@devries-cameron20 / t3code Mirror · read-only / commits / becf853def
Fork 0

The best way to control Claude Code, Codex and more from any computer or phone

Mirrored from https://github.com/Camo67/t3code.git

Synced Nov 22, 15:45 UTC

fix(web): PDF search finds phrases that wrap across lines (#25702)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Julius Marminge committed Oct 26, 2026 at 18:02 UTC becf853defb6127dbe4447860383777d69346b6f
3 files changed +92 -13
apps/web/src/components/files/PdfViewer.tsx
+2 -1
@@ -15,6 +15,7 @@ import { readBoundedDocument } from "./readBoundedDocument";
15 import { openPdf } from "./pdfDocument";
16 import {
17 currentPdfPage,
18 + pdfFindIncludes,
19 pdfFindMatchRuns,
20 pdfPageOffsets,
21 pdfPageScrollTop,
@@ -247,7 +248,7 @@ export default function PdfViewer(props: { src: string; title: string }) {
248 throw new Error(
249 "This PDF has too much text to search. Save it to search in another app.",
250 );
250 - if (text.toLowerCase().includes(query.toLowerCase())) found.push(index);
251 + if (pdfFindIncludes(text, query)) found.push(index);
252 }
253 if (cancelled) return;
254 setRatios(sizes);
apps/web/src/components/files/pdfLayout.test.ts
+31
@@ -1,6 +1,7 @@
1 import { describe, expect, it } from "vite-plus/test";
2 import {
3 currentPdfPage,
4 + pdfFindIncludes,
5 pdfFindMatchRuns,
6 pdfPageOffsets,
7 pdfPageScrollTop,
@@ -105,6 +106,36 @@ describe("pdfFindMatchRuns", () => {
106 ]);
107 });
108
109 + it("matches a space in the query across a line end", () => {
110 + expect(
111 + pdfFindMatchRuns(["Q3 quarterly", "\n", "outlook", " ahead"], "quarterly outlook"),
112 + ).toEqual([true, true, true, false]);
113 + });
114 +
115 + it("matches a word hyphenated across a line end", () => {
116 + expect(pdfFindMatchRuns(["the quar-", "\n", "terly report"], "quarterly")).toEqual([
117 + true,
118 + false,
119 + true,
120 + ]);
121 + });
122 +
123 + it("drops a line-end hyphen after a letter outside the BMP", () => {
124 + expect(pdfFindMatchRuns(["\u{10400}-", "\n", "\u{10401}"], "\u{10400}\u{10401}")).toEqual([
125 + true,
126 + false,
127 + true,
128 + ]);
129 + });
130 +
131 + it("finds the same pages the highlights do", () => {
132 + const page = "Q3 quarterly\noutlook ahead, the quar-\nterly report";
133 + expect(pdfFindIncludes(page, "Quarterly OUTLOOK")).toBe(true);
134 + expect(pdfFindIncludes(page, "quarterly report")).toBe(true);
135 + expect(pdfFindIncludes(page, "bud-get")).toBe(false);
136 + expect(pdfFindIncludes(page, " ")).toBe(false);
137 + });
138 +
139 it("highlights nothing for a blank or missing query", () => {
140 expect(pdfFindMatchRuns(["a b", "c"], " ")).toEqual([false, false]);
141 expect(pdfFindMatchRuns(["a b", "c"], "")).toEqual([false, false]);
apps/web/src/components/files/pdfLayout.ts
+59 -12
@@ -56,18 +56,65 @@ export function pdfPageText(
56 return items.map((item) => ("str" in item ? item.str + (item.hasEOL ? "\n" : "") : "")).join("");
57 }
58
59 +/**
60 + * Lowercased search text with each whitespace run, line ends included, read as one space, and a
61 + * hyphen that ends a line between letters dropped, as pdf.js's own find does. `owner` maps each
62 + * character back to the run it came from.
63 + */
64 +function pdfFindText(runs: readonly string[]) {
65 + const chars = runs.flatMap((run, index) =>
66 + Array.from(run.toLowerCase(), (char) => ({ char, run: index })),
67 + );
68 + let text = "";
69 + let last = "";
70 + const owner: number[] = [];
71 + for (let index = 0; index < chars.length; index++) {
72 + const { char, run } = chars[index]!;
73 + if (/\s/.test(char)) {
74 + if (!text.endsWith(" ")) {
75 + text += " ";
76 + last = " ";
77 + owner.push(run);
78 + }
79 + } else if (
80 + char === "-" &&
81 + chars[index + 1]?.char === "\n" &&
82 + /\p{L}/u.test(last) &&
83 + /\p{L}/u.test(chars[index + 2]?.char ?? "")
84 + ) {
85 + index++;
86 + } else {
87 + text += char;
88 + last = char;
89 + for (let unit = 0; unit < char.length; unit++) owner.push(run);
90 + }
91 + }
92 + return { text, owner };
93 +}
94 +
95 +/** Whether a page's text holds the query, under the same rules as `pdfFindMatchRuns`, without
96 + * mapping characters back to runs: the page scan checks every page on each query. */
97 +export function pdfFindIncludes(text: string, query: string) {
98 + const needle = query.toLowerCase().replace(/\s+/g, " ");
99 + if (!needle.trim()) return false;
100 + return text
101 + .toLowerCase()
102 + .replace(/(\p{L})-\n(?=\p{L})/gu, "$1")
103 + .replace(/\s+/g, " ")
104 + .includes(needle);
105 +}
106 +
107 /** Which text runs hold part of a case-insensitive match, when pdf.js splits words across spans. */
108 export function pdfFindMatchRuns(runs: readonly string[], query: string) {
61 - const texts = runs.map((run) => run.toLowerCase());
62 - const text = texts.join("");
63 - const needle = query.toLowerCase();
64 - let match = needle.trim() ? text.indexOf(needle) : -1;
65 - let offset = 0;
66 - return texts.map((run) => {
67 - while (match !== -1 && match + needle.length <= offset)
68 - match = text.indexOf(needle, match + needle.length);
69 - const hit = match !== -1 && match < offset + run.length;
70 - offset += run.length;
71 - return hit;
72 - });
109 + const { text, owner } = pdfFindText(runs);
110 + const needle = query.toLowerCase().replace(/\s+/g, " ");
111 + const hits = runs.map(() => false);
112 + if (!needle.trim()) return hits;
113 + for (
114 + let match = text.indexOf(needle);
115 + match !== -1;
116 + match = text.indexOf(needle, match + needle.length)
117 + )
118 + for (let index = match; index < match + needle.length; index++) hits[owner[index]!] = true;
119 + return hits;
120 }