Releases: Effect-TS/effect
@effect/ai-openai@0.39.2
Patch Changes
49c5acdThanks @mollyegibson! - Previously, settingstrict: falseonOpenAiLanguageModelconfig caused a 400 "Unknown parameter: 'strict'" response from the OpenAI Responses API, because the flag was spread into the top-level request body instead of being consumed only by the tool and response_format schema builders. Thestrictflag is now stripped from the request body while still controllingstricton tool schemas (prepareTools) and json_schema response formats (prepareResponseFormat).
effect@3.21.1
Patch Changes
-
#6139
f99048eThanks @marbemac! - Fix batched request resolver defects causing consumer fibers to hang forever.When a
RequestResolver.makeBatchedresolver died with a defect, the requestDeferreds were never completed because the cleanup logic ininvokeWithInterruptusedflatMap(which only runs on success). Changed toensuringso uncompleted request entries are always resolved regardless of exit type.
@effect/cluster@0.58.1
Patch Changes
-
#6183
4708bb8Thanks @tim-smart! - backport cluster serialization fix for notify path -
Updated dependencies [
f99048e]:- effect@3.21.1
@effect/cli@0.75.1
Patch Changes
-
#6144
ec5c505Thanks @LikiosSedo! - Fix--log-level=valueequals syntax incorrectly swallowing the next argument. Only skip the next arg when the previous arg is exactly--log-level(space-separated form). -
Updated dependencies [
f99048e]:- effect@3.21.1
@effect/ai-openrouter@0.10.1
Patch Changes
-
#6145
6c39a34Thanks @LikiosSedo! - Fix typo in HTTP header name:HTTP-Referrer→HTTP-Referer. The HTTP spec spells it "Referer" (single r), and OpenRouter expects this exact header name for app attribution. -
Updated dependencies [
f99048e]:- effect@3.21.1
@effect/ai-openai@0.39.1
Patch Changes
-
47f0439Thanks @aayushbaluni! - fix(ai-openai): deduplicate response.output items to prevent invalid JSON concatenation -
#6187
b63fdb8Thanks @alex-dixon! - Change 'in-memory' to 'in_memory' in prompt cache enums -
#6174
739f077Thanks @mollyegibson! - Make 'strict' mode configurable for tool definitions passed to the OpenAI model -
Updated dependencies [
f99048e]:- effect@3.21.1
@effect/sql-pg@0.52.1
Patch Changes
- #6140
4767f86Thanks @tim-smart! - Use a dedicated PostgreSQL connection for LISTEN / UNLISTEN so active listeners do not consume a pooled query connection.
effect@3.21.0
Minor Changes
-
#5780
f7bb09bThanks @kitlangton! - AddCron.prevand reverse iteration support, aligning next/prev lookup tables, fixing DST handling symmetry, and expanding cron backward/forward test coverage. -
#5780
bd7552aThanks @mattiamanzati! - Add type-level utils to asserting layer types -
#5780
ad1a7ebThanks @schickling! - RcMap: support dynamicidleTimeToLivevalues per keyThe
idleTimeToLiveoption can now be a function that receives the key and returns a duration, allowing different TTL values for different resources.const map = yield * RcMap.make({ lookup: (key: string) => acquireResource(key), idleTimeToLive: (key: string) => { if (key.startsWith("premium:")) return Duration.minutes(10) return Duration.minutes(1) } })
-
#5780
0d32048Thanks @mikearnaldi! - Fix annotateCurrentSpan, add Effect.currentPropagatedSpan
Patch Changes
-
#5780
0d32048Thanks @mikearnaldi! - Add logs to first propagated span, in the following case before this fix the log would not be added to thepspan becauseEffect.fnadds a fake span for the purpose of adding a stack frame.import { Effect } from "effect" const f = Effect.fn(function* () { yield* Effect.logWarning("FooBar") return yield* Effect.fail("Oops") }) const p = f().pipe(Effect.withSpan("p"))