Files
OmniRoute/tests/unit/auto-combo-context-advertising.test.ts
Diego Rodrigues de Sa e Souza 9350a5d6c6 Release v3.8.22 (#3623)
* 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>
2026-06-11 18:52:29 -03:00

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");
});