fix(pds): normalize JSON blob refs for correct dag-cbor encoding (#119)

* fix(pds): normalize JSON blob refs for correct dag-cbor encoding

Incoming API records contain blob references with nested $link objects
(e.g. { "$type": "blob", "ref": { "$link": "bafk..." } }). Without
normalization, @atproto/repo's lexToIpld walks these as plain objects,
encoding the ref as a CBOR map instead of a CID tag. This produces
incorrect block hashes that cause blob resolution failures downstream.

Add normalizeRecordLinks() to convert $link → CID and $bytes →
Uint8Array before records hit repo.applyWrites(). Applied on all
write paths: createRecord, putRecord, applyWrites.

Also handle Uint8Array → $bytes serialization on the read path.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor: use @atproto/lexicon jsonToLex instead of hand-rolled normalization

Replace the custom normalizeRecordLinks implementation with a re-export
of jsonToLex from @atproto/lexicon. This is better because:
- Battle-tested against edge cases the custom impl may miss
- Also converts blob refs to proper BlobRef instances (not just CIDs)
- One line instead of 75

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor: use jsonToLex directly instead of format.ts alias

Import jsonToLex from @atproto/lexicon directly in account-do.ts instead of
re-exporting it as normalizeRecordLinks from format.ts. Removes the
unnecessary alias and the test file that was just testing library behavior.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* refactor: use @atproto/lex-json directly, remove @atproto/lexicon dep

* fix: regenerate lockfile without spurious dep resolution changes

The previous lockfile bumped @atproto/common-web from 0.4.7 to 0.4.8
inside @atproto/lexicon, which likely caused CI test failures in
vitest-pool-workers. Reset from main and cleanly added only
@atproto/lex-json.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* test(pds): add blob ref normalization integration tests

Verifies that JSON-serialized blob refs ({"$link": "..."}) are properly
normalized to CID objects when stored in the repo via createRecord,
putRecord, and applyWrites. Uses raw repo access to confirm internal
representation matches expectations.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* Lock freshen

* Use vitest beta

---------

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

Author: ascorbic

.changeset/normalize-blob-refs.md

@@ -0,0 +1,16 @@
+---
+"@getcirrus/pds": patch
+---
+
+Normalize JSON blob references for correct dag-cbor encoding
+
+Incoming API records contain blob references with nested `$link` objects
+(for example, `{ "$type": "blob", "ref": { "$link": "bafk..." } }`). These
+must be converted to actual CID instances before CBOR encoding, otherwise
+the blob ref's `ref` field gets encoded as a map instead of a proper CID tag.
+This causes incorrect block hashes, which can lead to blob resolution failures
+on the Bluesky network.
+
+Uses `jsonToLex` from `@atproto/lex-json` to convert `$link` → CID and
+`$bytes` → Uint8Array on all record write paths (createRecord, putRecord,
+applyWrites).

package.json

@@ -20,7 +20,13 @@
 		"knip": "^5.78.0",
 		"prettier": "^3.7.4",
 		"typescript": "^5.9.3",
-		"vitest": "^3.2.4"
+		"vitest": "4.1.0-beta.1"
 	},
-	"packageManager": "pnpm@10.26.2"
+	"packageManager": "pnpm@10.26.2",
+	"pnpm": {
+		"overrides": {
+			"wrangler": "^4.63.0",
+			"miniflare": "^4.20260205.0"
+		}
+	}
 }

packages/oauth-provider/package.json

@@ -30,7 +30,7 @@
 		"publint": "^0.3.16",
 		"tsdown": "^0.18.3",
 		"typescript": "^5.9.3",
-		"vitest": "^4.0.0"
+		"vitest": "4.1.0-beta.1"
 	},
 	"repository": {
 		"type": "git",

packages/pds/package.json

@@ -34,6 +34,7 @@
 		"@atproto/crypto": "^0.4.5",
 		"@atproto/lex-cbor": "^0.0.3",
 		"@atproto/lex-data": "^0.0.3",
+		"@atproto/lex-json": "^0.0.11",
 		"@atproto/repo": "^0.8.12",
 		"@clack/prompts": "^0.11.0",
 		"@getcirrus/oauth-provider": "workspace:*",
@@ -59,7 +60,7 @@
 		"tsx": "^4.21.0",
 		"typescript": "^5.9.3",
 		"vite": "^6.4.1",
-		"vitest": "^4.0.0",
+		"vitest": "4.1.0-beta.1",
 		"wrangler": "^4.54.0",
 		"ws": "^8.18.3"
 	},

packages/pds/src/account-do.ts

@@ -27,6 +27,7 @@ import {
 	type CommitData,
 } from "./sequencer";
 import { BlobStore, type BlobRef } from "./blobs";
+import { jsonToLex } from "@atproto/lex-json";
 import type { PDSEnv } from "./types";
 
 /**
@@ -330,7 +331,7 @@ export class AccountDurableObject extends DurableObject<PDSEnv> {
 			action: WriteOpAction.Create,
 			collection,
 			rkey: actualRkey,
-			record: record as RepoRecord,
+			record: jsonToLex(record) as RepoRecord,
 		};
 
 		const prevRev = repo.commit.rev;
@@ -471,18 +472,19 @@ export class AccountDurableObject extends DurableObject<PDSEnv> {
 		const existing = await repo.getRecord(collection, rkey);
 		const isUpdate = existing !== null;
 
+		const normalizedRecord = jsonToLex(record) as RepoRecord;
 		const op: RecordWriteOp = isUpdate
 			? ({
 					action: WriteOpAction.Update,
 					collection,
 					rkey,
-					record: record as RepoRecord,
+					record: normalizedRecord,
 				} as RecordUpdateOp)
 			: ({
 					action: WriteOpAction.Create,
 					collection,
 					rkey,
-					record: record as RepoRecord,
+					record: normalizedRecord,
 				} as RecordCreateOp);
 
 		const prevRev = repo.commit.rev;
@@ -581,7 +583,7 @@ export class AccountDurableObject extends DurableObject<PDSEnv> {
 					action: WriteOpAction.Create,
 					collection: write.collection,
 					rkey,
-					record: write.value as RepoRecord,
+					record: jsonToLex(write.value) as RepoRecord,
 				};
 				ops.push(op);
 				results.push({
@@ -598,7 +600,7 @@ export class AccountDurableObject extends DurableObject<PDSEnv> {
 					action: WriteOpAction.Update,
 					collection: write.collection,
 					rkey: write.rkey,
-					record: write.value as RepoRecord,
+					record: jsonToLex(write.value) as RepoRecord,
 				};
 				ops.push(op);
 				results.push({
@@ -1528,6 +1530,15 @@ function serializeRecord(obj: unknown): unknown {
 		return { $link: cid.toString() };
 	}
 
+	// Convert Uint8Array to { $bytes: "<base64>" }
+	if (obj instanceof Uint8Array) {
+		let binary = "";
+		for (let i = 0; i < obj.length; i++) {
+			binary += String.fromCharCode(obj[i]!);
+		}
+		return { $bytes: btoa(binary) };
+	}
+
 	if (Array.isArray(obj)) {
 		return obj.map(serializeRecord);
 	}