mirror of
				https://github.com/ethereum/solidity
				synced 2023-10-03 13:03:40 +00:00 
			
		
		
		
	
		
			
				
	
	
		
			252 lines
		
	
	
		
			7.0 KiB
		
	
	
	
		
			Python
		
	
	
		
			Executable File
		
	
	
	
	
			
		
		
	
	
			252 lines
		
	
	
		
			7.0 KiB
		
	
	
	
		
			Python
		
	
	
		
			Executable File
		
	
	
	
	
| #! /usr/bin/env python3
 | |
| import random
 | |
| import re
 | |
| import os
 | |
| import getopt
 | |
| import sys
 | |
| from os import path
 | |
| 
 | |
| ENCODING = "utf-8"
 | |
| SOURCE_FILE_PATTERN = r"\b\d+_error\b"
 | |
| 
 | |
| 
 | |
| def read_file(file_name):
 | |
|     content = None
 | |
|     try:
 | |
|         with open(file_name, "r", encoding=ENCODING) as f:
 | |
|             content = f.read()
 | |
|     finally:
 | |
|         if content == None:
 | |
|             print(f"Error reading: {file_name}")
 | |
|     return content
 | |
| 
 | |
| 
 | |
| def write_file(file_name, content):
 | |
|     with open(file_name, "w", encoding=ENCODING) as f:
 | |
|         f.write(content)
 | |
| 
 | |
| 
 | |
| def in_comment(source, pos):
 | |
|     slash_slash_pos = source.rfind("//", 0, pos)
 | |
|     lf_pos = source.rfind("\n", 0, pos)
 | |
|     if slash_slash_pos > lf_pos:
 | |
|         return True
 | |
|     slash_star_pos = source.rfind("/*", 0, pos)
 | |
|     star_slash_pos = source.rfind("*/", 0, pos)
 | |
|     return slash_star_pos > star_slash_pos
 | |
| 
 | |
| 
 | |
| def find_ids_in_source_file(file_name, ids):
 | |
|     source = read_file(file_name)
 | |
|     for m in re.finditer(SOURCE_FILE_PATTERN, source):
 | |
|         if in_comment(source, m.start()):
 | |
|             continue
 | |
|         underscore_pos = m.group(0).index("_")
 | |
|         id = m.group(0)[0:underscore_pos]
 | |
|         if id in ids:
 | |
|             ids[id] += 1
 | |
|         else:
 | |
|             ids[id] = 1
 | |
| 
 | |
| 
 | |
| def get_used_ids(file_names):
 | |
|     used_ids = {}
 | |
|     for file_name in file_names:
 | |
|         find_ids_in_source_file(file_name, used_ids)
 | |
|     return used_ids
 | |
| 
 | |
| 
 | |
| def get_id(available_ids, used_ids):
 | |
|     while len(available_ids) > 0:
 | |
|         k = random.randrange(len(available_ids))
 | |
|         id = list(available_ids.keys())[k]
 | |
|         del available_ids[id]
 | |
|         if id not in used_ids:
 | |
|             return id
 | |
|     assert False, "Out of IDs"
 | |
| 
 | |
| 
 | |
| def fix_ids_in_file(file_name, available_ids, used_ids):
 | |
|     source = read_file(file_name)
 | |
| 
 | |
|     k = 0
 | |
|     destination = []
 | |
|     for m in re.finditer(SOURCE_FILE_PATTERN, source):
 | |
|         destination.extend(source[k:m.start()])
 | |
| 
 | |
|         underscore_pos = m.group(0).index("_")
 | |
|         id = m.group(0)[0:underscore_pos]
 | |
| 
 | |
|         # incorrect id or id has a duplicate somewhere
 | |
|         if not in_comment(source, m.start()) and (len(id) != 4 or id[0] == "0" or used_ids[id] > 1):
 | |
|             assert id in used_ids
 | |
|             new_id = get_id(available_ids, used_ids)
 | |
|             used_ids[id] -= 1
 | |
|         else:
 | |
|             new_id = id
 | |
| 
 | |
|         destination.extend(new_id + "_error")
 | |
|         k = m.end()
 | |
| 
 | |
|     destination.extend(source[k:])
 | |
| 
 | |
|     destination = ''.join(destination)
 | |
|     if source != destination:
 | |
|         write_file(file_name, destination)
 | |
|         print(f"Fixed file: {file_name}")
 | |
| 
 | |
| 
 | |
| def fix_ids(used_ids, file_names):
 | |
|     available_ids = {str(id): None for id in range(1000, 10000)}
 | |
|     for file_name in file_names:
 | |
|         fix_ids_in_file(file_name, available_ids, used_ids)
 | |
| 
 | |
| 
 | |
| def find_files(top_dir, sub_dirs, extensions):
 | |
|     """Builds a list of files with given extensions in specified subdirectories"""
 | |
| 
 | |
|     source_file_names = []
 | |
|     for dir in sub_dirs:
 | |
|         for root, _, file_names in os.walk(os.path.join(top_dir, dir), onerror=lambda e: exit(f"Walk error: {e}")):
 | |
|             for file_name in file_names:
 | |
|                 _, ext = path.splitext(file_name)
 | |
|                 if ext in extensions:
 | |
|                     source_file_names.append(path.join(root, file_name))
 | |
| 
 | |
|     return source_file_names
 | |
| 
 | |
| 
 | |
| def find_ids_in_test_file(file_name):
 | |
|     source = read_file(file_name)
 | |
|     pattern = r"^// (.*Error|Warning) \d\d\d\d:"
 | |
|     return {m.group(0)[-5:-1] for m in re.finditer(pattern, source, flags=re.MULTILINE)}
 | |
| 
 | |
| 
 | |
| def find_ids_in_test_files(file_names):
 | |
|     used_ids = set()
 | |
|     for file_name in file_names:
 | |
|         used_ids |= find_ids_in_test_file(file_name)
 | |
|     return used_ids
 | |
| 
 | |
| 
 | |
| def print_ids(ids):
 | |
|     for k, id in enumerate(sorted(ids)):
 | |
|         if k % 10 > 0:
 | |
|             print(" ", end="")
 | |
|         elif k > 0:
 | |
|             print()
 | |
|         print(id, end="")
 | |
| 
 | |
| 
 | |
| def examine_id_coverage(top_dir, used_ids):
 | |
|     test_sub_dirs = [
 | |
|         path.join("test", "libsolidity", "errorRecoveryTests"),
 | |
|         path.join("test", "libsolidity", "smtCheckerTests"),
 | |
|         path.join("test", "libsolidity", "syntaxTests")
 | |
|     ]
 | |
|     test_file_names = find_files(
 | |
|         top_dir,
 | |
|         test_sub_dirs,
 | |
|         [".sol"]
 | |
|     )
 | |
|     covered_ids = find_ids_in_test_files(test_file_names)
 | |
| 
 | |
|     print(f"IDs in source files: {len(used_ids)}")
 | |
|     print(f"IDs in test files  : {len(covered_ids)} ({len(covered_ids) - len(used_ids)})")
 | |
|     print()
 | |
| 
 | |
|     unused_covered_ids = covered_ids - used_ids
 | |
|     if len(unused_covered_ids) != 0:
 | |
|         print("Error. The following error codes found in tests, but not in sources:")
 | |
|         print_ids(unused_covered_ids)
 | |
|         return 1
 | |
| 
 | |
|     used_uncovered_ids = used_ids - covered_ids
 | |
|     if len(used_uncovered_ids) != 0:
 | |
|         print("The following error codes found in sources, but not in tests:")
 | |
|         print_ids(used_uncovered_ids)
 | |
|         print("\n\nPlease make sure to add appropriate tests.")
 | |
|         return 1
 | |
| 
 | |
|     return 0
 | |
| 
 | |
| 
 | |
| def main(argv):
 | |
|     # pylint: disable=too-many-branches, too-many-locals
 | |
| 
 | |
|     check = False
 | |
|     fix = False
 | |
|     noconfirm = False
 | |
|     examine_coverage = False
 | |
|     opts, args = getopt.getopt(argv, "", ["check", "fix", "noconfirm", "examine-coverage"])
 | |
| 
 | |
|     for opt, arg in opts:
 | |
|         if opt == '--check':
 | |
|             check = True
 | |
|         elif opt == "--fix":
 | |
|             fix = True
 | |
|         elif opt == '--noconfirm':
 | |
|             noconfirm = True
 | |
|         elif opt == '--examine-coverage':
 | |
|             examine_coverage = True
 | |
| 
 | |
|     if not check and not fix and not examine_coverage:
 | |
|         print("usage: python error_codes.py --check | --fix [--noconfirm] | --examine-coverage")
 | |
|         exit(1)
 | |
| 
 | |
|     cwd = os.getcwd()
 | |
| 
 | |
|     source_file_names = find_files(
 | |
|         cwd,
 | |
|         ["libevmasm", "liblangutil", "libsolc", "libsolidity", "libsolutil", "libyul", "solc"],
 | |
|         [".h", ".cpp"]
 | |
|     )
 | |
|     used_ids = get_used_ids(source_file_names)
 | |
| 
 | |
|     ok = True
 | |
|     for id in sorted(used_ids):
 | |
|         if len(id) != 4:
 | |
|             print(f"ID {id} length != 4")
 | |
|             ok = False
 | |
|         if id[0] == "0":
 | |
|             print(f"ID {id} starts with zero")
 | |
|             ok = False
 | |
|         if used_ids[id] > 1:
 | |
|             print(f"ID {id} appears {used_ids[id]} times")
 | |
|             ok = False
 | |
| 
 | |
|     if examine_coverage:
 | |
|         if not ok:
 | |
|             print("Incorrect IDs has to be fixed before applying --examine-coverage")
 | |
|         res = examine_id_coverage(cwd, used_ids.keys())
 | |
|         exit(res)
 | |
| 
 | |
|     if ok:
 | |
|         print("No incorrect IDs found")
 | |
|         exit(0)
 | |
| 
 | |
|     if check:
 | |
|         exit(1)
 | |
| 
 | |
|     assert fix, "Unexpected state, should not come here without --fix"
 | |
| 
 | |
|     if not noconfirm:
 | |
|         answer = input(
 | |
|             "\nDo you want to fix incorrect IDs?\n"
 | |
|             "Please commit current changes first, and review the results when the script finishes.\n"
 | |
|             "[Y/N]? "
 | |
|         )
 | |
|         while len(answer) == 0 or answer not in "YNyn":
 | |
|             answer = input("[Y/N]? ")
 | |
|         if answer not in "yY":
 | |
|             exit(1)
 | |
| 
 | |
|     random.seed()
 | |
|     fix_ids(used_ids, source_file_names)
 | |
|     print("Fixing completed")
 | |
|     exit(2)
 | |
| 
 | |
| 
 | |
| if __name__ == "__main__":
 | |
|     main(sys.argv[1:])
 |