import sys def get_lines(file_name): code_lines = [] pulling_from = None current_name = None processing_typedef = False for linenum, line in enumerate(open(sys.argv[1]).readlines()): line = line.rstrip() if pulling_from is None and len(line) > 0 and line[0] == '#' and line[-1] == '`': current_name = line[line[:-1].rfind('`') + 1: -1] if line[:9] == '```python': assert pulling_from is None pulling_from = linenum + 1 elif line[:3] == '```': if pulling_from is None: pulling_from = linenum else: if processing_typedef: assert code_lines[-1] == '}' code_lines[-1] = '})' pulling_from = None processing_typedef = False else: if pulling_from == linenum and line == '{': code_lines.append('%s = SSZType({' % current_name) processing_typedef = True elif pulling_from is not None: code_lines.append(line) elif pulling_from is None and len(line) > 0 and line[0] == '|': row = line[1:].split('|') if len(row) >= 2: for i in range(2): row[i] = row[i].strip().strip('`') if '`' in row[i]: row[i] = row[i][:row[i].find('`')] eligible = True if row[0][0] not in 'ABCDEFGHIJKLMNOPQRSTUVWXYZ_': eligible = False for c in row[0]: if c not in 'ABCDEFGHIJKLMNOPQRSTUVWXYZ_0123456789': eligible = False if eligible: code_lines.append(row[0] + ' = ' + (row[1].replace('**TBD**', '0x1234567890123567890123456789012357890'))) return code_lines