CSS file of 320 bytes takes 10+ seconds to parse

[vdmit]

I've encountered CSS file which cssutils library takes lots of seconds to parse (in fact, almost infinite).
After some investigation I've minified problematic sample (as possible) and prepared reproducible code example.
On my laptop (core i7-8550U) it takes 14 seconds to parse file with just one line and 320 bytes of text.

Original file has 20kBytes of text and parser does not finish it in several hours.

Code:
https://gist.github.com/vdmit/ef9007170fa1c616cf5aba1fcebfce87
I'm using:

• latest stable cssutils release 2.3.0 from pypi repo
• python 3.8.10
• Ubuntu 20.04 linux distribution.

Notes:

• original CSS file was malformed and even contained non-printable characters. My minified example is still not valid CSS file, but it is a plain ANSI file.
• Adding line break in any place fixes problem;
• Removing of last style expression (.s11... reduces execution time from 14 seconds to 3 seconds).
• Adding of one more style expression increases execution time from 14 seconds to 50, 2 more styles yields ~260 seconds.

So, there is a exponential complexity somewhere, which is strange.

Traceback of interrupted script is following:

Traceback (most recent call last):
  File "cssutils_infinite_loop_bug_example.py", line 10, in <module>
    sheet = css_parser.parseString(cssText=css_text)
  File "/home/vdmit/.local/lib/python3.8/site-packages/cssutils/parse.py", line 147, in parseString
    sheet._setCssTextWithEncodingOverride(
  File "/home/vdmit/.local/lib/python3.8/site-packages/cssutils/css/cssstylesheet.py", line 408, in _setCssTextWithEncodingOverride
    self.cssText = cssText
  File "/home/vdmit/.local/lib/python3.8/site-packages/cssutils/css/cssstylesheet.py", line 331, in _setCssText
    wellformed, expected = self._parse(
  File "/home/vdmit/.local/lib/python3.8/site-packages/cssutils/util.py", line 484, in _parse
    expected = p(expected, seq, token, tokenizer)
  File "/home/vdmit/.local/lib/python3.8/site-packages/cssutils/css/cssstylesheet.py", line 313, in ruleset
    rule.cssText = self._tokensupto2(tokenizer, token)
  File "/home/vdmit/.local/lib/python3.8/site-packages/cssutils/util.py", line 343, in _tokensupto2
    for token in tokenizer:
  File "/home/vdmit/.local/lib/python3.8/site-packages/cssutils/tokenize2.py", line 172, in tokenize
    match = matcher(text, pos)  # if no match try next production
KeyboardInterrupt

[jaraco]

Thanks for the report.

I welcome investigation or a patch.

[InconceivableVizzini]

No patch at the moment, but I believe this is related to backtracking in the regular expressions parsing strings, for the specific input shared. The lengthy call is located at

cssutils/cssutils/tokenize2.py

Line 172 in d9d2bdc

match = matcher(text, pos) # if no match try next production

When the input is ' and the matcher is related to the string production,

cssutils/cssutils/cssproductions.py

Line 32 in d9d2bdc

'string': r'{string1}|{string2}',

[InconceivableVizzini]

This can be fixed by rewriting the string productions (r"'(\\.|[^'])*'") or using an alternative regular expression library that does not rely on backtracking, such as re2, which may be used as a drop-in replacement of re.

[jaraco]

This can be fixed by rewriting the string productions (r"'(\\.|[^'])*'") or using an alternative regular expression library that does not rely on backtracking, such as re2, which may be used as a drop-in replacement of re.

I'd prefer to see a rewrite of the string productions, especially if it can be done in a fairly straightforward way and without adding to much complication to the approach.