open-agreements
Version:
Open-source legal template filling CLI and library
63 lines • 3.05 kB
JavaScript
import { createHash } from 'node:crypto';
/**
* Cf-category and bidi/joiner invisibles that change a paragraph's bytes
* without changing its rendered surface in any reader Word/Pages/Google Docs
* actually ships. Stripping them before hashing means a paragraph that picked
* up a stray soft hyphen (cooperate vs cooperate) or a ZWJ (AB vs AB)
* during round-tripping still fingerprints to the same value as the clean copy.
*
* Range covered (each value is a single Unicode code point):
* U+00AD SOFT HYPHEN
* U+200B ZERO WIDTH SPACE
* U+200C ZERO WIDTH NON-JOINER
* U+200D ZERO WIDTH JOINER
* U+200E LEFT-TO-RIGHT MARK
* U+200F RIGHT-TO-LEFT MARK
* U+202A..U+202E bidi embedding/override controls (LRE, RLE, PDF, LRO, RLO)
* U+FE00..U+FE0F variation selectors VS1..VS16
* U+FEFF ZERO WIDTH NO-BREAK SPACE / BYTE ORDER MARK
*
* NOT stripped on purpose:
* - curly quotes (U+2018/U+2019/U+201C/U+201D), en/em dashes (U+2013/U+2014),
* ellipsis (U+2026), etc. Citation systems legitimately distinguish these
* from ASCII variants. NFKC already handles compatibility decompositions
* (ligatures, full-width Latin, NBSP) — that is its job, not ours.
*/
const INVISIBLE_FORMAT_CHARS_RE = /[--︀-️]/g;
/**
* Compute a portable content fingerprint for a paragraph's raw visible text.
*
* Algorithm: `sha256:nfkc:` + first 32 hex chars of
* `sha256( stripInvisibles(NFKC(text)).replace(/\s+/g, " ").trim() )`.
*
* Order matters:
* 1. NFKC first so compatibility whitespace (NBSP, ideographic space) is
* normalized to ASCII space and ligatures/full-width Latin decompose.
* 2. Strip Cf-category invisibles (soft hyphen, ZWJ, ZWNJ, LRM/RLM, bidi
* controls, variation selectors, BOM) so byte-level round-trip noise
* doesn't change the hash without changing the visible glyphs.
* 3. Whitespace collapse + trim.
*
* Case is preserved — citation systems need to distinguish "Section 5" from
* "section 5". Curly quotes / en-dashes / ellipses are NOT folded to ASCII;
* legal practice distinguishes them.
*
* The output is read-only metadata, not an edit anchor. Edit tools accept only
* `_bk_*` paragraph IDs.
*
* The 128-bit truncation (32 hex) is collision-safe at any realistic
* citation-corpus scale; the `sha256:nfkc:` prefix reserves space for future
* algorithm bumps (e.g. `sha256:nfkc-strip:`). Consumers SHOULD store and
* compare the full prefixed string so an algorithm bump cleanly invalidates
* old hashes.
*/
export function computeContentFingerprint(rawVisibleText) {
const normalized = rawVisibleText
.normalize('NFKC')
.replace(INVISIBLE_FORMAT_CHARS_RE, '')
.replace(/\s+/g, ' ')
.trim();
const hex = createHash('sha256').update(normalized).digest('hex').slice(0, 32);
return `sha256:nfkc:${hex}`;
}
//# sourceMappingURL=content_fingerprint.js.map