rpc_blob 2.3.0

Streamed blob storage contract and RPC wiring for rpc_dart (binary/object storage alongside rpc_data).

2.3.0

Changed

• Requires rpc_dart 6. See its changelog — notably that a handler's bare Exception text no longer reaches the caller, so a storage error now needs an RpcStatusException to say anything to the peer.

2.2.0

Changed

• Requires rpc_dart 5. See its changelog: flow control is on by default, an expired deadline is now RpcDeadlineExceededException on every shape, and a stream that ends without a trailer raises UNAVAILABLE.

2.1.0

• applyBulkDelete returns results positionally aligned with the request, the same as applyBulkHead. It used to emit version-checked ids first and the batched ones after, grouped by collection — one result per request, but not in request order, so a caller pairing the two by index read another id's verdict. Grouping is how the work is issued, not how it is reported.

2.0.0

Prepares the blob layer for a hosted S3, where a round trip is billed and a bucket is a scarce resource. The interface changes are breaking.

Breaking:

• IBlobRepository.collectionSize returns Future<int?>. Null means the backend cannot answer without walking the whole store; 0 still means empty. The cost of this call ranges from an indexed query to a full enumeration depending on the backend, and a plain int hid that well enough for it to end up on a request path. Callers that need the number regardless sum listBlobs themselves, where the cost is visible.
• CollectionSizeResponse.sizeBytes is nullable to match.
• Three methods join IBlobRepository, so any implementation outside this repository must add them: ensureCollection, deleteMany, headMany.

Batching:

• deleteMany and headMany let a backend answer a batch the way it can — one statement on SQLite, DeleteObjects on S3, bounded parallel HEADs where there is no batch metadata call.
• bulkDeleteBlob and bulkHeadBlob were loops over single-blob calls in both the RPC responder and the in-process repository client; both now share applyBulkDelete / applyBulkHead. Fixing only the responder would have left the in-process path — the one a server holding a repository directly uses — still walking ids one round trip at a time.
• Version-checked deletes stay one at a time: that check is per id, while a batch delete is unconditional.

Lifecycle:

• ensureCollection makes setting a collection up an explicit act, so stores no longer have to re-discover its absence on every write.

1.2.2

Web/dart2js correctness (verified on node):

• InMemoryBlobRepository blob-id generation falls back from Random.secure() to Random() when secure randomness is unavailable in the runtime (e.g. node/dart2js without a proper globalThis crypto binding). Blob ids are not secrets, so this is a safe cross-platform compromise that no longer throws at startup on web.
• Added test/web_smoke_test.dart (runs on vm || node) exercising the full RPC client round-trip (put/get/list) and a mid-download get server-stream cancel; confirmed the client surface has no dart:io leak and the streaming get cancel does not deadlock on dart2js.

1.2.1

Bug fix:

• BlobUploadChunk.toJson() and BlobDownloadFrame.toJson() now coerce bytes via Uint8List.fromList() to guarantee correct CBOR byte string encoding on dart2js.
• _bytesFromJsonValue handles List<dynamic> (from CBOR array decoding) as a fallback for cross-platform compatibility.
• Bumped rpc_dart dependency to ^3.1.0.

1.0.5

• Allow to pass raw sqlite db connection to adapter

1.0.4

• S3 adapter now auto-detects public buckets and returns plain download URLs when anonymous s3:GetObject is allowed; presigning is used only for private buckets, still honoring presign* host overrides.

1.0.3

• Added presign-only endpoint overrides (presignEndpoint/presignPort/presignUseSSL/presignPathStyle) for S3/MinIO to sign links on a public host behind a reverse proxy; removed downloadUrlMapper to avoid generating invalid signatures.

1.0.2

• S3 adapter can rewrite presigned download URLs via S3BlobStorageOptions.downloadUrlMapper (useful for serving through a proxy/CDN) and configure presign TTL via S3BlobStorageOptions.presignTtlSeconds; constructor params consolidated into S3BlobStorageOptions.

1.0.1

• S3 adapter now fetches object tags (?tagging) and merges them into BlobDescriptor.metadata without overwriting metadata provided on upload.
• Added xml dependency to parse S3 tag responses.

1.0.0

• Added S3BlobStorageAdapter (S3/MinIO/Ceph-compatible) storing blobs under <prefix><collection>/<id> with metadata-based versioning and optimistic checks.
• New S3BlobStorageAdapter.connect(...) helper for quick setup; list/Head/read/write/delete/listCollections wired to S3 operations; descriptors include a presigned download URL.
• README now documents S3/MinIO usage; pubspec updated with the MinIO client dependency.
• Initial version.