Why is my string length wrong in JavaScript?

JavaScript's .length counts UTF-16 code units, not what you see. One family emoji is 1 visible glyph but 7 codepoints, 11 UTF-16 units and 25 UTF-8 bytes. unicodex shows all four counts side by side, with which tools count which, so the mismatch stops being a mystery.

Why don't two identical-looking strings match?

Usually one of three causes: different normalization forms (a precomposed é vs e plus a combining accent), invisible characters (zero-width spaces, bidi controls, no-break spaces), or look-alike letters such as Cyrillic а standing in for Latin a. unicodex's compare mode reports raw-equal, NFC-equal and the first divergent codepoint, and flags the characters that explain a visual match that isn't a real match.

How do I find invisible characters in pasted text?

Paste the text into unicodex. Every flagged character — zero-width spaces, byte-order marks, bidi controls, look-alike spaces, line separators — is revealed inline in a dashed box with its abbreviation, plus its Unicode name, a one-line risk note and a citation to the Unicode document that defines it.

Does unicodex upload or store the text I paste?

No. unicodex runs entirely in your browser with a Content-Security-Policy of connect-src 'none' — the browser itself blocks any network request, so 'pasted text never leaves this browser' is enforced, not promised. Optional autosave keeps your last input in this browser's localStorage only, and one tap clears it.

unicodex — unicode character inspector: codepoints, bytes, invisible characters

Inspect text

Paste or type anything

Samples:

Autosave input in this browser

0 grapheme clusters what you see · Intl.Segmenter · cursor steps

0 UTF-16 units JS .length · Java/C# length · DOM APIs

0 codepoints Python 3 len() · Rust chars().count()

0 UTF-8 bytes on the wire · Go len() · DB storage/limits

Cluster by cluster

Each card is one grapheme cluster — one "character" as you see it. Flagged characters appear as dashed reveal boxes. Select a card for bytes, class chips and copy-as escapes.

Flagged characters

Scanned against a hand-verified catalog of ~70 risky characters (zero-width, bidi/Trojan Source, look-alike spaces, line separators, confusables). Every entry cites the Unicode document that defines it. Absence of a warning is not a "safe" verdict — the confusable list is a curated subset of UTS #39.

Normalization diff — NFC · NFD · NFKC · NFKD

Computed by this browser's String.normalize (UAX #15). "changed" means the form differs from your input; unstable clusters are marked.

Compare two strings

For "these look identical but don't match" bugs: raw equality, NFC equality, the first divergent codepoint, and any flagged characters that explain a visual match that isn't a real match.

String A String B

Clean copy

Mechanical, per-category fixes only — each toggle states exactly what it replaces, every pass is idempotent by test, and nothing here guesses intent or repairs encoding damage (mojibake).

Strip zero-width & format characters — removes ZWSP, WJ, BOM, SHY, CGJ, the four invisible math operators and tag characters. Keeps emoji-sequence ZWJ, ZWNJ between letters (required in Hindi/Persian) and variation selectors. …strip all, including emoji/Indic joiners — also removes every ZWJ, ZWNJ and variation selector (decomposes 👨‍👩‍👧‍👦 into four emoji; corrupts Hindi/Persian orthography). Look-alike spaces → ASCII space — replaces the 16 flagged spaces (NBSP, NNBSP, thin, ideographic…) with U+0020. Strip bidi controls — removes all 12 directional formatting characters (LRE…PDI, LRM, RLM, ALM). Apply NFC — canonical composition via String.normalize('NFC').

Cleaned result

Honest limits

Cluster counts follow this browser's Intl.Segmenter and its Unicode version — e.g. Devanagari conjuncts (नमस्ते) can segment differently on older engines. unicodex reports the host engine's answer; its self-tests pin only version-stable fixtures.
No name database: arbitrary codepoints show U+hex plus engine-derived class chips, not the official Unicode name. Names appear only for the ~70 flagged catalog entries.
Confusable flags are a curated ~24-letter subset of UTS #39 — absence of a warning is not proof a string is safe, and unicodex never says "safe".
Rendering varies by device fonts — a cluster shown here may draw as tofu or differently elsewhere. unicodex inspects what text is, not how any device draws it.
Clean-copy performs only the listed mechanical replacements; it never guesses intent and cannot repair encoding damage (mojibake).
Autosaved input lives in this browser's localStorage (toggleable, one-tap clear); clearing site data erases it. Nothing is ever transmitted — enforced by CSP connect-src 'none'.

Unicode Character Inspector — Codepoints, Bytes, Invisible-Character Detector & Normalization Diff

Inspect text

Cluster by cluster

Cluster detail

Flagged characters

Normalization diff — NFC · NFD · NFKC · NFKD

Compare two strings

Clean copy

Honest limits