You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

250 lines
8.5 KiB

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""This Python script formats nginx configuration files in consistent way.
Originally published under https://github.com/1connect/nginx-config-formatter
"""
import argparse
import codecs
import re
__author__ = "Michał Słomkowski"
__license__ = "Apache 2.0"
__version__ = "1.0.2"
INDENTATION = ' ' * 4
TEMPLATE_VARIABLE_OPENING_TAG = '___TEMPLATE_VARIABLE_OPENING_TAG___'
TEMPLATE_VARIABLE_CLOSING_TAG = '___TEMPLATE_VARIABLE_CLOSING_TAG___'
TEMPLATE_BRACKET_OPENING_TAG = '___TEMPLATE_BRACKET_OPENING_TAG___'
TEMPLATE_BRACKET_CLOSING_TAG = '___TEMPLATE_BRACKET_CLOSING_TAG___'
def strip_line(single_line):
"""Strips the line and replaces neighbouring whitespaces with single space (except when within quotation marks)."""
single_line = single_line.strip()
if single_line.startswith('#'):
return single_line
within_quotes = False
parts = []
for part in re.split('"', single_line):
if within_quotes:
parts.append(part)
else:
parts.append(re.sub(r'[\s]+', ' ', part))
within_quotes = not within_quotes
return '"'.join(parts)
def count_multi_semicolon(single_line):
"""count multi_semicolon (except when within quotation marks)."""
single_line = single_line.strip()
if single_line.startswith('#'):
return 0, 0
within_quotes = False
q = 0
c = 0
for part in re.split('"', single_line):
if within_quotes:
q = 1
else:
c += part.count(';')
within_quotes = not within_quotes
return q, c
def multi_semicolon(single_line):
"""break multi_semicolon into multiline (except when within quotation marks)."""
single_line = single_line.strip()
if single_line.startswith('#'):
return single_line
within_quotes = False
parts = []
for part in re.split('"', single_line):
if within_quotes:
parts.append(part)
else:
parts.append(part.replace(";", ";\n"))
within_quotes = not within_quotes
return '"'.join(parts)
def apply_variable_template_tags(line: str) -> str:
"""Replaces variable indicators ${ and } with tags, so subsequent formatting is easier."""
return re.sub(r'\${\s*(\w+)\s*}',
TEMPLATE_VARIABLE_OPENING_TAG + r"\1" + TEMPLATE_VARIABLE_CLOSING_TAG,
line,
flags=re.UNICODE)
def strip_variable_template_tags(line: str) -> str:
"""Replaces tags back with ${ and } respectively."""
return re.sub(TEMPLATE_VARIABLE_OPENING_TAG + r'\s*(\w+)\s*' + TEMPLATE_VARIABLE_CLOSING_TAG,
r'${\1}',
line,
flags=re.UNICODE)
def apply_bracket_template_tags(content: str) -> str:
""" Replaces bracket { and } with tags, so subsequent formatting is easier."""
result = ""
in_quotes = False
last_c = ""
for c in content:
if (c == "\'" or c == "\"") and last_c != "\\":
in_quotes = reverse_in_quotes_status(in_quotes)
if in_quotes:
if c == "{":
result += TEMPLATE_BRACKET_OPENING_TAG
elif c == "}":
result += TEMPLATE_BRACKET_CLOSING_TAG
else:
result += c
else:
result += c
last_c = c
return result
def reverse_in_quotes_status(status: bool) -> bool:
if status:
return False
return True
def strip_bracket_template_tags(content: str) -> str:
""" Replaces tags back with { and } respectively."""
content = content.replace(TEMPLATE_BRACKET_OPENING_TAG, "{", -1)
content = content.replace(TEMPLATE_BRACKET_CLOSING_TAG, "}", -1)
return content
def clean_lines(orig_lines) -> list:
"""Strips the lines and splits them if they contain curly brackets."""
cleaned_lines = []
for line in orig_lines:
line = strip_line(line)
line = apply_variable_template_tags(line)
if line == "":
cleaned_lines.append("")
continue
else:
if line.startswith("#"):
cleaned_lines.append(strip_variable_template_tags(line))
else:
q , c = count_multi_semicolon(line)
if q == 1 and c > 1:
ml = multi_semicolon(line)
cleaned_lines.extend(clean_lines(ml.splitlines()))
elif q != 1 and c > 1:
newlines = line.split(";")
cleaned_lines.extend(clean_lines(["".join([ln, ";"]) for ln in newlines if ln != ""]))
else:
if line.startswith("rewrite"):
cleaned_lines.append(strip_variable_template_tags(line))
else:
cleaned_lines.extend(
[strip_variable_template_tags(l).strip() for l in re.split(r"([{}])", line) if l != ""])
return cleaned_lines
def join_opening_bracket(lines):
"""When opening curly bracket is in it's own line (K&R convention), it's joined with precluding line (Java)."""
modified_lines = []
for i in range(len(lines)):
if i > 0 and lines[i] == "{":
modified_lines[-1] += " {"
else:
modified_lines.append(lines[i])
return modified_lines
def perform_indentation(lines):
"""Indents the lines according to their nesting level determined by curly brackets."""
indented_lines = []
current_indent = 0
for line in lines:
if not line.startswith("#") and line.endswith('}') and current_indent > 0:
current_indent -= 1
if line != "":
indented_lines.append(current_indent * INDENTATION + line)
else:
indented_lines.append("")
if not line.startswith("#") and line.endswith('{'):
current_indent += 1
return indented_lines
def format_config_contents(contents):
"""Accepts the string containing nginx configuration and returns formatted one. Adds newline at the end."""
contents = apply_bracket_template_tags(contents)
lines = contents.splitlines()
lines = clean_lines(lines)
lines = join_opening_bracket(lines)
lines = perform_indentation(lines)
text = '\n'.join(lines)
text = strip_bracket_template_tags(text)
for pattern, substitute in ((r'\n{3,}', '\n\n\n'), (r'^\n', ''), (r'\n$', '')):
text = re.sub(pattern, substitute, text, re.MULTILINE)
return text + '\n'
def format_config_file(file_path, original_backup_file_path=None, verbose=True):
"""
Performs the formatting on the given file. The function tries to detect file encoding first.
:param file_path: path to original nginx configuration file. This file will be overridden.
:param original_backup_file_path: optional path, where original file will be backed up.
:param verbose: show messages
"""
encodings = ('utf-8', 'latin1')
encoding_failures = []
chosen_encoding = None
for enc in encodings:
try:
with codecs.open(file_path, 'r', encoding=enc) as rfp:
original_file_content = rfp.read()
chosen_encoding = enc
break
except ValueError as e:
encoding_failures.append(e)
if chosen_encoding is None:
raise Exception('none of encodings %s are valid for file %s. Errors: %s'
% (encodings, file_path, [e.message for e in encoding_failures]))
assert original_file_content is not None
with codecs.open(file_path, 'w', encoding=chosen_encoding) as wfp:
wfp.write(format_config_contents(original_file_content))
if verbose:
print("Formatted file '%s' (detected encoding %s)." % (file_path, chosen_encoding))
if original_backup_file_path:
with codecs.open(original_backup_file_path, 'w', encoding=chosen_encoding) as wfp:
wfp.write(original_file_content)
if verbose:
print("Original saved to '%s'." % original_backup_file_path)
if __name__ == "__main__":
arg_parser = argparse.ArgumentParser(description=__doc__)
arg_parser.add_argument("-v", "--verbose", action="store_true", help="show formatted file names")
arg_parser.add_argument("-b", "--backup-original", action="store_true", help="backup original config file")
arg_parser.add_argument("config_files", nargs='+', help="configuration files to format")
args = arg_parser.parse_args()
for config_file_path in args.config_files:
backup_file_path = config_file_path + '~' if args.backup_original else None
format_config_file(config_file_path, backup_file_path, args.verbose)