From 7ab5fc1f3b373fadc644448442b985c12597f8f0 Mon Sep 17 00:00:00 2001 From: Neels Hofmeyr Date: Thu, 15 Nov 2018 23:29:56 +0100 Subject: add contrib/struct_endianess.py In libosmocore (and likely elsewhere) we have scores of packed structs with sub-byte integer members that lack the necessary member reversal shims to be able to work on big endian architectures. Instead of manually editing each one of them and probably introduce errors in the process, this script handles the change automatically, and in the future allows us to verify correctness in gerrit verifications. Change-Id: I8e75b17d8071c7b3a2a171ba776fb76854b28a53 --- contrib/struct_endianess.py | 369 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 369 insertions(+) create mode 100755 contrib/struct_endianess.py (limited to 'contrib') diff --git a/contrib/struct_endianess.py b/contrib/struct_endianess.py new file mode 100755 index 00000000..be73fbe2 --- /dev/null +++ b/contrib/struct_endianess.py @@ -0,0 +1,369 @@ +#!/usr/bin/env python3 + +'''Using mad regexes, automatically make sure that all structs with sub-byte +integers have matching big-endian definitions. The idea is to save a lot of +manual effort, and to automatically verify that there are no errors. +This script most certainly has numerous holes and shortcomings, but actually, +if you hit problems with it, rather adjust your coding style so that this +script can deal with it...''' + +import re +import sys +import codecs +import os.path + +re_struct_start = re.compile(r'^struct\s*[a-zA-Z_][a-zA-Z_0-9]*\s*{\s*$') +re_struct_end = re.compile(r'^}[^;]*;\s*$') + +re_substruct_start = re.compile(r'^\s+struct\s*{\s*$') +re_substruct_end = re.compile(r'^\s+}\s*([^;]*\s)[a-zA-Z_][a-zA-Z_0-9]*\s*;\s*$') + +re_int_def = re.compile(r'(^\s*((const|unsigned|signed|char|int|long|int[0-9]+_t|uint[0-9]_t)\s+)+\s*)([^;]*;)', + re.DOTALL | re.MULTILINE) +re_int_members = re.compile(r'([a-zA-Z_][a-zA-Z_0-9]*|[a-zA-Z_][a-zA-Z_0-9]*\s*:\s*[0-9]+)\s*[,;]\s*', re.DOTALL | re.MULTILINE) + +re_little_endian_ifdef = re.compile(r'#\s*(if|elif)\s+OSMO_IS_LITTLE_ENDIAN\s*(==\s*1\s*|)'); +re_big_endian_ifdef = re.compile(r'#\s*(if|elif)\s+OSMO_IS_BIG_ENDIAN\s*'); +re_else = re.compile(r'#\s*else\s*'); +re_endif = re.compile(r'#\s*endif\s*'); + +re_c_comment = re.compile(r'(/\*[^*]+\*/|//.?$)') + +def remove_c_comments(code_str): + return ''.join(re_c_comment.split(code_str)[::2]) + +def section_struct_body(struct_body_lines): + '''divide a top-level-struct body into sections of + ['arbitrary string', ['body;\n', 'lines;\n'], 'arbitrary string', ...] + Aim: handle each sub-struct on its own, and if there already are ifdefs for + little and big endian, keep just the little endian bit and derive big + endian from it. + An arbitrary string is anything other than struct member definitions, like + a 'struct {', '} sub_name;', ... + "body lines" are lines that define struct members (possibly with comments). + Return: list of alternate arbitrary strings and variable definitions. + ''' + + # these globals are needed so that end_def() can change them from inside + # the function. Not very nice style, but easiest implementation. + global struct_body_parts + global arbitrary_part + global def_part + + struct_body_parts = [] + arbitrary_part = [] + def_part = [] + + def end_def(): + '''if there is any content, flush out recorded parts (def_part, + arbitrary_part) and start a new part. In short, cut a section + boundary.''' + global struct_body_parts + global arbitrary_part + global def_part + + if def_part: + struct_body_parts.append(arbitrary_part) + arbitrary_part = [] + struct_body_parts.append(def_part) + def_part = [] + + j = 0 + while j < len(struct_body_lines): + line = struct_body_lines[j] + + if (re_substruct_start.fullmatch(line) + or re_substruct_end.fullmatch(line)): + end_def() + arbitrary_part.append(line) + j += 1 + continue + + if re_big_endian_ifdef.fullmatch(line): + end_def() + # discard big endian section + j += 1 + while j < len(struct_body_lines): + line = struct_body_lines[j] + if re_endif.fullmatch(line): + end_def() + j += 1 + break; + if re_little_endian_ifdef.fullmatch(line): + end_def() + # keep that start of little endian section, not j++ + break; + if re_else.fullmatch(line): + # there's an '#else' after big-endian. Shim a little-endian header in just for the loop. + struct_body_lines[j] = '#if OSMO_IS_LITTLE_ENDIAN\n' + break; + j += 1 + continue + + if re_little_endian_ifdef.fullmatch(line): + end_def() + j += 1 + while j < len(struct_body_lines): + line = struct_body_lines[j] + if re_endif.fullmatch(line): + end_def() + j += 1 + break; + if re_big_endian_ifdef.fullmatch(line): + end_def() + # keep that start of big endian section, not j++ + break; + if re_else.fullmatch(line): + # there's an '#else' after little-endian. Shim a big-endian header in just for the loop. + struct_body_lines[j] = '#if OSMO_IS_BIG_ENDIAN\n' + break; + def_part.append(line) + j += 1 + + continue + + def_part.append(line) + j += 1 + + # flush the last section remaining that didn't see an explicit end + end_def() + # end_def() only flushes arbitrary_part if there was a def_part, so: + if arbitrary_part: + struct_body_parts.append(arbitrary_part) + + return struct_body_parts + +def struct_body_to_big_endian(body_str): + '''Input: a multi-line string containing the body of a struct, i.e. without + sub-structs and without #if OSMO_IS_BIG_ENDIAN. like + + '\tconst char *foo;\n\tuint8_t moo:3, goo:2;\n\tuint8_t loo:3;\n\tvoid *baz;\n' + + Return None to indicate that there is no little/big endian split + required, or return a multi-line string of the big-endian version of this + same struct body, where sub-byte ints are reversed at byte boundaries, and + all others are copied 1:1. If there are no sub-byte integers, return None, + to indicate that there is no little/big endian split required.''' + + # kick comments out of the code analysis. They will end up being stripped + # from big-endian only. + body_str = remove_c_comments(body_str) + + def_strs = body_str.split(';') + def_strs = ('%s;' % def_str for def_str in def_strs if def_str.strip()) + + # classify defs as containing sub-byte members or not + # defs = [ (true, 'uint8_t ', ('foo:3', 'bar:5')), + # (false, 'int baz;'),...] + defs = [] + any_sub_byte_ints = False + for one_def in def_strs: + + # does it have sub-string integers? + int_def = re_int_def.fullmatch(one_def) + if not int_def: + # not even a number, same for big and little endian + defs.append((False, one_def)) + continue + + int_type = int_def.group(1) + members_str = int_def.groups()[-1] + has_sub_byte_ints = False + + members = [] + for int_member in re_int_members.finditer(members_str): + member = int_member.group(1) + members.append(member) + if ':' in member: + has_sub_byte_ints = True + + if not has_sub_byte_ints: + defs.append((False, one_def)) + else: + defs.append((True, one_def, int_type, members)) + any_sub_byte_ints = True + + if not any_sub_byte_ints: + return None + + # now the interesting part, go over the defs, and reverse the sub-byte ints + # at byte boundaries. + + i = 0 + got_bits = 0 + byte_type = None + members_within_a_byte = [] + big_endian_defs = [] + + big_defs = [] + for classified_def in defs: + has_sub_byte_ints = classified_def[0] + + # now the big endian part + if has_sub_byte_ints: + _, one_def, int_type, members = classified_def + + if byte_type and byte_type.strip() != int_type.strip(): + raise Exception('mismatching type continuation after incomplete byte: %r %r to %r' + % (byte_type, members_within_a_byte, int_type)) + byte_type = int_type + + for member in members: + member_name, bits_str = member.split(':') + member_name = member_name.strip() + bits = int(bits_str) + member = '%s:%d' % (member_name, bits) + members_within_a_byte.append(member) + got_bits += bits + + if got_bits == 8: + # reverse these. + big_endian_defs.append('%s%s;' % (byte_type, ', '.join(reversed(members_within_a_byte)))) + members_within_a_byte = [] + byte_type = None + got_bits = 0 + + elif got_bits > 8: + raise Exception('sub-byte int breaks clean byte bounds: %s -- %d + %d = %d bits' + % (member, got_bits - bits, bits, got_bits)) + + elif not has_sub_byte_ints: + if got_bits: + raise Exception('sub-byte members do not add up to clean byte bounds: %r' % members_within_a_byte) + + big_endian_defs.append(classified_def[1]) + + # strip empty lines + lines = [l for l in (''.join(big_endian_defs).split('\n')) if l.strip()] + # clean lines' whitespace errors we might have taken in with the type names + for i in range(len(lines)): + line = lines[i] + while len(line) and line[-1] in ' \t': + line = line[:-1] + lines[i] = line + return '\n'.join(lines) + +def handle_struct_body(body_str): + + big_endian_body_str = struct_body_to_big_endian(body_str) + + if big_endian_body_str: + new_lines = ['#if OSMO_IS_LITTLE_ENDIAN\n'] + new_lines.append(body_str) + new_lines.append('#elif OSMO_IS_BIG_ENDIAN\n' + '/* auto-generated from the little endian part above (libosmocore/contrib/struct_endianess.py) */\n') + new_lines.append(big_endian_body_str) + new_lines.append('\n#endif\n') + return ''.join(new_lines) + else: + return body_str + +def _check_file(f): + if not (f.endswith('.h') or f.endswith('.c') or f.endswith('.cpp')): + return + + # section the file into + # [ ["no struct def"], ["struct {...};"], ["no struct def"], ... ] + sections = [] + in_struct = False + buf = [] + for line in codecs.open(f, "r", "utf-8").readlines(): + + if not in_struct and re_struct_start.fullmatch(line): + # flush whatever might still be in buf from before + sections.append(buf) + # start an in_struct section + buf = [line] + in_struct = True + elif in_struct and re_struct_end.fullmatch(line): + # add this end to the in_struct section and then start a non-struct section + buf.append(line) + sections.append(buf) + in_struct = False + buf = [] + else: + buf.append(line) + # flush any leftovers in buf + if buf: + sections.append(buf) + + # examine each struct, i.e. every second item in 'sections' + for i in range(len(sections)): + if not (i & 1): + continue + + struct = sections[i] + + # If the struct isn't packed, we need not bother. + # The practical use of this: in some structs we have booleans in the + # form of + # integer flag:1; + # and these don't add up to bytes, and cause errors. So let's skip all + # non-packed structs, then all of those are out of the picture. + if not 'packed' in struct[-1]: + continue + + try: + + # assume the 'struct foo {' is on the first line, the closing brace + # '} __attribute...;' on the last, and the rest are individual + # definitions split by ';'. + struct_body_lines = struct[1:-1] + struct_body_parts = section_struct_body(struct_body_lines) + + new_struct_body_parts = [] + for j in range(len(struct_body_parts)): + part = ''.join(struct_body_parts[j]) + if not (j & 1): + new_struct_body_parts.append(part) + else: + new_struct_body_parts.append(handle_struct_body(part)) + + new_struct = [struct[0], ''.join(new_struct_body_parts), struct[-1]] + sections[i] = new_struct + except Exception as e: + raise Exception('ERROR in struct %r' % struct[0]) + + # phew. result. + result = ''.join((''.join(s) for s in sections)) + + # see if osmocom/core/endian.h is needed and included. + if (not f.endswith('endian.h') + and 'OSMO_IS_LITTLE_ENDIAN' in result + and '#include ' not in result): + # add the include after the last 'osmocom/core' include + last_include_start = result.rfind('#include ' + result[insert_at:] + + with codecs.open(f, "w", "utf-8") as fd: + fd.write(result) + +def check_file(f): + try: + _check_file(f) + except Exception as e: + raise Exception('ERROR IN FILE %r' % f) + +args = sys.argv[1:] +if not args: + args = ['.'] + +for f in args: + if os.path.isdir(f): + for parent_path, subdirs, files in os.walk(f, None, None): + for ff in files: + check_file(os.path.join(parent_path, ff)) + else: + check_file(f) + +# vim: tabstop=4 shiftwidth=4 expandtab -- cgit v1.2.3