Repository navigation
Conversation
|
Thanks for this — verified all three fixes locally against
One thing worth flagging before merge: this branch is stacked on #10778, which is still open. Nice narrow, TDD-proven drain of the #9985 shards. |
e53bf4b to
fdea797
Compare
|
Hi @diegosouzapw, thanks for the detailed review! Quick update: we just pushed a new commit ( Could you please take a look at the latest changes when you have a moment? |
30b5493 to
8f414e2
Compare
0d72426 to
bd87b9a
Compare
bd87b9a to
5799ab6
Compare
…egram, and Discord formatters (cheaper-inference#9985)
|
Closing with credit — this drain PR's substance has been absorbed by the release tip:
Your single-value PRs (#11078/#11079/#11117/#11120/#11122/#11123/#11125/#11132) are the right shape and are being landed individually. Thank you @rqzbeh! |
…s before cached fallback Two release/v3.8.50 base-red unit failures in tests/unit/cc-compatible-provider.test.ts (part of the cheaper-inference#9985 "release branch not green" set), re-derived and fixed against the current tip rather than cherry-picked, since the base-red set has drifted since the originating PRs were written. 1. models route (genuine code bug; ref cheaper-inference#10795 fix cheaper-inference#1, closes the gap in cheaper-inference#10863): GET /api/providers/:id/models returned HTTP 200 with an empty/cached model list for a Claude-Code-compatible (anthropic-compatible-cc-*) connection instead of the documented HTTP 400 "does not support models listing". The isClaudeCodeCompatibleProvider check sat AFTER maybeReturnCachedDiscovery() / maybeReturnAutoFetchDisabled() inside the isAnthropicCompatibleProvider block, so a freshly-created CC connection with no synced models hit the auto-fetch-disabled 200 fallback first. Hoisted the CC 400 check to the top of that block so the unsupported-listing contract always wins. 2. stale test assertion (test-only; the code is correct per cheaper-inference#10684): the "preserves client cache markers" test asserted a client message marker sent with no ttl defaults to 1h even when a system block carries an explicit 5m marker. cheaper-inference#10684 ("advance Claude cache breakpoints on growing tails") changed normalizeCacheControlTtl to "default missing ttl to 5m after a 5m breakpoint" (sections processed tools -> system -> messages) and updated claude-code-parity.test.ts / chatcore-translation-paths.test.ts to match, but missed this third assertion. Updated it to expect 5m, matching the intended breakpoint semantics. Verified: tests/unit/cc-compatible-provider.test.ts 27/27, claude-code-parity 40/40, chatcore-translation-paths 75/75, typecheck:core clean. Refs cheaper-inference#9985, cheaper-inference#10795, cheaper-inference#10863, cheaper-inference#10684. (cherry picked from commit 674606963abc8ed3e9ba89f2544e9f3aad6c8fa5)
Summary
Stacked on PR #10778 to drain the remaining 5 unit test base-red failures on
release/v3.8.50(#9985):src/app/api/providers/[id]/models/route.ts: MovedisClaudeCodeCompatibleProvidercheck before cached discovery lookups so CC-compatible connections return HTTP 400 instead of returning HTTP 200 with cached models.src/i18n/messages/pt-BR.json: Added 6 missing Portuguese translation keys to matchen.json(settings.compressionOutputStyle.i-have-adhd.label,description,reasoningTokenBuffer,reasoningTokenBufferDesc,zeroLatencyOptimizations,zeroLatencyOptimizationsDesc).tests/unit/hard-session-lease-bypass-inventory.test.ts: UpdatedEXPECTED.credentialandEXPECTED.executorinventory baselines for 7 new connection/credential call sites (speechCombo.ts,videoCombo.ts,videoModelResolution.ts,transcriptions/route.ts,classify/route.ts,segment/route.ts,geminiWeb.ts).Verification
node --import tsx/esm --test tests/unit/i18n-pt-br.test.ts(3/3 passed)node --import tsx/esm --test tests/unit/cc-compatible-provider.test.ts(27/27 passed)node --import tsx/esm --test tests/unit/hard-session-lease-bypass-inventory.test.ts(3/3 passed)