refactor(files): share the whitespace class and split the display-name tests

Author: waleedlatif1

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

@@ -28,6 +28,7 @@ vi.mock('@/lib/uploads/core/setup.server', () => ({
 
 import { fileParseBodySchema } from '@/lib/api/contracts/storage-transfer'
 import * as inputValidation from '@/lib/core/security/input-validation.server'
+import { textArrayLiteral } from '@/lib/db/arrays'
 import { executeFileParserOperation } from '@/lib/internal/file/parser'
 import {
   createKnowledgeAclFixtureIds,
@@ -204,29 +205,34 @@ describe('workspace file names in PostgreSQL', () => {
     }
   })
 
+  // Names stored before upload normalization: U+202F before AM/PM, doubled and edge whitespace,
+  // a decomposed accent, a control character, and a BOM.
+  const unnormalizedNames = [
+    'Screenshot 2026-01-15 at 9.41.07\u202fAM.png',
+    'Quarterly  Report.pdf',
+    ' padded.txt\t',
+    'Cafe\u0301 menu.png',
+    'ring\u0007ing.png',
+    '\ufeffbom.md',
+  ]
+
+  it('normalizes a display name in SQL exactly as normalizeDisplaySegment does', async () => {
+    const rows = await db.execute<{ name: string; key: string }>(
+      sql`SELECT name, ${displaySegmentKey(sql`name`)} AS key FROM unnest(${textArrayLiteral(unnormalizedNames)}) AS name`
+    )
+    expect(rows).toHaveLength(unnormalizedNames.length)
+    for (const row of rows) expect(row.key).toBe(normalizeDisplaySegment(row.name))
+  })
+
   it('resolves a stored name that differs from its VFS path through the display-name index', async () => {
     const fixture = await seedWorkspace()
     const folder = await createWorkspaceFileFolder({
       workspaceId: fixture.workspaceId,
       userId: fixture.aliceId,
       name: 'Reports',
     })
-    // Names stored before upload normalization: U+202F before AM/PM, doubled and edge
-    // whitespace, a decomposed accent, a control character, and a BOM.
-    const storedNames = [
-      'Screenshot 2026-01-15 at 9.41.07 AM.png',
-      'Quarterly  Report.pdf',
-      ' padded.txt\t',
-      'Café menu.png',
-      'ring\u0007ing.png',
-      'bom.md',
-    ]
-    for (const name of storedNames) {
-      const [row] = await db.execute(sql`SELECT ${displaySegmentKey(sql`${name}::text`)} AS key`)
-      expect(row.key).toBe(normalizeDisplaySegment(name))
-    }
 
-    const files = storedNames.map((name, n) => ({
+    const files = unnormalizedNames.map((name, n) => ({
       name,
       id: `wf_legacy_${n}_${fixture.workspaceId}`,
       folderId: n === 0 ? folder.id : null,

apps/sim/lib/vfs/path.ts

@@ -1,10 +1,7 @@
-import { escapeRegExp } from '@sim/utils/string'
+import { escapeRegExp, WHITESPACE_CHARACTER_CLASS } from '@sim/utils/string'
 
 const CONTROL_CHARS = /[\x00-\x1f\x7f]/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(`${WHITESPACE_CLASS}+`, 'g')
+const WHITESPACE = new RegExp(`${WHITESPACE_CHARACTER_CLASS}+`, 'g')
 
 export class VfsPathError extends Error {
   constructor(message: string) {
@@ -29,7 +26,7 @@ export function normalizeDisplaySegment(segment: string): string {
  */
 export function displaySegmentPattern(name: string): string {
   const words = name.split(' ').map(escapeRegExp)
-  return `^${WHITESPACE_CLASS}*${words.join(`${WHITESPACE_CLASS}+`)}${WHITESPACE_CLASS}*$`
+  return `^${WHITESPACE_CHARACTER_CLASS}*${words.join(`${WHITESPACE_CHARACTER_CLASS}+`)}${WHITESPACE_CHARACTER_CLASS}*$`
 }
 
 export function encodeVfsSegment(segment: string): string {

packages/db/migrations/0407_workspace_files_display_name_index.sql

@@ -1,7 +1,5 @@
--- A name reference that misses the exact-name lookup read every live file in its workspace, and
--- a workflow's new-file existence check always misses, so a large workspace paid a full read per
--- file written. This index serves that fallback by displayed name. It is partial on live
--- workspace files, a small slice of the table, so the build writes little WAL.
+-- Serves name references that miss the exact-name lookup (see displaySegmentKey in schema.ts).
+-- Partial on live workspace files, so the concurrent build is small.
 COMMIT;--> statement-breakpoint
 SET lock_timeout = 0;--> statement-breakpoint
 -- migration-safe: replay replaces only this new index to recover an interrupted concurrent build; existing indexes remain available.

packages/db/schema.ts

@@ -1,3 +1,4 @@
+import { WHITESPACE_CHARACTER_CLASS } from '@sim/utils/string'
 import { type SQL, sql } from 'drizzle-orm'
 import {
   type AnyPgColumn,
@@ -69,10 +70,6 @@ export function foldedEmail(column: AnyPgColumn | SQL): SQL<string> {
   return sql<string>`lower(btrim(${column}))`
 }
 
-/** The characters JavaScript's `trim()` and `\s` treat as whitespace, as a PostgreSQL class. */
-const DISPLAY_WHITESPACE =
-  '[ \\t\\n\\v\\f\\r\\u00a0\\u1680\\u2000-\\u200a\\u2028\\u2029\\u202f\\u205f\\u3000\\ufeff]'
-
 /**
  * A file name reduced to the VFS path segment it displays as, in SQL: NFC-composed, trimmed,
  * control characters removed, whitespace runs collapsed to one space. It is the exact
@@ -82,7 +79,7 @@ const DISPLAY_WHITESPACE =
  * are inlined with `sql.raw` because a bound parameter would not match the index expression.
  */
 export function displaySegmentKey(column: AnyPgColumn | SQL): SQL<string> {
-  const ws = sql.raw(DISPLAY_WHITESPACE)
+  const ws = sql.raw(WHITESPACE_CHARACTER_CLASS)
   return sql<string>`regexp_replace(regexp_replace(regexp_replace(normalize(${column}, NFC), '^${ws}+|${ws}+$', '', 'g'), '[\\x01-\\x1f\\x7f]', '', 'g'), '${ws}+', ' ', 'g')`
 }
 
@@ -2822,12 +2819,7 @@ export const workspaceFiles = pgTable(
       .where(
         sql`${table.deletedAt} IS NULL AND ${table.context} = 'workspace' AND ${table.workspaceId} IS NOT NULL`
       ),
-    /**
-     * Serves name references that miss the exact-name lookup (a stored name whose whitespace,
-     * composition, or control characters differ from its VFS path, or a bare name in a folder).
-     * Without it each miss reads every live file in the workspace — one per new file a workflow
-     * writes, since the existence check misses by definition.
-     */
+    /** Serves name references that miss the exact-name lookup; see `displaySegmentKey`. */
     workspaceDisplayNameIdx: index('workspace_files_workspace_display_name_idx')
       .on(table.workspaceId, displaySegmentKey(table.originalName))
       .concurrently()

packages/utils/src/string.ts

@@ -14,6 +14,14 @@
  */
 const NUL_CHARACTER = '\u0000'
 
+/**
+ * The characters JavaScript's `\s` and `trim()` treat as whitespace, spelled out as a bracket
+ * expression both JavaScript and PostgreSQL regular expressions accept, so a SQL expression can
+ * normalize whitespace exactly as its TypeScript twin does.
+ */
+export const WHITESPACE_CHARACTER_CLASS =
+  '[ \\t\\n\\v\\f\\r\\u00a0\\u1680\\u2000-\\u200a\\u2028\\u2029\\u202f\\u205f\\u3000\\ufeff]'
+
 /** Reports whether `value` carries a `U+0000`. See {@link NUL_CHARACTER}. */
 export function containsNulCharacter(value: string): boolean {
   return value.includes(NUL_CHARACTER)