Dockery/Trilium
1
0
Fork 0
mirror of https://github.com/zadam/trilium.git synced 2026-07-07 08:42:16 +02:00
Code Issues Packages Projects Releases Activity

chore(ocr): optimize search algorithm

Browse Source 
OCRContentExpression now takes all tokens as an array (like NoteContentFulltextExp), iterates over the input note set from becca, and checks text representations in-memory — zero SQL queries.
parse.ts creates a single OCRContentExpression(tokens) instead of N separate instances.
The LIMIT 50 and the N+1 blob→note/attachment queries are gone entirely.
This commit is contained in:
Elian Doran
2026-04-02 20:54:22 +03:00
parent b747402352
commit 30ba36894d
2 changed files with 43 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files

109
apps/server/src/services/search/expressions/ocr_content.ts
View File

@@ -1,98 +1,69 @@
import Expression from "./expression.js";
import SearchContext from "../search_context.js";
import NoteSet from "../note_set.js";
import sql from "../../sql.js";
import becca from "../../../becca/becca.js";
import NoteSet from "../note_set.js";
import type SearchContext from "../search_context.js";
import Expression from "./expression.js";
/**
* Search expression for finding text within OCR-extracted content from images
* Search expression for finding text within OCR-extracted content (textRepresentation)
* from image notes and their attachments.
*/
export default class OCRContentExpression extends Expression {
private searchText: string;
private tokens: string[];
constructor(searchText: string) {
constructor(tokens: string[]) {
super();
this.searchText = searchText;
this.tokens = tokens;
}
execute(inputNoteSet: NoteSet, executionContext: object, searchContext: SearchContext): NoteSet {
const resultNoteSet = new NoteSet();
const ocrResults = this.searchOCRContent(this.searchText);
for (const ocrResult of ocrResults) {
// Find notes that use this blob
const notes = sql.getRows<{noteId: string}>(`
SELECT noteId FROM notes
WHERE blobId = ? AND isDeleted = 0
`, [ocrResult.blobId]);
for (const noteRow of notes) {
const note = becca.getNote(noteRow.noteId);
if (note && !note.isDeleted && inputNoteSet.hasNoteId(note.noteId)) {
resultNoteSet.add(note);
}
}
// Find attachments that use this blob and their parent notes
const attachments = sql.getRows<{ownerId: string}>(`
SELECT ownerId FROM attachments
WHERE blobId = ? AND isDeleted = 0
`, [ocrResult.blobId]);
for (const attachmentRow of attachments) {
const note = becca.getNote(attachmentRow.ownerId);
if (note && !note.isDeleted && inputNoteSet.hasNoteId(note.noteId)) {
resultNoteSet.add(note);
}
for (const note of inputNoteSet.notes) {
if (this.noteMatchesOCR(note.noteId)) {
resultNoteSet.add(note);
}
}
// Add highlight tokens for OCR matches
if (ocrResults.length > 0) {
const tokens = this.extractHighlightTokens(this.searchText);
searchContext.highlightedTokens.push(...tokens);
if (resultNoteSet.notes.length > 0) {
const highlightTokens = this.tokens
.filter(token => token.length > 2)
.map(token => token.toLowerCase());
searchContext.highlightedTokens.push(...highlightTokens);
}
return resultNoteSet;
}
/**
* Check if a note (or its attachments) has OCR text matching all tokens.
*/
private noteMatchesOCR(noteId: string): boolean {
const note = becca.notes[noteId];
if (!note) return false;
private searchOCRContent(searchText: string): Array<{
blobId: string;
textRepresentation: string;
}> {
try {
// Search in blobs table for OCR text
const query = `
SELECT blobId, textRepresentation
FROM blobs
WHERE textRepresentation LIKE ?
AND textRepresentation IS NOT NULL
AND textRepresentation != ''
LIMIT 50
`;
const params = [`%${searchText}%`];
// Collect all textRepresentation values for this note
const texts: string[] = [];
return sql.getRows<{
blobId: string;
textRepresentation: string;
}>(query, params);
} catch (error) {
console.error('Error searching OCR content:', error);
return [];
const noteBlob = becca.getBlob({ blobId: note.blobId });
if (noteBlob?.textRepresentation) {
texts.push(noteBlob.textRepresentation.toLowerCase());
}
}
for (const attachment of note.getAttachments()) {
const blob = becca.getBlob({ blobId: attachment.blobId });
if (blob?.textRepresentation) {
texts.push(blob.textRepresentation.toLowerCase());
}
}
private extractHighlightTokens(searchText: string): string[] {
// Split search text into words and return them as highlight tokens
return searchText
.split(/\s+/)
.filter(token => token.length > 2)
.map(token => token.toLowerCase());
if (texts.length === 0) return false;
// All tokens must appear in at least one of the text representations
const combined = texts.join(" ");
return this.tokens.every(token => combined.includes(token.toLowerCase()));
}
toString(): string {
return `OCRContent('${this.searchText}')`;
return `OCRContent('${this.tokens.join("', '")}')`;
}
}
}

13
apps/server/src/services/search/services/parse.ts
View File

@@ -50,24 +50,17 @@ function getFulltext(_tokens: TokenData[], searchContext: SearchContext, leading
new NoteContentFulltextExp("=", { tokens, flatText: true })
];
// Add OCR content search for each token
for (const token of tokens) {
exactMatchExpressions.push(new OCRContentExpression(token));
}
exactMatchExpressions.push(new OCRContentExpression(tokens));
return new OrExp(exactMatchExpressions);
}
const searchExpressions: Expression[] = [
new NoteFlatTextExp(tokens),
new NoteContentFulltextExp(operator, { tokens, flatText: true })
new NoteContentFulltextExp(operator, { tokens, flatText: true }),
new OCRContentExpression(tokens)
];
// Add OCR content search for each token
for (const token of tokens) {
searchExpressions.push(new OCRContentExpression(token));
}
return new OrExp(searchExpressions);
}
return new NoteFlatTextExp(tokens);