fix(files): match chat upload names exactly in SQL with one bounded row

Author: waleedlatif1

apps/sim/lib/uploads/contexts/workspace/__integration__/file-names.integration.ts

@@ -226,6 +226,7 @@ describe('workspace file names in PostgreSQL', () => {
       'Quarterly  Report.pdf',
       'Cafe\u0301 menu.png',
       'ring\u0007ing.png',
+      'trail.png \u0007',
     ]
     for (const name of names) {
       const row = await trackUpload(fixture.workspaceId, fixture.aliceId, chatId, name)
@@ -257,7 +258,12 @@ describe('workspace file names in PostgreSQL', () => {
     await trackUpload(fixture.workspaceId, fixture.aliceId, chatId, 'notes 100%.png')
     await trackUpload(fixture.workspaceId, fixture.aliceId, chatId, 'back\\slash.png')
 
-    for (const reference of ['uploads/a_b.png', 'uploads/notes.png', 'uploads/notes%20%25.png']) {
+    for (const reference of [
+      'uploads/a_b.png',
+      'uploads/a.b.png',
+      'uploads/notes.png',
+      'uploads/notes%20%25.png',
+    ]) {
       expect(
         await resolveWorkspaceFileReference(fixture.workspaceId, reference, {
           includeChatUploads: true,

apps/sim/lib/uploads/contexts/workspace/workspace-file-manager.ts

@@ -16,7 +16,6 @@ import {
 } from '@sim/utils/errors'
 import { generateShortId } from '@sim/utils/id'
 import { omit } from '@sim/utils/object'
-import { escapeLikePattern } from '@sim/utils/string'
 import { and, desc, eq, inArray, isNotNull, isNull, or, type SQL, sql } from 'drizzle-orm'
 import type { ShareRecord } from '@/lib/api/contracts/public-shares'
 import type { V2FileSortBy } from '@/lib/api/contracts/v2/files'
@@ -101,7 +100,7 @@ import {
 import { getWorkspaceFileSize, MAX_WORKSPACE_FILE_SIZE } from '@/lib/uploads/shared/types'
 import { isMarkdownFile } from '@/lib/uploads/utils/file-utils'
 import type { ServableFile } from '@/lib/uploads/utils/file-utils.server'
-import { normalizeDisplaySegment } from '@/lib/vfs/path'
+import { displaySegmentPattern } from '@/lib/vfs/path'
 import { SIM_PAGE_CONTENT_TYPE } from '@/lib/workspace-files/page-compile'
 import {
   MAX_SIM_PAGE_UPLOAD_SNIFF_BYTES,
@@ -1580,36 +1579,33 @@ export function parseChatUploadReference(fileReference: string): string | null {
  * Display names are unique per chat. Mothership supplies that namespace; callers
  * without a chat scope retain the workspace-wide newest-name lookup.
  *
- * `name` is in VFS display form ({@link normalizeDisplaySegment}), since that is what the
- * upload notice's path encodes, while the stored name keeps the uploaded spelling (a macOS
- * screenshot carries U+202F before AM/PM). The query narrows to names that, composed and
- * stripped of control characters, hold each of `name`'s words in order — a superset of the
- * exact match, which then compares the stored name in display form.
+ * `name` is decoded from the path the upload notice prints, which VFS encoding normalizes
+ * (NFC, control characters removed, whitespace runs collapsed and trimmed), while the stored
+ * name keeps the uploaded spelling: a macOS screenshot carries U+202F before AM/PM. The stored
+ * name is composed and stripped of control characters in SQL, and
+ * {@link displaySegmentPattern} matches its whitespace the way the encoding collapses it.
  */
 async function getChatUploadByName(
   workspaceId: string,
   name: string,
   chatId?: string
 ): Promise<WorkspaceFileRecord | null> {
-  const storedName = sql<string>`coalesce(${workspaceFiles.displayName}, ${workspaceFiles.originalName})`
-  const wordsInOrder = `%${name.split(' ').map(escapeLikePattern).join('%')}%`
-  const candidates = await db
+  const storedName = sql`coalesce(${workspaceFiles.displayName}, ${workspaceFiles.originalName})`
+  const [file] = await db
     .select()
     .from(workspaceFiles)
     .where(
       and(
         eq(workspaceFiles.workspaceId, workspaceId),
         eq(workspaceFiles.context, 'mothership'),
         chatId === undefined ? undefined : eq(workspaceFiles.chatId, chatId),
-        sql`regexp_replace(normalize(${storedName}, NFC), '[\\x01-\\x1f\\x7f]', '', 'g') LIKE ${wordsInOrder} ESCAPE '\\'`,
+        sql`regexp_replace(normalize(${storedName}, NFC), '[\\x01-\\x1f\\x7f]', '', 'g') ~ ${displaySegmentPattern(name)}`,
         isNull(workspaceFiles.deletedAt)
       )
     )
     .orderBy(desc(workspaceFiles.uploadedAt))
+    .limit(1)
 
-  const file = candidates.find(
-    (candidate) => normalizeDisplaySegment(candidate.displayName ?? candidate.originalName) === name
-  )
   return file ? mapChatUploadRecord(file, workspaceId) : null
 }
 

apps/sim/lib/vfs/path.ts

@@ -1,5 +1,10 @@
+import { escapeRegExp } from '@sim/utils/string'
+
 const CONTROL_CHARS = /[\x00-\x1f\x7f]/g
-const WHITESPACE = /\s+/g
+/** The characters `\s` matches, spelled out so a PostgreSQL pattern can share the class. */
+const WHITESPACE_CLASS =
+  '[ \\t\\n\\v\\f\\r\\u00a0\\u1680\\u2000-\\u200a\\u2028\\u2029\\u202f\\u205f\\u3000\\ufeff]'
+const WHITESPACE = new RegExp(`${WHITESPACE_CLASS}+`, 'g')
 
 export class VfsPathError extends Error {
   constructor(message: string) {
@@ -8,12 +13,18 @@ export class VfsPathError extends Error {
   }
 }
 
+function normalizeDisplaySegment(segment: string): string {
+  return segment.normalize('NFC').trim().replace(CONTROL_CHARS, '').replace(WHITESPACE, ' ')
+}
+
 /**
- * The display form every VFS path segment is built from and decoded to: NFC, trimmed,
- * control characters removed, and each whitespace run (including U+202F) one space.
+ * Anchored regular expression, valid in JavaScript and PostgreSQL, that matches every
+ * NFC-composed, control-character-free name whose segment decodes to `name`: the words of
+ * `name` (already in decoded form) separated by whitespace runs, with any whitespace around them.
  */
-export function normalizeDisplaySegment(segment: string): string {
-  return segment.normalize('NFC').trim().replace(CONTROL_CHARS, '').replace(WHITESPACE, ' ')
+export function displaySegmentPattern(name: string): string {
+  const words = name.split(' ').map(escapeRegExp)
+  return `^${WHITESPACE_CLASS}*${words.join(`${WHITESPACE_CLASS}+`)}${WHITESPACE_CLASS}*$`
 }
 
 export function encodeVfsSegment(segment: string): string {