regex（regex 搜索）

Greedy vs Lazy

• .* is greedy—matches as much as possible; .*? is lazy—matches minimum
• Greedy often overshoots: <.*> on <a>b</a> matches entire string, not <a>
• Default quantifiers + * {n,} are greedy—add ? for lazy: +? *? {n,}?

Escaping

• Metacharacters need escape: \. \* \+ \? \[ \] \( \) \{ \} \| \\ \^ \$
• Inside character class []: only ], \, ^, - need escape (and ^ only at start, - only mid)
• Literal backslash: \\ in regex, but in strings often need \\\\ (double escape)

Anchors

• ^ start, $ end—but behavior changes with multiline flag
• Multiline mode: ^ $ match line starts/ends; without, only string start/end
• \A always string start, \Z always string end (not all engines)
• Word boundary \b matches position, not character—\bword\b for whole words

Character Classes

• [abc] matches one of a, b, c; [^abc] matches anything except a, b, c
• Ranges: [a-z] [0-9]—but [a-Z] is invalid (ASCII order matters)
• Shorthand: \d digit, \w word char, \s whitespace; uppercase negates: \D \W \S
• . matches any char except newline—use [\s\S] for truly any, or s flag if available

Groups

• Capturing () vs non-capturing (?:)—use (?:) when you don't need backreference
• Named groups: (?<name>...) or (?P<name>...) depending on engine
• Backreferences: \1 \2 refer to captured groups in same pattern
• Groups also establish scope for alternation: cat|dog vs ca(t|d)og

Lookahead & Lookbehind

• Positive lookahead (?=...): assert what follows, don't consume
• Negative lookahead (?!...): assert what doesn't follow
• Positive lookbehind (?<=...): assert what precedes
• Negative lookbehind (?<!...): assert what doesn't precede
• Lookbehinds must be fixed-width in most engines—no * or + inside

Flags

• i case-insensitive, m multiline (^$ match lines), g global (find all)
• s (dotall): . matches newline—not supported everywhere
• u unicode: enables \p{} properties, proper surrogate handling
• Flags syntax varies: /pattern/flags (JS), (?flags) inline, or function arg (Python re.I)

Engine Differences

• JavaScript: no lookbehind until ES2018; no \A \Z; no possessive quantifiers
• Python re: uses (?P<name>) for named groups; no \p{} without regex module
• PCRE (PHP, grep -P): full features; possessive ++ *+; recursive patterns
• Go: RE2 engine, no backreferences, no lookahead—guaranteed linear time

Performance

• Catastrophic backtracking: (a+)+ against aaaaaaaaaab is exponential—avoid nested quantifiers
• Possessive quantifiers ++ *+ prevent backtracking—use when backtracking pointless
• Atomic groups (?>...) don't give back chars—similar to possessive
• Anchor patterns when possible—^prefix is O(1), unanchored prefix is O(n)

Common Mistakes

• Email validation: RFC-compliant regex is 6000+ chars—use simple check or library
• URL matching: edge cases are endless—use URL parser, regex for quick extraction only
• Don't use regex for HTML/XML—use a parser; regex can't handle nesting
• Forgetting to escape user input—regex injection is real; use literal escaping functions

Testing

• Test edge cases: empty string, special chars, unicode, very long input
• Visualize with tools: regex101.com shows matches and explains
• Check which engine documentation you're reading—features vary significantly