fix(db): drop displayName backfill, accept legacy NULLs

Make display_name nullable so the migration is a clean two-statement
ADD COLUMN + CREATE UNIQUE INDEX with no backfill. Legacy rows keep
display_name = NULL; readers coalesce to original_name. Pre-existing
collisions persist (the upload already happened — no fix possible
without rewriting history) but new uploads get full disambiguation
via the partial unique index, which treats NULLs as distinct so legacy
rows don't block index creation or new inserts.

Replaces the prior 0202_hard_grey_gargoyle migration (window-function
backfill). Resolver now matches WHERE display_name = ? OR (display_name
IS NULL AND original_name = ?), and the normalize-segment fallback
coalesces too.

Author: TheodoreSpeaks

apps/sim/lib/copilot/tools/handlers/materialize-file.ts

@@ -21,7 +21,7 @@ function toFileRecord(row: typeof workspaceFiles.$inferSelect) {
   return {
     id: row.id,
     workspaceId: row.workspaceId || '',
-    name: row.displayName,
+    name: row.displayName ?? row.originalName,
     key: row.key,
     path: `${pathPrefix}${encodeURIComponent(row.key)}?context=mothership`,
     size: row.size,
@@ -45,7 +45,7 @@ async function executeSave(fileName: string, chatId: string): Promise<ToolCallRe
 
   const [updated] = await db
     .update(workspaceFiles)
-    .set({ context: 'workspace', chatId: null, originalName: row.displayName })
+    .set({ context: 'workspace', chatId: null, originalName: row.displayName ?? row.originalName })
     .where(and(eq(workspaceFiles.id, row.id), isNull(workspaceFiles.deletedAt)))
     .returning({ id: workspaceFiles.id, originalName: workspaceFiles.originalName })
 

apps/sim/lib/copilot/tools/handlers/upload-file-reader.ts

@@ -2,20 +2,29 @@ import { db } from '@sim/db'
 import { workspaceFiles } from '@sim/db/schema'
 import { createLogger } from '@sim/logger'
 import { toError } from '@sim/utils/errors'
-import { and, asc, eq, isNull } from 'drizzle-orm'
+import { and, asc, eq, isNull, or } from 'drizzle-orm'
 import { type FileReadResult, readFileRecord } from '@/lib/copilot/vfs/file-reader'
 import { normalizeVfsSegment } from '@/lib/copilot/vfs/normalize-segment'
 import { getServePathPrefix } from '@/lib/uploads'
 import type { WorkspaceFileRecord } from '@/lib/uploads/contexts/workspace/workspace-file-manager'
 
 const logger = createLogger('UploadFileReader')
 
+/**
+ * VFS-visible name. Pre-displayName-column rows have NULL displayName, so we coalesce
+ * to originalName for them (those rows may still collide with each other — acceptable
+ * since the upload already happened; only new rows get collision protection).
+ */
+function vfsName(row: typeof workspaceFiles.$inferSelect): string {
+  return row.displayName ?? row.originalName
+}
+
 function toWorkspaceFileRecord(row: typeof workspaceFiles.$inferSelect): WorkspaceFileRecord {
   const pathPrefix = getServePathPrefix()
   return {
     id: row.id,
     workspaceId: row.workspaceId || '',
-    name: row.displayName,
+    name: vfsName(row),
     key: row.key,
     path: `${pathPrefix}${encodeURIComponent(row.key)}?context=mothership`,
     size: row.size,
@@ -29,10 +38,10 @@ function toWorkspaceFileRecord(row: typeof workspaceFiles.$inferSelect): Workspa
 }
 
 /**
- * Resolve a mothership upload row by `displayName` (the collision-disambiguated name
- * exposed via `uploads/<displayName>` in the VFS). Prefers an exact DB match; falls back
- * to a normalized scan when the model passes a visually-equivalent name (e.g. macOS U+202F
- * narrow no-break space vs ASCII space in screenshot filenames).
+ * Resolve a mothership upload row by VFS name (the collision-disambiguated `displayName`
+ * for new rows, or `originalName` for legacy rows that predate the column). Prefers an
+ * exact DB match; falls back to a normalized scan when the model passes a visually
+ * equivalent name (e.g. macOS U+202F vs ASCII space in screenshot filenames).
  */
 export async function findMothershipUploadRowByChatAndName(
   chatId: string,
@@ -45,7 +54,10 @@ export async function findMothershipUploadRowByChatAndName(
       and(
         eq(workspaceFiles.chatId, chatId),
         eq(workspaceFiles.context, 'mothership'),
-        eq(workspaceFiles.displayName, fileName),
+        or(
+          eq(workspaceFiles.displayName, fileName),
+          and(isNull(workspaceFiles.displayName), eq(workspaceFiles.originalName, fileName))
+        ),
         isNull(workspaceFiles.deletedAt)
       )
     )
@@ -67,7 +79,7 @@ export async function findMothershipUploadRowByChatAndName(
     )
 
   const segmentKey = normalizeVfsSegment(fileName)
-  return allRows.find((r) => normalizeVfsSegment(r.displayName) === segmentKey) ?? null
+  return allRows.find((r) => normalizeVfsSegment(vfsName(r)) === segmentKey) ?? null
 }
 
 /**

packages/db/migrations/0202_hard_grey_gargoyle.sql

@@ -0,0 +1,2 @@
+ALTER TABLE "workspace_files" ADD COLUMN "display_name" text;--> statement-breakpoint
+CREATE UNIQUE INDEX "workspace_files_chat_display_name_active_unique" ON "workspace_files" USING btree ("chat_id","display_name") WHERE "workspace_files"."deleted_at" IS NULL AND "workspace_files"."context" = 'mothership' AND "workspace_files"."chat_id" IS NOT NULL;

packages/db/migrations/0202_superb_abomination.sql

@@ -1,5 +1,5 @@
 {
-  "id": "549270ba-08fa-4690-af38-84494e559936",
+  "id": "0440b68d-47b7-4104-af1b-15f3206e9d7c",
   "prevId": "f7dc3eab-9b08-4ed7-99b0-d952797d86eb",
   "version": "7",
   "dialect": "postgresql",
@@ -14513,7 +14513,7 @@
           "name": "display_name",
           "type": "text",
           "primaryKey": false,
-          "notNull": true
+          "notNull": false
         },
         "content_type": {
           "name": "content_type",

packages/db/migrations/meta/0202_snapshot.json

@@ -1412,8 +1412,8 @@
     {
       "idx": 202,
       "version": "7",
-      "when": 1778009467652,
-      "tag": "0202_hard_grey_gargoyle",
+      "when": 1778014005108,
+      "tag": "0202_superb_abomination",
       "breakpoints": true
     }
   ]

packages/db/migrations/meta/_journal.json

@@ -1149,10 +1149,13 @@ export const workspaceFiles = pgTable(
     /**
      * Collision-disambiguated name exposed to the copilot VFS as `uploads/<displayName>`.
      * For mothership chat uploads, identical originalNames within a chat get suffixed
-     * `(2)`, `(3)`, ... in upload order. For other contexts, equals originalName.
-     * Stable for the row's lifetime; the partial unique index below makes that enforceable.
+     * `(2)`, `(3)`, ... in upload order so the VFS path is unique per chat.
+     * NULL on legacy rows that predate this column — readers must coalesce to originalName.
+     * Stable for the row's lifetime; the partial unique index below enforces uniqueness
+     * for new (non-NULL) rows. NULLs are treated as distinct in PG unique indexes, so
+     * legacy collisions remain (acceptable: those uploads have already happened).
      */
-    displayName: text('display_name').notNull(),
+    displayName: text('display_name'),
     contentType: text('content_type').notNull(),
     size: integer('size').notNull(),
     deletedAt: timestamp('deleted_at'),