]+[a-zA-Z_][a-zA-Z0-9_]*[ \t]*;)', '\\1\n\\2', source_text, 0, re.I | re.M) source_text = re.sub(blank_lines_between_returns_pattern, '\\1\n\\2', source_text, 0, re.I | re.M) source_text = source_text.strip() + '\n' # change TOML_LIB_SINGLE_HEADER to 1 source_text = re.sub( '#\s*define\s+TOML_LIB_SINGLE_HEADER\s+[0-9]+', '#define TOML_LIB_SINGLE_HEADER 1', source_text, 0, re.I ) # extract library version library_version = { 'major': 0, 'minor': 0, 'patch': 0 } match = re.search(r'^\s*#\s*define\s+TOML_LIB_MAJOR\s+([0-9]+)\s*$', source_text, re.I | re.M) if match is not None: library_version['major'] = match.group(1) match = re.search(r'^\s*#\s*define\s+TOML_LIB_MINOR\s+([0-9]+)\s*$', source_text, re.I | re.M) if match is not None: library_version['minor'] = match.group(1) match = re.search(r'^\s*#\s*define\s+TOML_LIB_(?:REVISION|PATCH)\s+([0-9]+)\s*$', source_text, re.I | re.M) if match is not None: library_version['patch'] = match.group(1) # build the preamble (license etc) preamble = [] preamble.append(''' // toml++ v{major}.{minor}.{patch} // https://github.com/marzer/tomlplusplus // SPDX-License-Identifier: MIT'''.format(**library_version)) preamble.append(''' // - THIS FILE WAS ASSEMBLED FROM MULTIPLE HEADER FILES BY A SCRIPT - PLEASE DON'T EDIT IT DIRECTLY - // // If you wish to submit a contribution to toml++, hooray and thanks! Before you crack on, please be aware that this // file was assembled from a number of smaller files by a python script, and code contributions should not be made // against it directly. You should instead make your changes in the relevant source file(s). The file names of the files // that contributed to this header can be found at the beginnings and ends of the corresponding sections of this file.''') preamble.append(''' // TOML Language Specifications: // latest: https://github.com/toml-lang/toml/blob/master/README.md // v1.0.0-rc.3: https://toml.io/en/v1.0.0-rc.3 // v1.0.0-rc.2: https://toml.io/en/v1.0.0-rc.2 // v1.0.0-rc.1: https://toml.io/en/v1.0.0-rc.1 // v0.5.0: https://toml.io/en/v0.5.0 // changelog: https://github.com/toml-lang/toml/blob/master/CHANGELOG.md''') preamble.append(utils.read_all_text_from_file(path.join(utils.get_script_folder(), '..', 'LICENSE'))) # write the output with StringIO(newline='\n') as output: # build in a string buffer write = lambda txt, end='\n': print(txt, file=output, end=end) if (len(preamble) > 0): write(utils.make_divider()) for pre in preamble: write('//') for line in pre.strip().splitlines(): if len(line) == 0: write('//') continue if not line.startswith('//'): write('// ', end = '') write(line) write('//') write(utils.make_divider()) write(source_text) write('') output_str = output.getvalue().strip() # analyze the output to find any potentially missing #undefs re_define = re.compile(r'^\s*#\s*define\s+([a-zA-Z0-9_]+)(?:$|\s|\()') re_undef = re.compile(r'^\s*#\s*undef\s+([a-zA-Z0-9_]+)(?:$|\s|//)') defines = dict() for output_line in output_str.splitlines(): defined = True m = re_define.match(output_line) if not m: defined = False m = re_undef.match(output_line) if m: defines[m.group(1)] = defined ignore_list = ( # macros that are meant to stay public (user configs etc) 'INCLUDE_TOMLPLUSPLUS_H', 'TOML_API', 'TOML_UNRELEASED_FEATURES', 'TOML_LARGE_FILES', 'TOML_PARSER', 'TOML_WINDOWS_COMPAT', 'TOML_EXCEPTIONS', 'TOML_LIB_SINGLE_HEADER', 'TOML_LIB_MAJOR', 'TOML_LIB_MINOR', 'TOML_LIB_PATCH', 'TOML_LANG_MAJOR', 'TOML_LANG_MINOR', 'TOML_LANG_PATCH', 'TOML_UNDEF_MACROS', 'TOML_HEADER_ONLY', 'TOML_ALL_INLINE' ) set_defines = [] for define, currently_set in defines.items(): if currently_set and define not in ignore_list: set_defines.append(define) if len(set_defines) > 0: set_defines.sort() print(f"Potentially missing #undefs:") for define in set_defines: print(f"\t#undef {define}") # write the output file output_file_path = path.join(utils.get_script_folder(), '..', 'toml.hpp') print("Writing to {}".format(output_file_path)) with open(output_file_path,'w', encoding='utf-8', newline='\n') as output_file: print(output_str, file=output_file) if __name__ == '__main__': utils.run(main)