mirror of
https://github.com/diegosouzapw/OmniRoute.git
synced 2026-07-26 09:52:11 +03:00
* chore(release): open v3.8.22 development cycle * refactor(dashboard): extract ProviderDetailPageClient — #3501 Phase 0 (#3633) #3501 Phase 0: extract ProviderDetailPageClient + smoke test. Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * refactor(dashboard): extract auth-import modals — #3501 Phase 1a (#3634) #3501 Phase 1a: extract 3 auth-import modal clusters. Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * fix(db): reclassify localDb unexported modules as intentionally-internal (#3499) (#3635) Closes #3499 — reclassify localDb unexported modules as intentionally-internal (audit + honest gate framing). * refactor(db): move call_logs aggregations into callLogStats db module (#3500) (#3636) #3500 slice 1: call_logs aggregations → src/lib/db/callLogStats.ts (Rule #5). Byte-identical queries; TDD 6/6. * refactor(dashboard): extract EditCompatibleNodeModal — #3501 Phase 1b (#3638) #3501 Phase 1b: extract EditCompatibleNodeModal (cycle-safe via leaf constants module). Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * refactor(db): move community_servers SQL into gamification db module (#3500 slice 3) (#3639) #3500 slice 3: community_servers SQL → gamification db module. * refactor(db): move usage_history SQL into usageAnalytics module (#3500 slice 2) (#3644) #3500 slice 2: usage_history/daily_usage_summary SQL → usageAnalytics db module. * refactor(db): move skills UPDATE + db-backups SQL into db modules (#3500 slice 5) (#3647) #3500 slice 5: skills UPDATE (allowlist) + db-backups SQL → db modules. * refactor(db): move usage_logs/semantic_cache/proxy_logs SQL into db modules (#3500 slice 4) (#3648) #3500 slice 4: usage_logs/semantic_cache/proxy_logs SQL → db modules. All internal routes done (2 external by-design remain). * chore(db-gate): reclassify external-DB reads, fully close #3500 (#3649) Closes #3500: reclassify external-DB reads; all internal raw-SQL migrated to db/ modules. * refactor(dashboard): extract pure helpers to providerPageHelpers — #3501 Phase 2 (#3653) #3501 Phase 2: extract pure helpers to providerPageHelpers (leaf, cycle-safe). Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * refactor(dashboard): extract remaining shared helpers to providerPageHelpers — #3501 Phase 2b (#3658) #3501 Phase 2b: extract remaining shared helpers to providerPageHelpers (leaf, cycle-safe). Heavy modals unblocked. Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * fix(reasoning): replay reasoning_content on plain DeepSeek turns (#1682) (#3632) Integrated into release/v3.8.22 * fix(kiro): route enterprise IAM Identity Center accounts to their regional endpoint (#3631) Integrated into release/v3.8.22 * refactor: small code cleanup (#3523) Integrated into release/v3.8.22 * fix(combo): skip same-provider targets on 408/500/502/503/504/524 errors (#3637) Integrated into release/v3.8.22 — circuit-breaker guard added in review (#1731v2) * feat(providers): add MiMoCode free-tier provider with bootstrap JWT auth (#3659) Integrated into release/v3.8.22 — page.tsx conflict resolved + NoAuthAccountCard re-applied to ProviderDetailPageClient in review. MiMoCode endpoint validated live. * Log Responses WebSocket calls in history (#3616) Integrated into release/v3.8.22 — Codex Responses WebSocket call history logging. * Add Claude Code routing preference for unprefixed Claude models (#3540) Integrated into release/v3.8.22 — page.tsx conflict resolved (re-applied toggle to ProviderDetailPageClient) + disable-test updated for catalog drift in review. * docs(changelog): credit #3632/#3631/#3637/#3659/#3540/#3616/#3523 (v3.8.22 targeted review round) * fix(mimocode): add required authHeader:"none" to registry entry (#3659 follow-up) The mimocode RegistryEntry omitted the required authHeader field, which broke typecheck:core (TS2741). Match the no-auth convention (authType:"none" + authHeader:"none") used by veoaifree-web and other free providers. Follow-up to #3659 (@pizzav-xyz). * fix(responses): detect stream readiness for tool-call-only and object-less chunks (#3612) (#3661) Closes #3612 * fix(mitm): remove duplicated 'Command failed:' error prefix (#3641) (#3662) Closes #3641 * fix(cli): honor HERMES_HOME for Hermes Agent config path (#3628) (#3663) Closes #3628 * fix(api): fetch live OpenCode model catalog for no-auth model picker (#3611) (#3664) Closes #3611 * fix(api): flag provider topology error state by current status, not stale history (#3619) (#3666) Closes #3619 * fix(electron): launch peer-stamping server-ws.mjs entrypoint to avoid 403 LOCAL_ONLY (#3386) (#3665) Closes #3386 * fix(dashboard): restore home topology live in-flight pulse (#3507) (#3667) Closes #3507 * fix(oauth): name Kiro/AWS auto-imported accounts and dedupe by profileArn (#3615) (#3671) Closes #3615 * fix(resilience): clear stale transient connection cooldowns on startup (#3625) (#3672) Closes #3625 * fix(i18n): use logical CSS direction utilities for sidebar and key overlays (RTL #3541) (#3670) Closes #3541 * fix(dashboard): honor auto-hide and switch to visible filter on passthrough Test-all (#3610) (#3669) Closes #3610 * refactor(dashboard): extract AddApiKeyModal + EditConnectionModal — #3501 Phase 1c (#3674) #3501 Phase 1c: extract AddApiKeyModal, EditConnectionModal, WebSessionCredentialGuide into components/; god-component 10,166->8,092 LOC. Reconciles the v3.8.22 file-size drift for this file. Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * docs(changelog): reconcile v3.8.22 — credit #3621/#3622 + MiMoCode follow-up roll-up * refactor(dashboard): extract ConnectionRow + ModelCompatPopover + SiliconFlowEndpointModal — #3501 Phase 1d (#3676) #3501 Phase 1d: god-component 8,092->6,838 LOC. Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * feat(obsidian): add WebDAV config route + encrypt creds at rest (#3485 part 1) (#3677) Part 1 of #3485. Adds /api/settings/obsidian/webdav (GET/POST/DELETE) wiring the ready obsidianSync lib, encrypts webdav password + obsidian token at rest, removes the duplicate UI block, drops the KNOWN_MISSING entry. WebDAV file server is part 2. * feat(obsidian): add /api/v1/webdav file server for Obsidian vault sync (#3485 part 2) (#3678) Part 2 of #3485. WebDAV server (PROPFIND/GET/PUT/DELETE/MKCOL/MOVE/OPTIONS) handled in the custom server layer (standalone-server-ws.mjs) since the App Router cannot export WebDAV methods. Basic-Auth (constant-time), path-traversal hardened, password decrypt ported from encryption.ts (parity-tested), DATA_DIR resolution parity-tested against dataPaths.ts. End-to-end Obsidian-over-Tailscale validation is a live VPS step (Rule #18). * fix(combo): stop premature context compaction — real auto-combo windows + per-target compression limit (#3680) Integrated into release/v3.8.22 * feat(dashboard): deactivate/activate accounts from the quota overview (#3675) Integrated into release/v3.8.22 * fix(dashboard): close review gaps in bulk provider connection actions (#3271 follow-up) (#3673) Integrated into release/v3.8.22 — page.tsx conflict (god-component split #3501) resolved by re-applying the bulk-action deltas to ProviderDetailPageClient.tsx * refactor(dashboard): extract useModelCompatState hook + model sections — #3501 Phase 1e (#3683) #3501 Phase 1e: extract useModelCompatState hook (unblocks the model sections) + ModelRow/PassthroughModelsSection/PassthroughModelRow/CustomModelsSection/CompatibleModelsSection. god-component 6,838->4,921 LOC. Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * refactor(dashboard): extract useProviderConnections/Settings/Models hooks — #3501 Phase 1f (#3684) #3501 Phase 1f: god-component 4,948->4,062 LOC. Connection state+handlers, settings, and model metadata moved into hooks/. Co-authored-by: oyi77 <oyi77@users.noreply.github.com> * chore(release): v3.8.22 CHANGELOG + env-doc sync - Set release date in CHANGELOG [3.8.22] to 2026-06-11 - Add HERMES_HOME to .env.example (from #3628/#3663) - Add HERMES_HOME + OMNIROUTE_PREFER_CLAUDE_CODE_FOR_UNPREFIXED_CLAUDE_MODELS to ENVIRONMENT.md (#3628/#3540) * docs(changelog): credit #3673 + #3675 — leninejunior bulk-actions + quota-toggle --------- Co-authored-by: oyi77 <oyi77@users.noreply.github.com> Co-authored-by: Abhishek Divekar <adivekar@utexas.edu> Co-authored-by: NOXX - Commiter <artur1992123@mail.ru> Co-authored-by: Nicolas Lorin <androw95220@gmail.com> Co-authored-by: Hernan Javier Ardila Sanchez <hjasgr@gmail.com> Co-authored-by: PizzaV <103120356+pizzav-xyz@users.noreply.github.com> Co-authored-by: kkkayye <98376609+kkkayye@users.noreply.github.com> Co-authored-by: Witroch4 <witalo_rocha@hotmail.com> Co-authored-by: Lenine Júnior <lenine@engrene.com.br>
211 lines
8.4 KiB
TypeScript
211 lines
8.4 KiB
TypeScript
/**
|
|
* TDD regression tests — auto-combo context-window advertising + per-target
|
|
* combo compression limit (the "premature auto compaction" bug).
|
|
*
|
|
* Bug chain (discussion report: coding agents "keep forgetting things"):
|
|
* 1. /api/combos/auto never exposed context_length, so the opencode plugin
|
|
* advertised `limit: { context: 0 }` for auto combos. opencode disables
|
|
* its smart auto-compaction entirely when context === 0, letting the
|
|
* conversation grow until OmniRoute's destructive purifyHistory() drops
|
|
* old messages silently.
|
|
* 2. chatCore's proactive-compression block overrode the per-target context
|
|
* limit with min(...allComboTargets) even though chatCore always executes
|
|
* with the CONCRETE target's provider/model (handleSingleModel resolves
|
|
* the target before calling chatCore) — compressing at the smallest
|
|
* target's window while running on the largest target.
|
|
*
|
|
* Fixes under test:
|
|
* - virtualFactory.computeAdvertisedLimits(): MAX of candidates' known
|
|
* context windows (the auto-combo context pre-filter routes oversized
|
|
* requests to large-window candidates, so MAX is safe to advertise).
|
|
* - GET /api/combos/auto includes context_length / max_output_tokens.
|
|
* - contextManager.resolveComboContextLimit(): prefers the executing
|
|
* target's own limit; min(...targets) only as a defensive fallback when
|
|
* the current provider/model resolves no specific limit.
|
|
*/
|
|
import test from "node:test";
|
|
import assert from "node:assert/strict";
|
|
import fs from "node:fs";
|
|
import os from "node:os";
|
|
import path from "node:path";
|
|
|
|
const TEST_DATA_DIR = fs.mkdtempSync(path.join(os.tmpdir(), "omniroute-auto-combo-ctx-"));
|
|
process.env.DATA_DIR = TEST_DATA_DIR;
|
|
process.env.API_KEY_SECRET = process.env.API_KEY_SECRET ?? "auto-combo-ctx-test-secret";
|
|
|
|
const core = await import("../../src/lib/db/core.ts");
|
|
const settingsDb = await import("../../src/lib/db/settings.ts");
|
|
|
|
const virtualFactory = await import("../../open-sse/services/autoCombo/virtualFactory.ts");
|
|
const contextManager = await import("../../open-sse/services/contextManager.ts");
|
|
const combosAutoRoute = await import("../../src/app/api/combos/auto/route.ts");
|
|
|
|
test.after(() => {
|
|
core.resetDbInstance();
|
|
try {
|
|
fs.rmSync(TEST_DATA_DIR, { recursive: true, force: true });
|
|
} catch {
|
|
// best-effort cleanup
|
|
}
|
|
});
|
|
|
|
// ── virtualFactory.computeAdvertisedLimits ───────────────────────────────────
|
|
|
|
test("computeAdvertisedLimits returns MAX of candidates' known context windows", () => {
|
|
const { computeAdvertisedLimits } = virtualFactory as unknown as {
|
|
computeAdvertisedLimits: (candidates: Array<{ provider: string; model: string }>) => {
|
|
contextLength: number | null;
|
|
maxOutputTokens: number | null;
|
|
};
|
|
};
|
|
assert.equal(
|
|
typeof computeAdvertisedLimits,
|
|
"function",
|
|
"virtualFactory should export computeAdvertisedLimits()"
|
|
);
|
|
|
|
// gemini has registry defaultContextLength=1048576; claude has 200000.
|
|
const result = computeAdvertisedLimits([
|
|
{ provider: "claude", model: "claude-sonnet-4-6" },
|
|
{ provider: "gemini", model: "gemini-2.5-pro" },
|
|
]);
|
|
assert.equal(result.contextLength, 1048576, "MAX of candidate windows should win");
|
|
assert.ok(
|
|
typeof result.maxOutputTokens === "number" && result.maxOutputTokens > 0,
|
|
"maxOutputTokens should be a positive number"
|
|
);
|
|
});
|
|
|
|
test("computeAdvertisedLimits returns null limits for an empty candidate pool", () => {
|
|
const { computeAdvertisedLimits } = virtualFactory as unknown as {
|
|
computeAdvertisedLimits: (candidates: Array<{ provider: string; model: string }>) => {
|
|
contextLength: number | null;
|
|
maxOutputTokens: number | null;
|
|
};
|
|
};
|
|
const result = computeAdvertisedLimits([]);
|
|
assert.equal(result.contextLength, null);
|
|
assert.equal(result.maxOutputTokens, null);
|
|
});
|
|
|
|
test("computeAdvertisedLimits never returns 0 for a non-empty pool (unknown models fall back)", () => {
|
|
const { computeAdvertisedLimits } = virtualFactory as unknown as {
|
|
computeAdvertisedLimits: (candidates: Array<{ provider: string; model: string }>) => {
|
|
contextLength: number | null;
|
|
maxOutputTokens: number | null;
|
|
};
|
|
};
|
|
const result = computeAdvertisedLimits([
|
|
{ provider: "totally-unknown-provider", model: "mystery-model" },
|
|
]);
|
|
assert.ok(
|
|
typeof result.contextLength === "number" && result.contextLength > 0,
|
|
`unknown candidates should fall back to a positive default, got ${result.contextLength}`
|
|
);
|
|
});
|
|
|
|
// ── GET /api/combos/auto advertises context_length ──────────────────────────
|
|
|
|
test("GET /api/combos/auto includes positive context_length for combos with candidates", async () => {
|
|
await settingsDb.updateSettings({ requireLogin: false });
|
|
|
|
const req = new Request("http://localhost/api/combos/auto", { method: "GET" });
|
|
const res = await combosAutoRoute.GET(req as never);
|
|
const body = await res.json();
|
|
|
|
assert.equal(res.status, 200);
|
|
assert.ok(Array.isArray(body.combos), "body.combos should be an array");
|
|
assert.ok(body.combos.length > 0, "should list at least the default auto combo");
|
|
|
|
for (const combo of body.combos) {
|
|
if ((combo.candidateCount ?? 0) > 0) {
|
|
assert.ok(
|
|
typeof combo.context_length === "number" && combo.context_length > 0,
|
|
`combo ${combo.id} with ${combo.candidateCount} candidates must advertise a positive context_length, got ${combo.context_length}`
|
|
);
|
|
assert.ok(
|
|
typeof combo.max_output_tokens === "number" && combo.max_output_tokens > 0,
|
|
`combo ${combo.id} must advertise a positive max_output_tokens, got ${combo.max_output_tokens}`
|
|
);
|
|
}
|
|
}
|
|
});
|
|
|
|
// ── contextManager.resolveComboContextLimit (per-target compression limit) ──
|
|
|
|
test("resolveComboContextLimit prefers the executing target's own limit over combo min", () => {
|
|
const { resolveComboContextLimit } = contextManager as unknown as {
|
|
resolveComboContextLimit: (opts: {
|
|
provider: string;
|
|
model: string | null;
|
|
comboTargetLimits: number[];
|
|
}) => { limit: number; source: string };
|
|
};
|
|
assert.equal(
|
|
typeof resolveComboContextLimit,
|
|
"function",
|
|
"contextManager should export resolveComboContextLimit()"
|
|
);
|
|
|
|
// Executing on gemini (1048576 provider default) while the combo also has
|
|
// a tiny 32k target: compression must use the EXECUTING target's window.
|
|
const result = resolveComboContextLimit({
|
|
provider: "gemini",
|
|
model: "gemini-2.5-pro",
|
|
comboTargetLimits: [32000, 1048576],
|
|
});
|
|
assert.equal(result.limit, 1048576, "must not regress to min(...targets) on a known target");
|
|
assert.equal(result.source, "target");
|
|
});
|
|
|
|
test("resolveComboContextLimit regression: claude target must not be compressed at an 8k sibling", () => {
|
|
const { resolveComboContextLimit } = contextManager as unknown as {
|
|
resolveComboContextLimit: (opts: {
|
|
provider: string;
|
|
model: string | null;
|
|
comboTargetLimits: number[];
|
|
}) => { limit: number; source: string };
|
|
};
|
|
const result = resolveComboContextLimit({
|
|
provider: "claude",
|
|
model: "claude-sonnet-4-6",
|
|
comboTargetLimits: [8000],
|
|
});
|
|
assert.equal(result.limit, 200000);
|
|
assert.equal(result.source, "target");
|
|
});
|
|
|
|
test("resolveComboContextLimit falls back to combo min when the target has no specific limit", () => {
|
|
const { resolveComboContextLimit } = contextManager as unknown as {
|
|
resolveComboContextLimit: (opts: {
|
|
provider: string;
|
|
model: string | null;
|
|
comboTargetLimits: number[];
|
|
}) => { limit: number; source: string };
|
|
};
|
|
const result = resolveComboContextLimit({
|
|
provider: "totally-unknown-provider",
|
|
model: "mystery-model",
|
|
comboTargetLimits: [32000, 200000],
|
|
});
|
|
assert.equal(result.limit, 32000, "unknown target should defensively use min of combo targets");
|
|
assert.equal(result.source, "combo-min");
|
|
});
|
|
|
|
test("resolveComboContextLimit uses generic fallback when nothing else is known", () => {
|
|
const { resolveComboContextLimit } = contextManager as unknown as {
|
|
resolveComboContextLimit: (opts: {
|
|
provider: string;
|
|
model: string | null;
|
|
comboTargetLimits: number[];
|
|
}) => { limit: number; source: string };
|
|
};
|
|
const result = resolveComboContextLimit({
|
|
provider: "totally-unknown-provider",
|
|
model: "mystery-model",
|
|
comboTargetLimits: [],
|
|
});
|
|
assert.equal(result.limit, 128000, "generic default fallback");
|
|
assert.equal(result.source, "fallback");
|
|
});
|