Find and replace text — literal, regex, capture groups and a list of rules

You get a sentence, and the tab keeps working. Some patterns are not slow but exponential: (a+)+b against a run of a characters with no b at the end makes the engine try every way of dividing that run between the two quantifiers, and thirty characters is already more than a billion attempts. A regular expression running on a page’s own thread cannot be interrupted — no timer fires, no error is thrown, control never comes back, and an input limit or a match limit never gets the chance to run. So regex mode here does not run on this page’s thread. The pattern, the flags and your text go to a Web Worker, the page starts a 1.5-second clock, and if the worker has not answered by then the page calls terminate() on it and tells you the pattern was too expensive, naming catastrophic backtracking as the likely cause. Terminating a worker is the only thing in a browser that reliably stops a running regular expression. Your text is untouched and still in the box. Literal mode never needs any of this, because a fully escaped literal pattern cannot backtrack.

The sentence under the box is the browser’s own message and it usually names the problem. Three catch nearly everyone: “[” opens a character class that has to be closed with “]”; “(” opens a group that has to be closed with “)”; and a “*” or “+” at the very start has nothing in front of it to repeat. If you want one of . * + ? ^ $ { } ( ) | [ ] \ to mean itself, put a backslash in front of it — or switch regex off, where every character already means itself and nothing has to be escaped.

Round brackets in the pattern remember the text they matched, and $1, $2 and so on write it back out in the replacement. To swap “Lim, Ahmad” into “Ahmad Lim”, find (\w+), (\w+) and replace with $2 $1. $& stands for the whole match, so replacing with [$&] wraps every match in square brackets, and $$ produces one literal dollar sign — which matters on this page, because the sample text is full of currency. Paste those two names into the box above and watch the preview highlight the two captured runs before you press anything.

Two reasons. Wrapping your pattern in a boundary group would renumber your capture groups, so $1 would silently stop meaning what you wrote — a bug that produces plausible wrong output rather than an error. And in regex mode you already have \b, alternation and lookahead to describe exactly the boundary you want. There is a third thing worth knowing: \b in JavaScript is defined over [A-Za-z0-9_] only, so it is wrong the moment an accented letter is involved. /\bcafé\b/.test("café") is false, which you can check in a console in one line. The Whole word switch in literal mode is built on \p{L} and \p{N} instead, so café and Müller behave the way you would expect.

i ignores case. m makes ^ and $ match at the start and end of every line instead of the whole text, which is what you want for anything line-oriented. s makes . match a line break, which it otherwise never does — this is the flag people are missing when a pattern works within one line and returns nothing across two. u turns on Unicode mode, and it changes meaning rather than adding it: \p{L} means “any letter” only when u is on, and without it \p is just the letter p. g is always on here, because replacing only the first match is not what a Replace all should do.

On a page that reuses one regular expression object, that is exactly what happens: a /g regex carries a lastIndex from one call to the next, so the second call starts searching from where the first one stopped. It is a one-line demonstration — const re = /a/g; re.test("a") is true, and running re.test("a") again is false. This page builds a fresh RegExp for every run and for every rule, so the answer never depends on what you did before. If a result really does change here, it is because a rule earlier in your list changed the text that the later rule sees.

Yes, but not with a dot on its own. “.” matches anything except a line break, which is why a pattern that works inside one line finds nothing across two. Turn on the s flag and “.” includes line breaks. To match the break itself, write \n — or \r?\n if the text might have come from Windows, where a line ends with two characters rather than one. And if you want ^ and $ to mean the start and end of a line rather than of the whole text, that is the m flag, not s.

Applying a fixed list of substitutions in one pass: renaming a set of terms across a document, converting a column of codes, normalising the several spellings a form has collected. Every rule runs over the output of the rule above it, which is the point and also the one trap. Replace cat with dog, then dog with bird, and you finish with no cats and no dogs — because by the time the second rule runs, the cats have already become dogs. Order the list so that nothing you produce is something a later rule is hunting for, or route the risky one through a placeholder string that appears nowhere else.

Nowhere. Matching and replacing both run in JavaScript this page has already loaded, there is no upload, and nothing is written to localStorage — reload and the box is empty. It is worth stating plainly given what gets pasted into a find-and-replace: log extracts, configuration files, SQL dumps and CSV exports, which routinely carry names, addresses, internal hostnames and keys.