Regex New Added 19 September 2026

Python: re.error: look-behind requires fixed-width pattern

The standard re module only implements lookbehind when every alternative is the same known length, so a quantifier or an alternation of different widths inside (?<=...) is rejected at compile time. The pattern is valid in .NET and in the third party regex module, which is why one copied from a tutorial can fail here.

Quick fix

Read the commands before running them. Anything that restarts a service, deletes data or changes permissions should be tried on a non-production system first.

Quick fix

# Fails: \s* has no fixed width
import re
re.compile(r'(?<=Total:\s*)\d+')

# Capture instead of looking behind, and read group 1
m = re.search(r'Total:\s*(\d+)', text)
print(m.group(1) if m else None)

# Alternations are allowed only if every branch is the same length
re.compile(r'(?<=cat|dog)s')     # fine, both are three characters
re.compile(r'(?<=cat|bird)s')    # rejected

# Keep the lookbehind and drop the variable part into a fixed one
re.compile(r'(?<=Total: )\d+')

# Or use the regex module, which supports variable width lookbehind
# pip install regex
import regex
regex.search(r'(?<=Total:\s*)\d+', text)

# In sub(), a capture group is usually simpler than any lookaround:
re.sub(r'(Total:\s*)\d+', r'\g<1>0', text)

How to diagnose Regex errors

Regex problems come in two flavours: it does not match what you expect (usually escaping, greediness, or an engine feature difference) and it matches but takes forever. The second is catastrophic backtracking, a genuine denial-of-service vector known as ReDoS, caused by nested quantifiers over overlapping character classes, such as (a+)+b. Any regex applied to untrusted input should be checked for it.

If the quick fix above does not resolve it, work through these steps. They apply to this whole class of error, not just to this one message, which is usually what saves the time.

Test against a visualiser that shows backtracking steps, and always test with input that fails to match. That is where catastrophic backtracking appears, not on successful matches.
Eliminate nested quantifiers over overlapping classes. Rewrite (a+)+ as a+, and prefer possessive quantifiers or atomic groups where the engine supports them.
Know your engine: lookbehind, named groups and Unicode property escapes differ between PCRE, RE2, JavaScript, Python and Go. RE2 (used by Go) deliberately has no backtracking and rejects some patterns outright.
Anchor patterns with ^ and $ where you mean a whole-string match. Unanchored patterns silently match substrings.
Set a timeout or use a linear-time engine for user-supplied patterns. Never run an untrusted regex on a request thread.

Tools worth reaching for

regex101.com
re2 / RE2J
Python re.DEBUG
grep -P for PCRE testing
recheck / redos linters

Authoritative references

Primary documentation for this error, worth reading before applying any fix in production.

docs.python.org docs.python.org

Related Regex errors

See all 11 Regex errors →

Browse other categories

Something missing or wrong?

This entry is maintained by hand. If the fix is out of date, incomplete, or you have a better one, email a correction and it will be reviewed.