31 lines
1.5 KiB
Dart
31 lines
1.5 KiB
Dart
|
// lib/editor/search/search_text.dart
|
||
|
|
//
|
||
|
|
// Pure text normalization + matching for full-text search (F8). PDF text layers
|
||
|
|
// and OCR output are full of hard line breaks and irregular whitespace, so a
|
||
|
|
// query like "hello world" won't substring-match raw extracted text that reads
|
||
|
|
// "hello\nworld". Normalizing both sides (lowercase + collapse every whitespace
|
||
|
|
// run to a single space + trim) fixes that.
|
||
|
|
//
|
||
|
|
// CJK NOTE: this user writes Chinese. We deliberately do NOT word-tokenize —
|
||
|
|
// Chinese has no inter-word spaces, so a whitespace/punctuation tokenizer would
|
||
|
|
// mangle it. Substring matching over normalized text is correct for both Latin
|
||
|
|
// and CJK; word/段 segmentation belongs in the DB FTS tokenizer (trigram /
|
||
|
|
// unicode61), not here.
|
||
|
|
|
||
|
|
/// Matches any run of Unicode whitespace (spaces, tabs, newlines, NBSP, …).
|
||
|
|
final RegExp _whitespaceRun = RegExp(r'\s+');
|
||
|
|
|
||
|
|
/// Normalize [text] for indexing/matching: lowercase, collapse whitespace runs
|
||
|
|
/// (incl. the hard newlines PDF/OCR insert mid-sentence) to single spaces, trim.
|
||
|
|
String normalizeForIndex(String text) {
|
||
|
|
return text.toLowerCase().replaceAll(_whitespaceRun, ' ').trim();
|
||
|
|
}
|
||
|
|
|
||
|
|
/// Whether [source] contains [query] after both are normalized — so a match can
|
||
|
|
/// span the line breaks present in the raw text. Empty query never matches.
|
||
|
|
bool matchesNormalized(String source, String query) {
|
||
|
|
final q = normalizeForIndex(query);
|
||
|
|
if (q.isEmpty) return false;
|
||
|
|
return normalizeForIndex(source).contains(q);
|
||
|
|
}
|