Effect-TS/effect effect@4.0.7

on GitHub

Patch Changes

• #6743 ac3485a Thanks @IMax153! - Add a scoped CliOutput.Presenter service for suppressing, redirecting, or replacing built-in CLI help, invalid-invocation, and version output.

• #9873 3a4c691 Thanks @effect-bot! - Require tool parameter schema services in Response.AllParts, Response.Part and Response.StreamPart, including for toolkits with handlers, and add Tool.ParametersDecodingServices.

• #9861 4f0600e Thanks @effect-bot! - Add a toModelOutput option to Tool.make and Tool.dynamic for sending text, images and files to the model in place of the serialized tool result. Anthropic and OpenAI Responses send native content blocks; OpenAI-compatible APIs and OpenRouter send text with a placeholder for each file, via the new Prompt.toolModelOutputToText.

  OpenAI Responses now sends PDF file parts given as base64, data URL or HTTP(S) URL strings instead of dropping them.

• #9377 110fd22 Thanks @gcanti! - Add a concurrency option to Arbitrary.checkEffect and global check configuration to run independent property evaluations concurrently. Results remain deterministic for deterministic generators and repeatable properties, while shrinking and replay stay sequential. The default concurrency is 1.

• #9811 aa58cda Thanks @effect-bot! - Speed up Base64.encode and Base64Url.encode on large inputs and reduce the memory they allocate.

• #9913 8e257e7 Thanks @Marve10s! - Fix duplicate interrupt reasons when interrupting a fiber inside Effect.withSpan or Effect.fn.

• #9846 e62a149 Thanks @mkantautas! - Fix ChildProcessSpawner helpers hanging when a command fills a pipe the helper never reads. exitCode now reads and discards stdout and stderr until the process exits, and string, lines, streamString and streamLines read and discard stderr unless includeStderr is set. A failure reading that output while the helper runs, such as from an output Sink, now fails the helper.

• #9794 dcd43a5 Thanks @effect-bot! - Clarify that ChildProcess output sinks for stdout, stderr and output additionalFds only receive data while the corresponding handle stream is consumed, and that the handle stream emits the sink's results.

• #9921 3388cdd Thanks @Marve10s! - Fix Bash completion of file and directory paths containing spaces, quotes or glob characters.

• #9976 81b0e14 Thanks @effect-bot! - Add Completions.fromCommand to effect/cli. It converts an existing Command tree into a CommandDescriptor, so build tooling can generate shell completion scripts with Completions.generate without running the CLI.

• #9881 a046435 Thanks @effect-bot! - Skip Flag.withFallbackPrompt and Argument.withFallbackPrompt fallbacks when TerminalCapabilities.canPrompt is false, failing with the original MissingOption or MissingArgument error instead of rendering a prompt or waiting for input.

• #9922 f6d8968 Thanks @Marve10s! - Show command help for invalid global flag values instead of failing with a raw parse error.

• #8349 5741c3a Thanks @ArjunCodess! - Add the CliError.userMessage annotation so Command.run renders tagged handler errors without wrapping them in CliError.UserError.

• #9959 381d5c6 Thanks @baptisteArno! - Record ClickHouse migrations only after they succeed, so a failed migration and the ones after it are retried on the next run.

• #10001 d47478a Thanks @tim-smart! - Stop a runner that lost a shard lock from storing the final reply of a persisted request on that shard. Before storing it, the runner refreshes its shard locks and checks that the shard is still among them. If it is not, the request is left for the shard's new owner.

• #10037 383716d Thanks @tim-smart! - Release a frozen runner's PostgreSQL advisory shard locks after shardLockExpiration. On PostgreSQL 14+, SqlRunnerStorage sets idle_session_timeout on its lock connection, so a runner that stops refreshing loses its locks instead of holding them until the server drops its TCP connection.

  With advisory locks on PostgreSQL and MySQL, SqlRunnerStorage refresh now only confirms locks its current lock session holds. After the lock connection is replaced, shards held by the previous session are reported lost instead of being silently reacquired.

• #9817 46dc30a Thanks @effect-bot! - Fix cluster shutdown waiting for entityTerminationTimeout when a workflow is waiting on a persisted activity request that has not started. During runner shutdown the caller is now abandoned so the request replays under the next owner, and a durable caller abandoned this way is left for replay instead of being answered with an interrupt. With preemptiveShutdown: false, entities drain before the runner is marked as shutting down, so such a caller still waits for entityTerminationTimeout.

• #9985 3d5c862 Thanks @tim-smart! - Fix a cluster workflow replay ignoring a durable deferred completion that is persisted but not yet processed by the new owner. A DurableDeferred.raceAll could pick a different winner on replay and run the wrong branch. Workflow replays now read the stored completion from the unprocessed request when it has no reply yet.

• #9845 93ebc4f Thanks @effect-bot! - Fix ClusterWorkflowEngine running a completed step's compensation twice when interruptUnsafe interrupts a workflow during a running activity.

• #9938 c7b161a Thanks @effect-bot! - Fix a forced shard release stalling forever when a child workflow wakes its still-running parent. The wake's cluster abandonment was retried, so it waited on the retired run and blocked the shard from being reacquired. The abandonment now interrupts the wake, so the activation unwinds, the release completes, and the parent is resumed under the next owner.

• #10034 e55945d Thanks @jpowersdev! - Fix HttpClientRequest.compressBody replacing the request's Content-Type with the body's content type. Compression now only updates the Content-Encoding and Content-Length headers.

• #10035 1a53d6a Thanks @jpowersdev! - Fix HttpMiddleware.compression sending an uncompressed response when Accept-Encoding contains an empty list member, such as gzip,. Empty members are now ignored as RFC 9110 requires.

• #9769 5e25cc9 Thanks @Marve10s! - Fix Equal.equals and Hash.hash treating any two same-length ArrayBuffers as equal. ArrayBuffer and SharedArrayBuffer are now compared and hashed by their bytes.

• #9799 9503d5c Thanks @effect-bot! - Fix Exit.hasFails, Exit.hasDies, Exit.hasInterrupts, and AsyncResult.isInterrupted narrowing away every failure when they return false. They now return boolean, like the matching Cause checks; use Exit.isFailure or AsyncResult.isFailure to narrow.

• #10026 13efb72 Thanks @tim-smart! - Prevent later Semaphore.take calls from overtaking eligible queued requests while preserving mixed-size overtaking.

• #9767 8ca6abb Thanks @Marve10s! - Fix Queue.offerAll, MutableList.appendAll and MutableList.prependAll keeping a reference to the caller's array. Changing that array afterward no longer changes the queued values.

• #9982 4a09dd1 Thanks @jpowersdev! - Fix JsonSchema.toDocumentDraft07 emitting enum: [], which the Draft-07 metaschema rejects. An empty enum now becomes an allOf entry that rejects every value.

• #9754 346fc92 Thanks @gcanti! - Fix Schema.isCapitalized and Schema.isUncapitalized accepting astral letters with the wrong case.

• #9967 20e5ba8 Thanks @gcanti! - Fix Schema.isULID to reject values exceeding the 128-bit ULID limit.

• #9878 f4b237e Thanks @Hona! - Fix Effect.fn stack frames retaining call arguments. A fiber forked inside an Effect.fn call inherits the frame, which kept the arguments of that call and of every traced call above it alive for as long as the fiber ran.

• #9887 1dd08ee Thanks @jayhemnani9910! - Fix HttpClient.followRedirects re-encoding the query of the Location it follows (%20 became +, flag became flag=, and invalid UTF-8 bytes were replaced with U+FFFD). The redirected request now uses the resolved Location URL without rewriting its query.

• #9948 940be42 Thanks @chiefcll! - Format values with own getters in Formatter.formatJson and Cause.pretty without Proxy, so engines that lack it no longer throw a ReferenceError. Getter results stay redacted on every engine.

• #10029 37be245 Thanks @jpowersdev! - Fix HttpStaticServer dropping Content-Location, Date, Expires and Vary from 304 responses.

• #9944 563fde0 Thanks @effect-bot! - Add HttpApiClient.decodeResponse to decode a raw endpoint response with the endpoint's declared success and error schemas, for example after a responseMode: "response-only" call.

• #9796 938cecb Thanks @effect-bot! - Reject . and .. path parameters in HttpApiClient so they cannot change the request path through URL normalization.

• #9869 cf197c1 Thanks @sarankumar1325! - Describe undeclared response statuses in HttpApiClient errors. The reason is still a DecodeError, now with the description undeclared status code, so the message reads Decode: undeclared status code (418 GET ...) instead of Decode error (418 GET ...).

• #9998 7ca71bb Thanks @tim-smart! - Make HttpApiClient close the upgraded connection of an HttpApiSchema.Upgrade endpoint when the 101 response headers fail to decode, instead of leaving it open.

• #10002 1e73219 Thanks @tim-smart! - Make HttpApiClient close the upgraded connection of an HttpApiSchema.Upgrade endpoint when it is interrupted right after taking the upgrade, before its cleanup was in place.

• #9941 4dd860d Thanks @effect-bot! - Allow HttpApiSchema.StreamUint8Array and HttpApiSchema.StreamNdjson to declare a list of content types. The first entry is the default response type, streaming payloads are accepted under every entry, and OpenAPI emits the stream under each one.

• #9855 4d24a11 Thanks @effect-bot! - Stop HttpClientError reasons from exposing request headers, query parameters and bodies when serialized or inspected. Their toJSON now prints the request as its method and redacted URL, and the response as its status, matching the error message. The full request and response remain available through the request and response properties.

• #9786 6bbc5ac Thanks @gcanti! - Fix OpenAPI incorrectly marking optional HTTP parameters and headers as nullable.

• #9770 5fd25f4 Thanks @Marve10s! - Fix HttpRouter.prefixed dropping the leading / from the local URL when only a query string follows the prefix. A handler for /api?x=1 now sees /?x=1 instead of ?x=1, and a router mounted under a prefix no longer returns 404 for /prefix?query.

• #10011 a461561 Thanks @tim-smart! - Format HttpServerError messages with only the request path, so query values and URL credentials can't leak into span errors or logs. The full request is still available on the error.

• #9994 832b3c6 Thanks @tim-smart! - Redact the URL in the failure that HttpMiddleware.tracer records for 5xx responses when the response ends, so exporters that read the message later no longer see query values that were only redacted during the request.

• #9874 837e906 Thanks @john-griffin! - Fix HttpApiClient method calls inside Effect.all widening their required services to any, and reading services off a non-streaming method's type giving unknown.

• #9942 cd1f2d4 Thanks @effect-bot! - Name both endpoints when HttpApiBuilder.layer finds two HttpApiEndpoints with the same method and path, and document how to declare request variants of one operation as a single endpoint.

• #9751 fb38ce6 Thanks @effect-bot! - Allow buffered payloads and typed 101 response headers on HttpApiSchema.Upgrade endpoints.

  • HttpApiSchema.Upgrade({ protocol, headers }) declares 101 headers, encoded like WithHeaders headers. Clients decode them and return { headers, socket }, and OpenAPI lists them on the 101.
  • Upgrade endpoints can declare buffered payloads. Streaming payloads are still rejected.
  • Handlers receive an upgrade function instead of socket: upgrade(), or upgrade({ headers }) when headers are declared.
  • 426 Upgrade Required is returned before the payload is decoded.
  • HttpServerRequest gains upgradeWith({ headers }), and upgrade is upgradeWith({}). Third-party HttpServerRequest implementations must add it. WebSocket upgrades fail when headers are given.
  • NodeHttpServer waits for the request body to end before writing the 101. On Node 25 and older, which do not parse upgrade request bodies, it reads Content-Length bodies from the connection and answers chunked ones with 411 Length Required.
  • NodeMultipart fails with a MultipartError when its source stream errors, instead of waiting for more input.

• #9815 446c7ac Thanks @tim-smart! - Ignore end, attribute, event, and addLinks calls on spans that have already ended. Logs and Effect.annotateCurrentSpan from fibers that outlive their span no longer modify the ended span, OtlpTracer no longer exports a span twice when end is called again, OtelSpan no longer triggers OpenTelemetry SDK warnings for late calls, and the DevTools tracer no longer sends events for ended spans.

• #10024 b48b7d8 Thanks @jpowersdev! - Ignore malformed Set-Cookie Max-Age attributes instead of parsing their numeric prefixes.

• #8180 dc66d90 Thanks @Ishkirat-Singh! - SchemaRepresentation.fromJsonSchemaDocument now supports $ref pointers below top-level definitions, including shared and recursive schemas, and rejects circular pointer aliases during import.

• #9999 d5a3dc3 Thanks @jpowersdev! - JsonPatch.apply now throws on malformed JSON Pointer escapes, such as /a~2 or a trailing ~, instead of treating them as literal keys.

• #10027 482e0d5 Thanks @tkr109! - Fix Match array patterns matching arrays shorter than the pattern.

• #8509 d84f518 Thanks @IMax153! - Add a controlMessages option to the JSON-RPC serializations in RpcSerialization. MCP servers disable it, so @effect/rpc/* notifications from MCP clients no longer stop stdio servers or write @effect/rpc/Pong to stdout.

• #9961 d34d399 Thanks @effect-bot! - Fix McpServer method-not-found, invalid-params and invalid-request errors including an internal _tag field. Strict MCP clients such as McpClient rejected these responses and the stdio transport stopped serving later requests.

• #9886 5bbd25e Thanks @jayhemnani9910! - Fix MultipartParser rejecting horizontal tabs in quoted Content-Disposition and Content-Type parameters.

• #9768 14eab6b Thanks @Marve10s! - MutableRef and Ref now compare by reference. Distinct instances with equal contents no longer compare Equal or share a HashSet or HashMap key.

• #9925 af8fcb3 Thanks @Marve10s! - Fix OpenApi.fromApi to require all stacked security middleware instead of documenting them as alternatives.

• #9829 6b82a28 Thanks @tim-smart! - Add the effect/observability/OtlpProxy module, which serves OTLP routes that forward browser telemetry to a collector or vendor and keep credentials on the server.

• #9914 3043aea Thanks @Marve10s! - Fix PartitionedSemaphore deadlocking when waiters in different partitions request multiple permits, by completing one waiter before rotating to the next partition.

• #9646 785577e Thanks @tim-smart! - Add the experimental effect/persistence/CircuitBreaker module with keyed circuits, consecutive and rate trip policies, an in-memory store and a Redis store, plus HttpClient.withCircuitBreaker and HttpClient.circuitBreakerClassifier.

• #9800 a8dbbce Thanks @effect-bot! - Fix Pool waiters starving under contention. Capacity freed by a release, an ended reservation, a new acquisition or a strategy reclaim now goes to the longest-waiting borrower instead of a borrower that arrives later.

• #9762 65094ac Thanks @godu! - Fix Stream.broadcastN dropping the last element for a subscriber that keeps up while another lags beyond capacity. PubSub.end no longer drops a backpressured message that a waiting subscriber has not received yet.

• #9903 113ec70 Thanks @jayhemnani9910! - Redact the AWS SigV4 query parameters X-Amz-Signature, X-Amz-Credential and X-Amz-Security-Token by default.

• #10032 ad3126e Thanks @jpowersdev! - Fix HttpClient.followRedirects keeping Content-Encoding, Content-Language and Content-Location headers on redirects that switch the request to GET.

• #10036 eedfb27 Thanks @effect-bot! - Release discarded redirect response bodies in HttpClient.followRedirects.

• #9828 45c467a Thanks @tim-smart! - Add Redis.reserve, which leases a dedicated connection from a pool, so blocking commands like XREAD BLOCK and BLPOP no longer hold up other commands on the shared connection. NodeRedis, BunRedis and DenoRedis layers accept maxConnections, minConnections, idleTimeout and acquireTimeout to size the pool.

  Redis.make now requires a connect option, and hand-written Redis services need a reserve member.

• #9872 c983fc1 Thanks @effect-bot! - Reject repeated occurrences of single-valued CLI flags instead of silently keeping the first value. Use Flag.atLeast, Flag.atMost or Flag.between to accept multiple values.

• #9900 58f164f Thanks @effect-bot! - Fix RpcClient scope closure hanging when a streamed chunk fails to decode while the socket disconnects and never reconnects. The best-effort Interrupt sent to the server for the failed request no longer blocks client shutdown, and is still sent if the client stays open.

• #9899 f6df1aa Thanks @effect-bot! - Treat synchronous throws from RPC handlers as request-level defects, so disableFatalDefects: true fails only the throwing request instead of every pending request on the client.

• #9931 c05355c Thanks @Marve10s! - Fix RpcSerialization.jsonRpc and ndJsonRpc dropping or reordering stream chunks when a stream request is sent in a JSON-RPC batch.

• #9898 ae7805c Thanks @effect-bot! - Add a disableFatalDefects option to RpcTest.makeClient, so a handler defect fails only its own request instead of every pending request on the client.

• #9917 9ffc621 Thanks @Marve10s! - Fix Schedule.fibonacci skipping the repeated initial delay. For a base of 100 millis the delays are now 100, 100, 200, 300, 500 ms instead of 100, 200, 300, 500, 800 ms, so every retry delay after the first is shorter than before.

• #9771 c97b331 Thanks @Marve10s! - Fix SchemaRepresentation.toCodeDocument generating code that fails to load when a schema identifier is Schema, RegExp, Date or Symbol.

• #9613 15297e5 Thanks @gcanti! - Restore encoded content schemas for Schema.fromJsonString in schema representations and JSON Schema, including SSE event data in OpenAPI.

  Schema.fromJsonString now ignores explicit contentMediaType and contentSchema annotations on its encoded String. Plain Schema.String schemas still honor those annotations.

  Content declarations and checks now participate in code generation and persistence; add any missing toCode callbacks or representation identities.

  The encoded projection of Schema.fromJsonString no longer carries a contentMediaType annotation.

• #9970 41f6738 Thanks @danieljvdm! - Keep caller stack frames in errors created with Schema.Error.make and Schema.TaggedError.make, including when only a decoder is registered.

• #9620 53f9352 Thanks @Marve10s! - Support Schema.Enum keys in Schema.Record, declaring each enum member value as a required property instead of throwing during construction.

• #9864 b72f6a2 Thanks @gcanti! - Schema.Record now consistently rejects optional index-signature values created with Schema.optionalKey or Schema.optional. Use Schema.UndefinedOr to allow undefined values.

• #9775 52e1716 Thanks @gcanti! - Preserve shared references in Schema.toCodecJson and Schema.toCodecStringTree, and retain httpApiStatus annotations when adding checks.

  If you relied on checks clearing HTTP statuses, add .annotate({ httpApiStatus: undefined }) explicitly.

• #9850 50c863a Thanks @gcanti! - Fix Schema.toJsonSchemaDocument throwing for named template-literal segments and emitting incorrect schemas for segment checks, exclusive unions, and empty unions.

• #9929 c148d27 Thanks @Marve10s! - Fix Snowflake.makeGenerator returning duplicate ids when the clock is frozen or moves backward.

• #9758 a11ecdd Thanks @bezff! - Fix IN ${sql.in([])} compiling to the invalid IN (). An empty list now matches no rows with IN and every row with NOT IN, when the keyword directly precedes the helper. On PostgreSQL, array-typed columns are not supported, and a comparison on the left of IN must be parenthesized: a = (b IN ${sql.in([])}).

• #10014 08e338e Thanks @effect-bot! - Split SqlEventJournal batch statements so large remote syncs and writeMany calls stay within database parameter limits, including the 100 parameters Durable Object SQLite allows per statement, still committing each batch in one transaction after a single callback.

• #9951 8450ad5 Thanks @tim-smart! - Keep SqlResolver batches separated by transactions started with SqlClient.makeWithTransaction when the caller's context has no SqlClient, or carries a different client. Requests outside such a transaction no longer join its batch, where a rollback could discard their writes.

• #9825 39da87a Thanks @MaxFreedomPollard! - Fix transformResultNames and transformQueryNames throwing on JSON arrays that contain null, and turning scalar items in those arrays into objects.

• #9759 a2adb31 Thanks @bezff! - Escape the sql.updateValues alias as an identifier, without applying name transforms. Pass a bare alias rather than pre-quoted SQL, and quote mixed-case PostgreSQL aliases where they are referenced. Custom onRecordUpdate compiler hooks now receive the alias already escaped.

• #9989 a6174d5 Thanks @jpowersdev! - Fix Sse.makeParser dropping the first event of a stream that starts with a UTF-8 BOM when an empty chunk is fed first.

• #9772 96e6004 Thanks @MichaelDeBoey! - Add the root $schema to the draft-07 and draft-2020-12 schemas returned by Schema.toStandardJSONSchemaV1.

• #9915 3ebd645 Thanks @Marve10s! - Fix Stream.groupBy and Stream.groupByKey groups not ending when the upstream ends or fails.

• #9916 3c3c79b Thanks @Marve10s! - Fix Stream.grouped and Stream.rechunk dropping buffered elements when the upstream fails.

• #9803 8c6d2fa Thanks @effect-bot! - Keep upstream stream failures out of sink error handlers in Stream.run, Stream.peel and Stream.aggregate/Stream.aggregateWithin. Upstream failures now reach the caller unchanged instead of being remapped or recovered by Sink.mapError, Sink.catch, Sink.catchCause or Sink.orElse, and Sink.onExit observes them as an interruption.

• #9954 1ff35d2 Thanks @effect-bot! - Preserve the throw location of defects raised in Effect.sync and Effect.suspend when rendering causes on V8.

• #9834 2b100e9 Thanks @effect-bot! - Emit smaller declarations for exported Data.taggedEnum constructors by returning the named Data.TaggedEnum.Constructor and new Data.TaggedEnum.GenericConstructor types instead of expanded anonymous objects.

  Libraries that export a generic tagged enum and are built against this version emit Data.TaggedEnum.GenericConstructor in their declarations, so their consumers need effect 4.0.7 or later.

• #10023 e3938e5 Thanks @jpowersdev! - Fix Cookies.fromSetCookie dropping cookies with spaces or tabs before =, such as session = abc, and keeping that whitespace at the start of the value in session= abc. Spaces and tabs around the name and value are now removed, as RFC 6265 requires, while percent-encoded spaces in the value are still decoded.

• #10028 f90101a Thanks @jpowersdev! - Fix HttpStaticServer ignoring If-Match and If-Unmodified-Since. Failed preconditions now return 412 before cache validation and range handling.

• #10022 306c9a1 Thanks @jpowersdev! - Fix Toml.parse treating everything after a bare carriage return as comment text. A carriage return not followed by a line feed now throws a SyntaxError; LF and CRLF line endings are still accepted.

• #10021 5b69c67 Thanks @jpowersdev! - Fix Toml.parse to reject forbidden raw control characters, including bare carriage returns.

• #10019 a12ffe3 Thanks @jpowersdev! - Fix Toml.parse to reject integers outside JavaScript's safe integer range instead of silently rounding them.

• #9904 f6cf0fd Thanks @FredAmartey! - Fix Toml.parse accepting a table defined twice, such as a [table] header for a table that dotted keys already defined.

• #9924 3541f65 Thanks @Marve10s! - Fix Toml.parse to reject Unicode escapes that are not scalar values, as required by TOML 1.0.

• #10025 d50aaa2 Thanks @jpowersdev! - Fix Cookies.parseHeader leaving quotes on values surrounded by whitespace.

• #10013 3731ddb Thanks @tim-smart! - Fix SqlRunnerStorage leaking shard locks on Vitess after releasing all shards. RELEASE_ALL_LOCKS() makes vtgate replace the session's lock connection, so the stored lock holder id went stale: later acquisitions took locks the runner already held a second time, and releasing such a shard left it locked. The storage now re-takes its pin lock after releasing all shards and refreshes the stored id.

• #9835 3b8090c Thanks @tim-smart! - Enforce MaxBodySize on buffered body reads (arrayBuffer, text, json, urlParamsBody) of web HttpServerRequests, as used by Bun, Deno and toWebHandler. Bodies over the limit fail with a BodyTooLarge RequestParseError, which responds with 413.

• #9986 137757a Thanks @tim-smart! - Replay a workflow on the next runner when shutdown interrupts an activity in a DurableDeferred.raceAll whose other branches are parked. Before, the run was stored as Suspended and never resumed.

• #9848 a658477 Thanks @tim-smart! - Preserve the app's error and service types when applying HttpMiddleware.xForwardedHeaders, instead of widening them to any.