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

apps/web/src/components/files/PdfViewer.tsx

@@ -15,6 +15,7 @@ import { readBoundedDocument } from "./readBoundedDocument";
 import { openPdf } from "./pdfDocument";
 import {
   currentPdfPage,
+  pdfFindIncludes,
   pdfFindMatchRuns,
   pdfPageOffsets,
   pdfPageScrollTop,
@@ -247,7 +248,7 @@ export default function PdfViewer(props: { src: string; title: string }) {
             throw new Error(
               "This PDF has too much text to search. Save it to search in another app.",
             );
-          if (text.toLowerCase().includes(query.toLowerCase())) found.push(index);
+          if (pdfFindIncludes(text, query)) found.push(index);
         }
         if (cancelled) return;
         setRatios(sizes);

apps/web/src/components/files/pdfLayout.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it } from "vite-plus/test";
 import {
   currentPdfPage,
+  pdfFindIncludes,
   pdfFindMatchRuns,
   pdfPageOffsets,
   pdfPageScrollTop,
@@ -105,6 +106,36 @@ describe("pdfFindMatchRuns", () => {
     ]);
   });
 
+  it("matches a space in the query across a line end", () => {
+    expect(
+      pdfFindMatchRuns(["Q3 quarterly", "\n", "outlook", " ahead"], "quarterly outlook"),
+    ).toEqual([true, true, true, false]);
+  });
+
+  it("matches a word hyphenated across a line end", () => {
+    expect(pdfFindMatchRuns(["the quar-", "\n", "terly report"], "quarterly")).toEqual([
+      true,
+      false,
+      true,
+    ]);
+  });
+
+  it("drops a line-end hyphen after a letter outside the BMP", () => {
+    expect(pdfFindMatchRuns(["\u{10400}-", "\n", "\u{10401}"], "\u{10400}\u{10401}")).toEqual([
+      true,
+      false,
+      true,
+    ]);
+  });
+
+  it("finds the same pages the highlights do", () => {
+    const page = "Q3 quarterly\noutlook ahead, the quar-\nterly report";
+    expect(pdfFindIncludes(page, "Quarterly  OUTLOOK")).toBe(true);
+    expect(pdfFindIncludes(page, "quarterly report")).toBe(true);
+    expect(pdfFindIncludes(page, "bud-get")).toBe(false);
+    expect(pdfFindIncludes(page, " ")).toBe(false);
+  });
+
   it("highlights nothing for a blank or missing query", () => {
     expect(pdfFindMatchRuns(["a b", "c"], " ")).toEqual([false, false]);
     expect(pdfFindMatchRuns(["a b", "c"], "")).toEqual([false, false]);

apps/web/src/components/files/pdfLayout.ts

@@ -56,18 +56,65 @@ export function pdfPageText(
   return items.map((item) => ("str" in item ? item.str + (item.hasEOL ? "\n" : "") : "")).join("");
 }
 
+/**
+ * Lowercased search text with each whitespace run, line ends included, read as one space, and a
+ * hyphen that ends a line between letters dropped, as pdf.js's own find does. `owner` maps each
+ * character back to the run it came from.
+ */
+function pdfFindText(runs: readonly string[]) {
+  const chars = runs.flatMap((run, index) =>
+    Array.from(run.toLowerCase(), (char) => ({ char, run: index })),
+  );
+  let text = "";
+  let last = "";
+  const owner: number[] = [];
+  for (let index = 0; index < chars.length; index++) {
+    const { char, run } = chars[index]!;
+    if (/\s/.test(char)) {
+      if (!text.endsWith(" ")) {
+        text += " ";
+        last = " ";
+        owner.push(run);
+      }
+    } else if (
+      char === "-" &&
+      chars[index + 1]?.char === "\n" &&
+      /\p{L}/u.test(last) &&
+      /\p{L}/u.test(chars[index + 2]?.char ?? "")
+    ) {
+      index++;
+    } else {
+      text += char;
+      last = char;
+      for (let unit = 0; unit < char.length; unit++) owner.push(run);
+    }
+  }
+  return { text, owner };
+}
+
+/** Whether a page's text holds the query, under the same rules as `pdfFindMatchRuns`, without
+ * mapping characters back to runs: the page scan checks every page on each query. */
+export function pdfFindIncludes(text: string, query: string) {
+  const needle = query.toLowerCase().replace(/\s+/g, " ");
+  if (!needle.trim()) return false;
+  return text
+    .toLowerCase()
+    .replace(/(\p{L})-\n(?=\p{L})/gu, "$1")
+    .replace(/\s+/g, " ")
+    .includes(needle);
+}
+
 /** Which text runs hold part of a case-insensitive match, when pdf.js splits words across spans. */
 export function pdfFindMatchRuns(runs: readonly string[], query: string) {
-  const texts = runs.map((run) => run.toLowerCase());
-  const text = texts.join("");
-  const needle = query.toLowerCase();
-  let match = needle.trim() ? text.indexOf(needle) : -1;
-  let offset = 0;
-  return texts.map((run) => {
-    while (match !== -1 && match + needle.length <= offset)
-      match = text.indexOf(needle, match + needle.length);
-    const hit = match !== -1 && match < offset + run.length;
-    offset += run.length;
-    return hit;
-  });
+  const { text, owner } = pdfFindText(runs);
+  const needle = query.toLowerCase().replace(/\s+/g, " ");
+  const hits = runs.map(() => false);
+  if (!needle.trim()) return hits;
+  for (
+    let match = text.indexOf(needle);
+    match !== -1;
+    match = text.indexOf(needle, match + needle.length)
+  )
+    for (let index = match; index < match + needle.length; index++) hits[owner[index]!] = true;
+  return hits;
 }