Compare commits

...

30 Commits

Author SHA1 Message Date
Jason 6cf49ee032 fix(proxy/gemini): prefer exact tool-call id over normalized-name fallback
The shadow-turn matcher used a three-branch `||` chain (id / full name /
normalized name). When two tools share a suffix (e.g. `server_a:search`
and `server_b:search`), the normalized-name clause could short-circuit
on an earlier turn whose id is actually wrong for the incoming tool_use,
mis-routing replay state (functionCall id / thoughtSignature) for later
tool_result resolution.

Split matching into two layers: when the incoming message carries any
tool_use ids, run id-based lookup first and return on the earliest hit.
Only fall back to full-name / normalized-name matching when the incoming
ids are absent or none of them resolve.

Add two regressions:

- shadow_replay_prefers_exact_id_match_over_normalized_name_collision
  Two shadow turns with colliding normalized names and two assistant
  messages whose ids cross the positional order; asserts each message
  replays the id-correct shadow turn (including thoughtSignature).

- shadow_replay_falls_back_to_name_when_ids_absent
  Shadow turn with no id and incoming tool_use with an empty id;
  asserts the name fallback still populates the replayed part.
2026-04-16 22:31:49 +08:00
Jason 06ff5e211a fix(proxy/gemini): reconcile synthesized tool-call ids with later real ids + preserve thoughtSignature
Three related findings on `streaming_gemini.rs` for Gemini's cumulative
`streamGenerateContent` stream, all centered on `merge_tool_call_snapshots`:

1. (P1) Match upgraded tool-call IDs by position.
   When Gemini delivers a `functionCall` without an id on chunk 1
   (cc-switch synthesizes `gemini_synth_*`) and then upgrades it to a
   real id on chunk 2, the `Some(incoming_id)` branch only matched by
   id and missed the existing synthesized snapshot. A second entry
   would be pushed, yielding duplicate `tool_use` content blocks at
   stream end — one with the synthesized id, one with the real id —
   which could trigger duplicate tool execution and break tool_result
   correlation. Add a positional fallback: when no id match exists but
   the same-position slot holds a synthesized id, merge into it.
   `or(preserved_id)` already lets the real id win the merge.

2. (P2) Preserve prior thoughtSignature when merging snapshots.
   `tool_call_snapshots[index] = tool_call` overwrote the slot
   entirely, dropping any `thoughtSignature` captured on an earlier
   chunk if the current cumulative snapshot omitted it. Since
   `build_shadow_assistant_parts` writes `thoughtSignature` into the
   shadow turn from `tool_call.thought_signature`, a dropped signature
   would cause later replay requests to Gemini to be rejected with
   invalid-signature errors. Preserve the existing signature when the
   incoming chunk does not carry one.

3. (P2) Document the part-order streaming trade-off.
   All `tool_use` content blocks are emitted after the final text
   `content_block_stop`, so interleaved [text, functionCall, text,
   functionCall] parts arrive at the Anthropic client as [text(concat),
   tool_use, tool_use] — different from the non-streaming transformer,
   which preserves part order. This is intentional given the cumulative
   snapshot model and the consumers we target (claude-code-like clients
   don't depend on strict interleaving for tool execution correctness).
   Add a block comment at the flush site describing the trade-off and
   what a strict-order fix would entail, so this isn't rediscovered as
   a bug later.

Regression tests:
- upgraded_real_id_merges_into_existing_synthesized_snapshot
- thought_signature_preserved_when_later_chunk_omits_it

Test count: 868 -> 870. clippy 1.95 clean. fmt clean.
2026-04-16 22:02:10 +08:00
Jason 1b0e259492 chore(lint): address clippy 1.95 findings in existing modules
CI upgraded to Rust 1.95 and flagged ten pre-existing warnings that
older toolchains did not enforce. None relate to the Gemini proxy
integration PR itself but they block CI on the feature branch, so
clean them up here as a separate commit for easy review:

collapsible_match:
- proxy/providers/gemini_schema.rs: `"items" if value.is_object()`
  match guard instead of nested if.
- proxy/providers/transform_responses.rs: fold
  `map_responses_stop_reason`'s `"completed"` / `"incomplete"` arms
  into match guards, relying on the existing `_ => "end_turn"` fall-
  through for non-matching guard conditions (semantics preserved).
- services/session_usage_codex.rs: fold
  `"session_meta" if state.session_id.is_none()` guard, relying on
  the existing `_ => {}` fall-through.

unnecessary_sort_by:
- services/provider/endpoints.rs: `sort_by_key(|ep| Reverse(ep.added_at))`.
- services/skill.rs (backup list): same Reverse idiom on `created_at`.
- services/skill.rs (skill listings x2): `sort_by_key(|s| s.name.to_lowercase())`.

useless_conversion:
- services/skill.rs: drop the explicit `.into_iter()` on `zip`'s argument.

while_let_loop:
- services/webdav_auto_sync.rs: `while let Some(wait_for) = ...`
  instead of `loop { let Some(...) = ... else { break }; ... }`.

All changes are mechanical and preserve behavior. `cargo test --lib`
remains green (868 passed).
2026-04-16 21:37:02 +08:00
Jason 6b2e5e07cc test(proxy/gemini): pin non-full-URL versioned relay base stripping
Adds two regression tests that lock in the intentional asymmetry
between full-URL and non-full-URL modes:

- Full-URL mode: opaque base path (e.g. `https://relay.example/custom/v1beta`)
  is preserved verbatim. Already covered by
  `preserves_opaque_full_url_with_bare_v1beta_suffix`.
- Non-full-URL mode: base path MUST strip `/v1`, `/v1beta`, etc. so the
  standard `/v1beta/models/{model}:method` endpoint can be appended
  without producing a doubled `/v1beta/v1beta/models/...` path.

The non-full-URL contract is "base URL + cc-switch appends the
canonical Gemini endpoint". A user who needs a relay's custom
namespace (e.g. `/v1/models/...`) must use full-URL mode and paste
the complete method path. This commit adds regression coverage so a
future attempt to mirror full-URL's host-whitelist gating into
`normalize_gemini_base_path` will fail the test suite immediately.
2026-04-16 21:36:48 +08:00
Jason 3349189864 Revert "fix(proxy/gemini): gate generic REST suffix stripping behind Google host in non-full-URL mode"
This reverts commit d19ff09cb7.
2026-04-16 21:32:04 +08:00
Jason d19ff09cb7 fix(proxy/gemini): gate generic REST suffix stripping behind Google host in non-full-URL mode
`build_gemini_native_url` unconditionally stripped `/v1`, `/v1beta`,
`/models`, and `/openai` suffixes from the base path regardless of
host. This worked for Google's own endpoints but silently rewrote
third-party relay URLs like `https://relay.example/custom/v1` to
`.../custom/v1beta/models/...`, breaking any relay that mounts its
Gemini-compatible namespace under a versioned prefix.

The result was also asymmetric with the previously-fixed full-URL
branch: toggling the "full URL" switch changed the outbound URL for
the same base_url, which is exactly the kind of invisible behavior
that makes debugging proxy deployments painful.

Align `normalize_gemini_base_path` with
`should_normalize_gemini_full_url`'s layered model:

- Unconditional: `/models/...:method` structured paths and deep
  OpenAI-compat endpoints (`/openai/chat/completions`,
  `/openai/responses` and their versioned variants) — these are
  unambiguous Gemini-specific grammar on any host.
- Google-host gated: generic `/v1`, `/v1beta`, `/models`, `/openai`
  suffixes only get stripped on `generativelanguage.googleapis.com`,
  `aiplatform.googleapis.com`, or `*-aiplatform.googleapis.com`.
  Other hosts preserve the prefix verbatim so relays keep their
  intended routing.

Adds seven regression tests for the non-full-URL flow: opaque relay
preservation (v1 / v1beta / models / openai suffix variants), Google
host normalization (counter-case), and boundary cases (structured
method path and deep OpenAI-compat endpoint stripped regardless of
host).

Test count: 864 -> 873.
2026-04-16 21:20:59 +08:00
Jason 6ef8d2f56a fix(proxy/gemini): trim API key before provider-type detection and OAuth parsing
Leading whitespace on a copied oauth_creds.json (e.g. trailing newline
when the user copies the file content as-is) would slip past the
`starts_with("ya29.") || starts_with('{')` prefix check in
`ClaudeAdapter::provider_type`, causing the provider to be misclassified
as raw-API-key Gemini and fall back to `x-goog-api-key` with the raw
JSON as the key — which upstream rejects with 401.

The frontend's `handleApiKeyChange` already trims on keystrokes but
deep-link imports, the JSON editor, and live-config backfill all bypass
that path. Trim at every backend extraction point so the coverage is
uniform:

- `ClaudeAdapter::extract_key` (5 env / fallback branches) gets
  `.map(str::trim)` before `.filter(|s| !s.is_empty())` so that
  whitespace-only values are also treated as missing.
- `GeminiAdapter::extract_key_raw` gets the same chain (including
  the `.filter` it was missing before).
- `GeminiAdapter::parse_oauth_credentials` gets a defensive
  `let key = key.trim();` at the entry as a belt-and-suspenders guard.

Adds two regression tests covering JSON and bare `ya29.` keys with
leading newline/space.
2026-04-16 21:20:51 +08:00
Jason 7a5c3725b0 fix(proxy/gemini): gate /v1beta behind Google host + normalize models/ model id prefix
Two related P2 corrections to the Gemini Native URL surface, both
folding into the existing Google-host-whitelist architecture.

## P2a — `/v1beta` suffix should not unconditionally trigger rewrite

`should_normalize_gemini_full_url` placed `/v1beta` and `/v1beta/models`
in the unconditional layer on the reasoning that `/v1beta` is
Google-specific. In practice an opaque relay fronting a non-Gemini
service at `https://relay.example/custom/v1beta` would still be
silently rewritten to `/v1beta/models/{model}:generateContent`,
breaking the deployment.

Move `/v1beta`, `/v1beta/models`, and `/v1beta/openai` into the
Google-host gated layer alongside `/v1`, `/models`, and friends. The
unconditional layer now only accepts paths whose grammar is
intrinsically Gemini — `/models/...:generateContent` method calls and
the deep OpenAI-compat endpoints like `/openai/chat/completions` and
`/openai/responses`. Pasted AI-Studio URLs such as
`https://generativelanguage.googleapis.com/v1beta` still normalize
because the host matches the whitelist.

## P2b — `model: "models/gemini-2.5-pro"` produced doubled path prefix

Gemini SDKs (and the official `list_models` response) commonly surface
model ids in resource-name form `models/gemini-2.5-pro`. Raw
interpolation into `format!("/v1beta/models/{model}:...")` produced
`/v1beta/models/models/gemini-2.5-pro:streamGenerateContent` which
upstream rejects — yielding false-negative health checks for otherwise
valid provider configs.

Introduce `normalize_gemini_model_id(&str) -> &str` in `gemini_url`
as the single source of truth: strips an optional leading `/` then an
optional `models/` prefix, leaving bare ids untouched. Apply in the
three call sites that build a Gemini method URL:
- `services/stream_check.rs::resolve_claude_stream_url` (unified path)
- `services/stream_check.rs::check_gemini_stream` (Gemini-only path)
- `proxy/forwarder.rs::rewrite_claude_transform_endpoint` (production)

Tests (9 new):
- `gemini_url`: 3 regressions for opaque vs Google-host `/v1beta*`
  handling + 5 unit tests pinning `normalize_gemini_model_id` behavior
  (strip prefix, leave bare id, preserve nested slashes past the one
  stripped prefix, tolerate leading slash, pass through empty input).
- `stream_check`: one end-to-end regression confirming
  `models/gemini-2.5-pro` collapses to the expected single-prefix URL.
- `forwarder`: one end-to-end regression on the production rewrite
  path.

All 864 lib tests pass; cargo fmt + clippy -D warnings clean.

Addresses Codex P2 feedback on #1918.
2026-04-16 20:50:07 +08:00
Jason 1e04f68745 refactor(proxy/gemini): share build_anthropic_usage between stream and non-stream paths
`streaming_gemini::anthropic_usage_from_gemini` and
`transform_gemini::build_anthropic_usage` were byte-for-byte identical
(32 lines each) — both converting Gemini `usageMetadata` into the
Anthropic `usage` shape including `cache_read_input_tokens` mapping.

Promote the non-streaming version to `pub(crate)` and reuse it from the
streaming SSE converter. Removes ~30 lines of duplication and guarantees
the two paths cannot drift apart.

No behavioral change; all 854 lib tests pass; cargo fmt + clippy -D
warnings clean.
2026-04-16 17:50:58 +08:00
Jason b02b8f3b1b fix(proxy/gemini): align shadow id with client-visible id in non-streaming path
When Gemini returns a `functionCall` without an id (common in 2.x
parallel calls), `gemini_to_anthropic_with_shadow_and_hints` previously
generated TWO independent synthesized UUIDs:

  1. Line 186-197 — synthesized id `A` used for the Anthropic-visible
     `content[tool_use].id` returned to the client.
  2. Line 850-881 — `extract_tool_call_meta` independently synthesized
     id `B ≠ A`, which populated `shadow_turn.tool_calls[i].id`.

`shadow_content` (line 225-228, cloned from `rectified_parts`) retained
the original missing/empty id. Result: the client sees id `A`, the
shadow store holds id `B`.

On the next turn, `convert_messages_to_contents` builds
`tool_name_by_id` from `build_tool_name_map_from_shadow_turns`, which
uses `tool_calls[i].id` — so the map contains `B → name` but not
`A → name`. When the client sends back `tool_result(tool_use_id=A)`,
resolution fails with:

  Unable to resolve Gemini functionResponse.name for tool_use_id `A`

This affects both truncated histories (client sends only the
tool_result) and full histories (shadow-replay branch at line 342-354
skips `convert_message_content_to_parts`, so the assistant tool_use
block never registers id `A` itself).

Fix: make `rectified_parts` the single source of truth. After
`rectify_tool_call_parts`, run a pre-pass that writes
`synthesize_tool_call_id()` back into any `functionCall` that lacks a
non-empty id. All three readers — the content builder (186-197), the
shadow_content clone (225-228), and `extract_tool_call_meta` — then
observe the same id. `shadow_parts()` already strips synthesized ids on
replay (line 616-628), so the internal identifier never leaks to
Gemini upstream.

This mirrors the streaming path, which already has single-source-of-
truth semantics via `tool_call_snapshots` in `streaming_gemini.rs` —
no change needed there.

Tests (5 new in `transform_gemini::tests`):
- `non_stream_shadow_id_matches_client_visible_id`: asserts
  `response.content[0].id == shadow.tool_calls[0].id ==
  shadow.assistant_content.parts[0].functionCall.id`.
- `non_stream_missing_id_scenario_a_truncated_history_resolves`: turn 2
  sends only `[tool_result(id=A)]`; resolution must succeed.
- `non_stream_missing_id_scenario_b_full_history_replay_resolves`: turn 2
  sends `[assistant(tool_use=A), tool_result(A)]`; shadow-replay branch
  strips the synth id from outgoing `functionCall` while still
  resolving the subsequent `tool_result`.
- `non_stream_preserves_original_gemini_id_when_present`: regression —
  genuine Gemini ids flow through unchanged.
- `non_stream_synthesized_id_not_leaked_to_gemini_via_shadow_replay`:
  defensive — shadow-replay path must strip synth ids from both
  `functionCall.id` and `functionResponse.id`.

All 854 lib tests pass; cargo fmt + clippy -D warnings clean.

Addresses Codex follow-up P1 on #1918.
2026-04-16 17:38:07 +08:00
Jason 1f3ebe9c62 fix(proxy/gemini): gate generic REST path suffixes behind Google host whitelist
`should_normalize_gemini_full_url` previously treated any full URL whose
path ends with `/v1`, `/v1/models`, `/models`, `/v1/openai`, or `/openai`
as a structured Gemini endpoint and rewrote it to
`/v1beta/models/{model}:generateContent`. These are ubiquitous REST
conventions — opaque relays such as `https://relay.example/custom/v1`
legitimately use them for fixed endpoints — so the rewrite silently
routed traffic to the wrong upstream path.

Split the predicate into two layers:

- **Unconditional**: `matches_structured_gemini_models_path` (i.e. a
  `/models/...:generateContent` method call anywhere in the path), the
  Google-specific `/v1beta*` family, and the deep OpenAI-compat paths
  (`/v1beta/openai/chat/completions`, `/openai/chat/completions`, and
  their `responses` siblings). These remain host-agnostic because the
  path grammar itself is Gemini-specific.
- **Google-host gated**: `/v1`, `/v1/models`, `/models`, `/v1/openai`,
  `/openai`. Only normalized when the host is one of
  `generativelanguage.googleapis.com`, `aiplatform.googleapis.com`, or a
  real `*-aiplatform.googleapis.com` Vertex regional endpoint. The match
  is exact/suffix (not `contains`), so lookalike hosts like
  `aiplatform.example.com` are correctly treated as opaque relays.

Tests (8 new in `gemini_url::tests`):
- Four opaque-relay cases: `/custom/v1`, `/custom/models`,
  `/custom/v1/models`, `/custom/openai` — all preserved as-is.
- Three Google-host counter-cases: `/v1`, `/models`, and
  `us-central1-aiplatform.googleapis.com/v1` still normalize.
- One lookalike safety case: `aiplatform.example.com/v1` is NOT
  treated as Google.

All 849 lib tests pass; cargo fmt + clippy -D warnings clean.

Addresses Codex review P2 on #1918.
2026-04-16 17:35:11 +08:00
Jason baaefcc2fc fix(proxy/gemini): treat empty-string functionCall id as missing in streaming path
Follow-up to the earlier P1 fix: some Gemini relays serialize an absent
functionCall id as `"id": ""` instead of omitting the field. The
non-streaming `extract_tool_call_meta` already filters these via
`.filter(|s| !s.is_empty())`, but the streaming counterpart
`extract_tool_calls` passed the empty string straight through
`function_call.get("id").and_then(|v| v.as_str())` into
`GeminiToolCallMeta::new`, producing a `Some("")` id.

Downstream, `merge_tool_call_snapshots` would then match two parallel
no-id calls against each other on their shared empty-string id,
collapsing them into a single snapshot (silent data loss for the first
call) and emitting an Anthropic `tool_use.id: ""` that breaks tool_result
correlation on the Claude Code client.

Fix:
- `extract_tool_calls`: apply the same `filter(|s| !s.is_empty())` guard
  used in the non-streaming path so empty strings become `None` before
  reaching the shadow meta.
- `merge_tool_call_snapshots`: defensively collapse any incoming
  `Some("")` to `None` up front — keeps the "missing vs present" invariant
  local to the merge step for future callers that might build
  `GeminiToolCallMeta` by hand.

Tests (2 new, both in streaming_gemini):
- `parallel_empty_string_id_calls_are_treated_as_missing_and_preserved`
  covers two parallel calls with explicit `"id": ""` — asserts both
  surface, no empty tool_use id leaks, and each gets a unique
  `gemini_synth_` id.
- `single_empty_string_id_tool_call_gets_synthesized_id` covers the
  non-parallel degraded-relay case.

All 841 lib tests pass; cargo fmt + clippy -D warnings clean.

Addresses Codex follow-up P1 on #1918.
2026-04-16 17:14:22 +08:00
Jason 8b65488a89 fix(proxy/gemini): narrow URL normalization + guard empty OAuth access_token
P2a — Preserve opaque relay URLs that contain `/v1/models/` prefixes.

`should_normalize_gemini_full_url` previously flagged any full URL whose
path merely contained `/v1beta/models/` or `/v1/models/` as a structured
Gemini endpoint, forcing rewrite to `.../v1beta/models/{model}:method`.
This silently dropped legitimate relay route segments (e.g.
`https://relay.example/v1/models/invoke` → `.../v1beta/models/...:generateContent`,
losing `/invoke`) and sent traffic to the wrong upstream path.

Replace the bare `contains(...)` checks with
`matches_structured_gemini_models_path`, which requires the
`/models/` segment to be followed by a canonical Gemini method call
(`*:generateContent` or `*:streamGenerateContent`). The
`matches_bare_gemini_models_path` helper is generalized (and renamed) to
handle both `/v1beta/models/` and `/v1/models/` alongside the original
bare `/models/` shape.

P2b — Reject empty Gemini OAuth access_tokens before they reach the
bearer header.

`GeminiAdapter::parse_oauth_credentials` accepts refresh-token-only JSON
(and surfaces `{"access_token": "", ...}` for expired credentials) with
`access_token` defaulting to `""`. The Claude adapter's GeminiCli branch
then called `AuthInfo::with_access_token(key, creds.access_token)`
unconditionally, so the bearer-header builder at
`AuthStrategy::GoogleOAuth` resolved to `Authorization: Bearer ` — a
deterministic 401 from upstream.

CC Switch does not currently exchange the refresh_token for a fresh
access_token (`OAuthCredentials::needs_refresh` / `can_refresh` are
annotated `#[allow(dead_code)]`). Until that exists, only attach
`access_token` when it is non-empty; fall back to plain GoogleOAuth
strategy with the raw key and log a warn pointing users at
`~/.gemini/oauth_creds.json` so the failure mode is observable.

Tests:
- gemini_url.rs: three new regressions — opaque `/v1/models/invoke`,
  opaque `/v1beta/models/route`, and the positive counter-case where a
  structured `/v1/models/...:generateContent` path still normalizes.
- claude.rs: three new `test_extract_auth_gemini_cli_*` tests covering
  refresh-only JSON, empty-string access_token JSON, and the valid-JSON
  pass-through.

All 839 lib tests pass; cargo fmt + clippy -D warnings clean.

Addresses Codex review P2 findings on #1918.
2026-04-16 16:55:11 +08:00
Jason 128d431084 fix(proxy/gemini): synthesize unique ids for no-id tool calls + enforce object params schema
P1 — Parallel tool calls without Gemini-assigned ids no longer collapse.
Gemini 2.x native parallel `functionCall` entries may omit the `id` field.
The previous `merge_tool_call_snapshots` fell back to matching by `name`,
which silently merged two parallel calls to the same function into one
entry — dropping the first call's args. The non-streaming path and shadow
store further bottlenecked on empty-string ids: multiple `tool_use` blocks
shared the same id, and `tool_name_by_id.get("")` could only return one
mapping, causing later `tool_result` round-trips to fail with
`Unable to resolve Gemini functionResponse.name` or bind to the wrong tool.

Fix: introduce `synthesize_tool_call_id()` producing `gemini_synth_<uuid>`.
Both streaming and non-streaming response paths now guarantee every
Anthropic-visible tool_use carries a unique id. `merge_tool_call_snapshots`
matches by id first, falling back to the `parts` array position (for the
cumulative-streaming case) while preserving the synthesized id across
chunks. `convert_message_content_to_parts` detects the synthetic prefix
and strips the id from outbound `functionCall`/`functionResponse` so the
internal identifier never leaks upstream. `shadow_parts` performs the
same strip when replaying a recorded assistant turn.

P2 — Vertex AI rejects empty `parameters` schemas. When an Anthropic tool
arrives with missing or empty `input_schema`, the proxy used to emit
`"parameters": {}` (no `type`), which fails Vertex AI validation with
`functionDeclaration parameters schema should be of type OBJECT`.
Contrary to the automated-review suggestion, the fix is not to omit
`parameters` (that too is rejected) but to normalize to the canonical
empty-object form `{type: "object", properties: {}}`.
Refs: google-gemini/generative-ai-python#423, BerriAI/litellm#5055.

Fix: new `ensure_object_schema` helper in `gemini_schema` promotes
missing `type` to `"object"` and adds empty `properties` when absent,
while leaving atomic (non-object) schemas untouched.

Tests: seven new regressions covering parallel no-id calls, cumulative
chunk id reuse, synthetic-id round-trip both directions, shadow replay
id stripping, and the three Vertex-AI schema shapes.

The two existing wrapper functions (`gemini_to_anthropic` and
`gemini_to_anthropic_with_shadow`) gain `#[allow(dead_code)]` to clear
a pre-existing clippy -D warnings failure — they are part of the public
transform API surface and intentionally kept for future callers.

Addresses Codex review P1/P2 on #1918.
2026-04-16 16:22:52 +08:00
Jason 2f4a680d70 style: apply cargo fmt to pass Backend Checks CI
Wrap prompt_cache_key chained call across lines per rustfmt default
formatting. Pure formatting change, no behavior difference.
2026-04-16 13:11:33 +08:00
YoVinchen 2309b49ed5 Keep Gemini tool replay stable across Claude request boundaries
Claude Code follow-up requests were still falling back to locally reconstructed functionCall parts, which dropped Gemini thought signatures and triggered INVALID_ARGUMENT errors from the official Gemini API. The replay path needed to survive real Claude request boundaries, not just idealized in-process test flows.

This change makes Claude requests reuse X-Claude-Code-Session-Id as the shadow session key, records streamed Gemini tool turns before tool_use events are fully drained, and matches assistant tool_use turns to shadow state by tool_use id and normalized tool name before positional fallback. Together these fixes keep thoughtSignature-bearing Gemini tool calls available for the next request in the loop.

Constraint: Claude Code sends a stable X-Claude-Code-Session-Id header while metadata.session_id may be absent on follow-up requests
Rejected: Rely on metadata-only Claude session extraction | generated fresh session ids and broke cross-request shadow replay
Rejected: Record Gemini shadow only after streaming completes | loses the race when the client sends the next request immediately after tool_use
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Preserve Gemini shadow continuity across requests by keying Claude sessions from the header first and persisting tool-call shadow before yielding tool_use events downstream
Tested: cargo fmt --manifest-path src-tauri/Cargo.toml --all; cargo test --manifest-path src-tauri/Cargo.toml test_extract_session_from_claude_header; cargo test --manifest-path src-tauri/Cargo.toml test_extract_session_from_claude_header_precedes_metadata; cargo test --manifest-path src-tauri/Cargo.toml stores_tool_shadow_before_tool_use_events_are_fully_drained; cargo test --manifest-path src-tauri/Cargo.toml shadow_replay_matches_tool_use_turn_by_id_when_position_drifts; cargo test --manifest-path src-tauri/Cargo.toml shadow_replay_aligns_to_latest_turns_after_client_truncation
Not-tested: Full src-tauri test suite without test filters; live end-to-end Gemini relay after this exact commit hash
2026-04-16 11:31:23 +08:00
YoVinchen 8bdc4d6326 Prevent Gemini review regressions in streaming and tool rectification
PR #1918 review feedback exposed two correctness issues in the Gemini Native adapter path. Gemini SSE buffering was still using lossy UTF-8 decoding, which could corrupt split multibyte payloads and drop streamed output. Tool arg rectification also removed top-level parameters eagerly, which broke tools that legitimately define a parameters field.

This change moves Gemini SSE buffering onto the existing append_utf8_safe path and makes parameters flattening conditional on the schema actually expecting nested extraction. The old Skill rectification path stays intact, and new regression tests cover both the preserved parameters case and UTF-8-split JSON payloads.

Constraint: Existing PR #1918 review feedback must be fixed without staging unrelated local docs and artifact files
Rejected: Keep String::from_utf8_lossy in Gemini SSE buffering | corrupts split multibyte payloads and can drop JSON chunks
Rejected: Always preserve the parameters wrapper | regresses the existing nested-parameters rectification path for Skill-style tools
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep Gemini SSE buffering on the UTF-8-safe accumulator path and only unwrap parameters when the target schema does not declare it as a legitimate field
Tested: cargo fmt --manifest-path src-tauri/Cargo.toml --all; cargo test --manifest-path src-tauri/Cargo.toml preserves_utf8_boundaries_when_json_payload_spans_chunks; cargo test --manifest-path src-tauri/Cargo.toml gemini_to_anthropic_rectifies_tool_args_from_schema_hints; cargo test --manifest-path src-tauri/Cargo.toml rectifies_streamed_skill_args_from_nested_parameters; cargo test --manifest-path src-tauri/Cargo.toml gemini_to_anthropic_preserves_legitimate_parameters_arg
Not-tested: Full src-tauri test suite; live end-to-end Gemini relay traffic against upstream services
2026-04-16 10:57:54 +08:00
YoVinchen a186f6baca Merge branch 'main' into feat/gemini-proxy-integration 2026-04-15 17:01:21 +08:00
YoVinchen 32e3cad141 Merge branch 'main' into feat/gemini-proxy-integration
# Conflicts:
#	src-tauri/src/proxy/providers/claude.rs
#	src/i18n/locales/en.json
#	src/i18n/locales/ja.json
#	src/i18n/locales/zh.json
2026-04-15 16:58:26 +08:00
YoVinchen 9bbc8bb789 fix: revert unrelated cache_key change in claude proxy transform
Restore .unwrap_or(&provider.id) fallback for cache_key to match main
branch behavior. Only gemini_native related changes should be in this branch.
2026-04-11 16:22:07 +08:00
YoVinchen d52a811016 Merge branch 'main' into feat/gemini-proxy-integration
Resolve conflict in claudeProviderPresets.ts: keep both Gemini Native
and Shengsuanyun presets.
2026-04-11 15:50:48 +08:00
YoVinchen 92668f610e Merge branch 'main' into feat/gemini-proxy-integration
# Conflicts:
#	src-tauri/src/proxy/providers/claude.rs
#	src-tauri/src/proxy/sse.rs
#	src-tauri/src/services/stream_check.rs
2026-04-10 09:08:39 +08:00
YoVinchen 44ac926210 fix(proxy): align shadow turns to tail on client history truncation 2026-04-06 23:42:47 +08:00
YoVinchen 7cf554c011 feat(proxy): update Gemini streaming and transformation logic 2026-04-06 12:23:19 +08:00
YoVinchen 8f300f4591 feat(proxy): add Gemini Native tool argument rectification 2026-04-06 11:53:06 +08:00
YoVinchen 70886d9847 feat(ui): add Gemini Native provider preset and api format option
- Add gemini_native to ClaudeApiFormat type and ProviderMeta.apiFormat
- Add "Gemini Native" provider preset with default Google AI endpoints
- Show Gemini-specific endpoint hints and full-URL mode guidance
- Add gemini_native option to API format selector in ClaudeFormFields
- Add i18n strings for zh/en/ja
2026-04-06 11:12:41 +08:00
YoVinchen 8603d4a373 feat(proxy): wire Gemini Native format into proxy core and Claude adapter
Integrate gemini_native api_format throughout the proxy pipeline:
- ClaudeAdapter: detect Gemini provider type, Google/GoogleOAuth auth
  strategies, and suppress Anthropic-specific headers for Gemini targets
- Forwarder: Gemini URL resolution, shadow store threading, endpoint
  rewriting to models/*:generateContent with stream/non-stream variants
- Handlers: route Gemini streaming through streaming_gemini adapter and
  non-streaming through transform_gemini converter
- Server/State: add GeminiShadowStore to shared ProxyState
- StreamCheck: support gemini_native health check with proper auth headers
2026-04-06 11:12:11 +08:00
YoVinchen 8c09d8c7d3 feat(proxy): add Gemini Native schema, shadow store, transform, and streaming
- gemini_schema: Gemini generateContent request/response type definitions
- gemini_shadow: session-scoped shadow store for thinking signature and
  tool-call state replay across streaming chunks
- transform_gemini: bidirectional Anthropic Messages ↔ Gemini Native
  request/response conversion with thinking block and tool-use support
- streaming_gemini: Gemini SSE → Anthropic SSE streaming adapter with
  incremental thinking/text/tool_use delta emission
2026-04-06 11:11:30 +08:00
YoVinchen 3ce8245f2d feat(proxy): add Gemini Native URL builder and full-URL resolver
Introduce gemini_url module that normalizes legacy Gemini/OpenAI-compatible
base URLs into canonical models/*:generateContent endpoints. Supports both
structured Gemini URLs (auto-normalized) and opaque relay URLs (pass-through
with query params only).
2026-04-06 11:10:42 +08:00
YoVinchen abbf43ea6f refactor(proxy): extract take_sse_block helper with CRLF delimiter support
Replace inline `buffer.find("\n\n")` SSE splitting logic across streaming,
streaming_responses, response_handler, and response_processor with a shared
`take_sse_block` function that handles both `\n\n` and `\r\n\r\n` delimiters.
2026-04-06 11:09:23 +08:00
33 changed files with 5542 additions and 94 deletions
+174 -9
View File
@@ -9,7 +9,10 @@ use super::{
failover_switch::FailoverSwitchManager,
log_codes::fwd as log_fwd,
provider_router::ProviderRouter,
providers::{get_adapter, AuthInfo, AuthStrategy, ProviderAdapter, ProviderType},
providers::{
gemini_shadow::GeminiShadowStore, get_adapter, AuthInfo, AuthStrategy, ProviderAdapter,
ProviderType,
},
thinking_budget_rectifier::{rectify_thinking_budget, should_rectify_thinking_budget},
thinking_rectifier::{
normalize_thinking_type, rectify_anthropic_request, should_rectify_thinking_signature,
@@ -43,12 +46,15 @@ pub struct RequestForwarder {
router: Arc<ProviderRouter>,
status: Arc<RwLock<ProxyStatus>>,
current_providers: Arc<RwLock<std::collections::HashMap<String, (String, String)>>>,
gemini_shadow: Arc<GeminiShadowStore>,
/// 故障转移切换管理器
failover_manager: Arc<FailoverSwitchManager>,
/// AppHandle,用于发射事件和更新托盘
app_handle: Option<tauri::AppHandle>,
/// 请求开始时的"当前供应商 ID"(用于判断是否需要同步 UI/托盘)
current_provider_id_at_start: String,
/// 代理会话 ID(用于 Gemini Native shadow replay
session_id: String,
/// 整流器配置
rectifier_config: RectifierConfig,
/// 优化器配置
@@ -66,9 +72,11 @@ impl RequestForwarder {
non_streaming_timeout: u64,
status: Arc<RwLock<ProxyStatus>>,
current_providers: Arc<RwLock<std::collections::HashMap<String, (String, String)>>>,
gemini_shadow: Arc<GeminiShadowStore>,
failover_manager: Arc<FailoverSwitchManager>,
app_handle: Option<tauri::AppHandle>,
current_provider_id_at_start: String,
session_id: String,
_streaming_first_byte_timeout: u64,
_streaming_idle_timeout: u64,
rectifier_config: RectifierConfig,
@@ -79,9 +87,11 @@ impl RequestForwarder {
router,
status,
current_providers,
gemini_shadow,
failover_manager,
app_handle,
current_provider_id_at_start,
session_id,
rectifier_config,
optimizer_config,
copilot_optimizer_config,
@@ -918,7 +928,7 @@ impl RequestForwarder {
let api_format = resolved_claude_api_format
.as_deref()
.unwrap_or_else(|| super::providers::get_claude_api_format(provider));
rewrite_claude_transform_endpoint(endpoint, api_format, is_copilot)
rewrite_claude_transform_endpoint(endpoint, api_format, is_copilot, &mapped_body)
} else {
(
endpoint.to_string(),
@@ -928,7 +938,13 @@ impl RequestForwarder {
)
};
let url = if is_full_url {
let url = if matches!(resolved_claude_api_format.as_deref(), Some("gemini_native")) {
super::gemini_url::resolve_gemini_native_url(
&base_url,
&effective_endpoint,
is_full_url,
)
} else if is_full_url {
append_query_to_full_url(&base_url, passthrough_query.as_deref())
} else {
adapter.build_url(&base_url, &effective_endpoint)
@@ -944,6 +960,8 @@ impl RequestForwarder {
mapped_body,
provider,
api_format,
Some(&self.session_id),
Some(self.gemini_shadow.as_ref()),
)?
} else {
adapter.transform_request(mapped_body, provider)?
@@ -1143,8 +1161,11 @@ impl RequestForwarder {
.ok()
.and_then(|u| u.authority().map(|a| a.to_string()));
let should_send_anthropic_headers = adapter.name() == "Claude"
&& matches!(resolved_claude_api_format.as_deref(), Some("anthropic"));
// 预计算 anthropic-beta 值(仅 Claude
let anthropic_beta_value = if adapter.name() == "Claude" {
let anthropic_beta_value = if should_send_anthropic_headers {
const CLAUDE_CODE_BETA: &str = "claude-code-20250219";
Some(if let Some(beta) = headers.get("anthropic-beta") {
if let Ok(beta_str) = beta.to_str() {
@@ -1264,8 +1285,10 @@ impl RequestForwarder {
// --- anthropic-version — 透传客户端值 ---
if key_str.eq_ignore_ascii_case("anthropic-version") {
saw_anthropic_version = true;
ordered_headers.append(key.clone(), value.clone());
if should_send_anthropic_headers {
saw_anthropic_version = true;
ordered_headers.append(key.clone(), value.clone());
}
continue;
}
@@ -1306,7 +1329,7 @@ impl RequestForwarder {
}
// anthropic-version:仅在缺失时补充默认值
if adapter.name() == "Claude" && !saw_anthropic_version {
if should_send_anthropic_headers && !saw_anthropic_version {
ordered_headers.append(
"anthropic-version",
http::HeaderValue::from_static("2023-06-01"),
@@ -1650,6 +1673,7 @@ fn rewrite_claude_transform_endpoint(
endpoint: &str,
api_format: &str,
is_copilot: bool,
body: &Value,
) -> (String, Option<String>) {
let (path, query) = split_endpoint_and_query(endpoint);
let passthrough_query = if is_claude_messages_path(path) {
@@ -1662,6 +1686,36 @@ fn rewrite_claude_transform_endpoint(
return (endpoint.to_string(), passthrough_query);
}
if api_format == "gemini_native" {
let model =
super::providers::transform_gemini::extract_gemini_model(body).unwrap_or("unknown");
// Accept both bare ids (`gemini-2.5-pro`) and the resource-name
// form (`models/gemini-2.5-pro`) that Gemini SDKs emit. See
// `normalize_gemini_model_id` for rationale.
let model = super::gemini_url::normalize_gemini_model_id(model);
let is_stream = body
.get("stream")
.and_then(|value| value.as_bool())
.unwrap_or(false);
let target_path = if is_stream {
format!("/v1beta/models/{model}:streamGenerateContent")
} else {
format!("/v1beta/models/{model}:generateContent")
};
let rewritten_query = merge_query_params(
passthrough_query.as_deref(),
if is_stream { Some("alt=sse") } else { None },
);
let rewritten = match rewritten_query.as_deref() {
Some(query) if !query.is_empty() => format!("{target_path}?{query}"),
_ => target_path,
};
return (rewritten, rewritten_query);
}
let target_path = if is_copilot && api_format == "openai_responses" {
"/v1/responses"
} else if is_copilot {
@@ -1680,6 +1734,26 @@ fn rewrite_claude_transform_endpoint(
(rewritten, passthrough_query)
}
fn merge_query_params(base_query: Option<&str>, extra_param: Option<&str>) -> Option<String> {
let mut params: Vec<String> = base_query
.into_iter()
.flat_map(|query| query.split('&'))
.filter(|pair| !pair.is_empty())
.filter(|pair| !pair.starts_with("alt="))
.map(ToString::to_string)
.collect();
if let Some(extra_param) = extra_param {
params.push(extra_param.to_string());
}
if params.is_empty() {
None
} else {
Some(params.join("&"))
}
}
fn append_query_to_full_url(base_url: &str, query: Option<&str>) -> String {
match query {
Some(query) if !query.is_empty() => {
@@ -1808,6 +1882,7 @@ mod tests {
"/v1/messages?beta=true&foo=bar",
"openai_chat",
false,
&json!({ "model": "gpt-5.4" }),
);
assert_eq!(endpoint, "/v1/chat/completions?foo=bar");
@@ -1820,6 +1895,7 @@ mod tests {
"/claude/v1/messages?beta=true&x-id=1",
"openai_responses",
false,
&json!({ "model": "gpt-5.4" }),
);
assert_eq!(endpoint, "/v1/responses?x-id=1");
@@ -1828,8 +1904,12 @@ mod tests {
#[test]
fn rewrite_claude_transform_endpoint_uses_copilot_path() {
let (endpoint, passthrough_query) =
rewrite_claude_transform_endpoint("/v1/messages?beta=true&x-id=1", "anthropic", true);
let (endpoint, passthrough_query) = rewrite_claude_transform_endpoint(
"/v1/messages?beta=true&x-id=1",
"anthropic",
true,
&json!({ "model": "claude-sonnet-4-6" }),
);
assert_eq!(endpoint, "/chat/completions?x-id=1");
assert_eq!(passthrough_query.as_deref(), Some("x-id=1"));
@@ -1841,12 +1921,60 @@ mod tests {
"/v1/messages?beta=true&x-id=1",
"openai_responses",
true,
&json!({ "model": "gpt-5.4" }),
);
assert_eq!(endpoint, "/v1/responses?x-id=1");
assert_eq!(passthrough_query.as_deref(), Some("x-id=1"));
}
#[test]
fn rewrite_claude_transform_endpoint_maps_gemini_generate_content() {
let (endpoint, passthrough_query) = rewrite_claude_transform_endpoint(
"/v1/messages?beta=true&x-id=1",
"gemini_native",
false,
&json!({ "model": "gemini-2.5-pro" }),
);
assert_eq!(
endpoint,
"/v1beta/models/gemini-2.5-pro:generateContent?x-id=1"
);
assert_eq!(passthrough_query.as_deref(), Some("x-id=1"));
}
/// Regression: body.model arriving as the resource-name form
/// `models/gemini-2.5-pro` must not produce a doubled
/// `/v1beta/models/models/...` path.
#[test]
fn rewrite_claude_transform_endpoint_strips_gemini_model_resource_prefix() {
let (endpoint, _) = rewrite_claude_transform_endpoint(
"/v1/messages",
"gemini_native",
false,
&json!({ "model": "models/gemini-2.5-pro" }),
);
assert_eq!(endpoint, "/v1beta/models/gemini-2.5-pro:generateContent");
}
#[test]
fn rewrite_claude_transform_endpoint_maps_gemini_streaming() {
let (endpoint, passthrough_query) = rewrite_claude_transform_endpoint(
"/v1/messages?beta=true",
"gemini_native",
false,
&json!({ "model": "gemini-2.5-flash", "stream": true }),
);
assert_eq!(
endpoint,
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
assert_eq!(passthrough_query.as_deref(), Some("alt=sse"));
}
#[test]
fn append_query_to_full_url_preserves_existing_query_string() {
let url = append_query_to_full_url("https://relay.example/api?foo=bar", Some("x-id=1"));
@@ -1854,6 +1982,43 @@ mod tests {
assert_eq!(url, "https://relay.example/api?foo=bar&x-id=1");
}
#[test]
fn build_gemini_native_url_uses_origin_when_base_ends_with_v1beta() {
let url = crate::proxy::gemini_url::build_gemini_native_url(
"https://generativelanguage.googleapis.com/v1beta",
"/v1beta/models/gemini-2.5-pro:generateContent",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:generateContent"
);
}
#[test]
fn build_gemini_native_url_uses_origin_when_base_already_contains_models_prefix() {
let url = crate::proxy::gemini_url::build_gemini_native_url(
"https://generativelanguage.googleapis.com/v1beta/models",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
#[test]
fn resolve_gemini_native_url_keeps_opaque_full_url_as_is() {
let url = crate::proxy::gemini_url::resolve_gemini_native_url(
"https://relay.example/custom/generate-content",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/generate-content?alt=sse");
}
#[test]
fn force_identity_for_stream_flag_requests() {
let headers = HeaderMap::new();
+665
View File
@@ -0,0 +1,665 @@
//! Gemini Native URL helpers.
//!
//! Normalizes legacy Gemini/OpenAI-compatible base URLs into the canonical
//! Gemini Native `models/*:generateContent` endpoints.
/// Normalize a Gemini model identifier to its bare form, stripping an
/// optional leading `models/` (and any leading `/`) so that the value can
/// be safely interpolated into a URL template like
/// `/v1beta/models/{model}:generateContent`.
///
/// Gemini SDKs and documentation commonly surface model ids as
/// `models/gemini-2.5-pro` (the resource-name form). Passing that value
/// through to the format string would otherwise yield a doubled prefix
/// like `/v1beta/models/models/gemini-2.5-pro:generateContent`, which is
/// rejected by the upstream API and turns any health check or live
/// request into a false negative.
pub fn normalize_gemini_model_id(model: &str) -> &str {
let trimmed = model.strip_prefix('/').unwrap_or(model);
trimmed.strip_prefix("models/").unwrap_or(trimmed)
}
pub fn resolve_gemini_native_url(base_url: &str, endpoint: &str, is_full_url: bool) -> String {
if !is_full_url || should_normalize_gemini_full_url(base_url) {
return build_gemini_native_url(base_url, endpoint);
}
let base_url = base_url
.split_once('#')
.map_or(base_url, |(base, _)| base)
.trim_end_matches('/');
let (base_without_query, base_query) = split_query(base_url);
let (_, endpoint_query) = split_query(endpoint);
let mut url = base_without_query.to_string();
if let Some(query) = merge_queries(base_query, endpoint_query) {
url.push('?');
url.push_str(&query);
}
url
}
pub fn build_gemini_native_url(base_url: &str, endpoint: &str) -> String {
let base_url = base_url
.split_once('#')
.map_or(base_url, |(base, _)| base)
.trim_end_matches('/');
let (base_without_query, base_query) = split_query(base_url);
let (endpoint_without_query, endpoint_query) = split_query(endpoint);
let endpoint_path = format!("/{}", endpoint_without_query.trim_start_matches('/'));
let (origin, raw_path) = split_origin_and_path(base_without_query);
let prefix_path = normalize_gemini_base_path(raw_path);
let mut url = if prefix_path.is_empty() {
format!("{origin}{endpoint_path}")
} else {
format!("{origin}{prefix_path}{endpoint_path}")
};
if let Some(query) = merge_queries(base_query, endpoint_query) {
url.push('?');
url.push_str(&query);
}
url
}
fn should_normalize_gemini_full_url(base_url: &str) -> bool {
let base_url = base_url
.split_once('#')
.map_or(base_url, |(base, _)| base)
.trim_end_matches('/');
let (base_without_query, _) = split_query(base_url);
let (origin, path) = split_origin_and_path(base_without_query);
if path.is_empty() || path == "/" {
return true;
}
let path = path.trim_end_matches('/');
let on_google_host = is_google_gemini_host(extract_host(origin));
// Unconditional layer: only paths whose grammar is *intrinsically*
// Gemini-specific — the `/models/...:generateContent` method-call
// shape and the deep OpenAI-compat endpoints (`/openai/chat/completions`,
// `/openai/responses`) that are implausibly used as a relay's fixed
// terminal path — get rewritten regardless of host.
if matches_structured_gemini_models_path(path)
|| path.ends_with("/v1beta/openai/chat/completions")
|| path.ends_with("/v1beta/openai/responses")
|| path.ends_with("/openai/chat/completions")
|| path.ends_with("/openai/responses")
|| path.ends_with("/v1/openai/chat/completions")
|| path.ends_with("/v1/openai/responses")
{
return true;
}
// All other version / resource-root suffixes — `/v1beta`, `/v1`,
// `/models`, `/openai`, and variants — could legitimately be an
// opaque relay's fixed endpoint (`https://relay.example/custom/v1beta`
// is a real deployment shape, even if uncommon outside Google's
// ecosystem). Only rewrite when the host itself is Google's Gemini
// or Vertex AI endpoint.
if on_google_host
&& (path.ends_with("/v1beta")
|| path.ends_with("/v1beta/models")
|| path.ends_with("/v1beta/openai")
|| path.ends_with("/v1")
|| path.ends_with("/v1/models")
|| path.ends_with("/models")
|| path.ends_with("/v1/openai")
|| path.ends_with("/openai"))
{
return true;
}
false
}
/// Extract the host portion of an origin like `https://host:port` or
/// `https://host`. Returns an empty string if no host can be found (e.g.
/// bare `http://`).
fn extract_host(origin: &str) -> &str {
let after_scheme = origin.split_once("://").map_or(origin, |(_, rest)| rest);
// authority may carry credentials (`user:pass@host`) and a port
// (`host:port`). Strip userinfo first, then port.
let without_userinfo = after_scheme
.rsplit_once('@')
.map_or(after_scheme, |(_, h)| h);
let without_port = without_userinfo
.split_once(':')
.map_or(without_userinfo, |(h, _)| h);
// Strip trailing `/` defensively (split_origin_and_path already handled
// it, but this helper may be reused elsewhere).
without_port.trim_end_matches('/')
}
/// Returns true when `host` is one of Google's Gemini / Vertex AI endpoints.
/// Case-insensitive. Requires exact match or a real `-aiplatform.googleapis.com`
/// subdomain suffix — not a substring match, so lookalikes like
/// `aiplatform.example.com` are rejected.
fn is_google_gemini_host(host: &str) -> bool {
if host.is_empty() {
return false;
}
let host_lower = host.to_ascii_lowercase();
host_lower == "generativelanguage.googleapis.com"
|| host_lower == "aiplatform.googleapis.com"
|| host_lower.ends_with("-aiplatform.googleapis.com")
}
fn split_query(input: &str) -> (&str, Option<&str>) {
input
.split_once('?')
.map_or((input, None), |(path, query)| (path, Some(query)))
}
fn split_origin_and_path(base_url: &str) -> (&str, &str) {
let Some(scheme_sep) = base_url.find("://") else {
return (base_url, "");
};
let authority_start = scheme_sep + 3;
let Some(path_start_rel) = base_url[authority_start..].find('/') else {
return (base_url, "");
};
let path_start = authority_start + path_start_rel;
(&base_url[..path_start], &base_url[path_start..])
}
fn normalize_gemini_base_path(path: &str) -> String {
let path = path.trim_end_matches('/');
if path.is_empty() || path == "/" {
return String::new();
}
for marker in ["/v1beta/models/", "/v1/models/", "/models/"] {
if let Some(index) = path.find(marker) {
return normalize_prefix(&path[..index]);
}
}
for suffix in [
"/v1beta/openai/chat/completions",
"/v1/openai/chat/completions",
"/openai/chat/completions",
"/v1beta/openai/responses",
"/v1/openai/responses",
"/openai/responses",
"/v1beta/openai",
"/v1/openai",
"/openai",
"/v1beta/models",
"/v1/models",
"/models",
"/v1beta",
"/v1",
] {
if path == suffix {
return String::new();
}
if let Some(prefix) = path.strip_suffix(suffix) {
return normalize_prefix(prefix);
}
}
path.to_string()
}
fn normalize_prefix(prefix: &str) -> String {
let prefix = prefix.trim_end_matches('/');
if prefix.is_empty() || prefix == "/" {
String::new()
} else {
prefix.to_string()
}
}
/// Returns true when `path` contains a `/models/` segment followed by a
/// canonical Gemini method call (`*:generateContent` or
/// `*:streamGenerateContent`). The `/models/` segment alone is not enough:
/// opaque relay routes such as `/v1/models/invoke` or `/custom/models/foo`
/// also contain `/models/` but are not Gemini-structured and must not be
/// rewritten.
fn matches_structured_gemini_models_path(path: &str) -> bool {
let mut cursor = path;
while let Some(idx) = cursor.find("/models/") {
let after = &cursor[idx + "/models/".len()..];
if after.contains(":generateContent") || after.contains(":streamGenerateContent") {
return true;
}
// Advance past this `/models/` occurrence so a later Gemini-style
// segment in the same path (unusual but cheap to handle) can still
// match.
cursor = &cursor[idx + "/models/".len()..];
}
false
}
fn merge_queries(base_query: Option<&str>, endpoint_query: Option<&str>) -> Option<String> {
let parts: Vec<&str> = [base_query, endpoint_query]
.into_iter()
.flatten()
.flat_map(|query| query.split('&'))
.filter(|part| !part.is_empty())
.collect();
if parts.is_empty() {
None
} else {
Some(parts.join("&"))
}
}
#[cfg(test)]
mod tests {
use super::{build_gemini_native_url, normalize_gemini_model_id, resolve_gemini_native_url};
#[test]
fn strips_version_root_for_official_base() {
let url = build_gemini_native_url(
"https://generativelanguage.googleapis.com/v1beta",
"/v1beta/models/gemini-2.5-pro:generateContent",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:generateContent"
);
}
#[test]
fn strips_openai_compat_path_for_official_base() {
let url = build_gemini_native_url(
"https://generativelanguage.googleapis.com/v1beta/openai/chat/completions",
"/v1beta/models/gemini-2.5-pro:generateContent",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:generateContent"
);
}
#[test]
fn preserves_custom_proxy_prefix_while_stripping_openai_suffix() {
let url = build_gemini_native_url(
"https://proxy.example.com/google/v1beta/openai/chat/completions",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
);
assert_eq!(
url,
"https://proxy.example.com/google/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
#[test]
fn strips_model_method_path_from_full_url_base() {
let url = build_gemini_native_url(
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:generateContent",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
#[test]
fn resolves_structured_full_url_by_normalizing_to_requested_method() {
let url = resolve_gemini_native_url(
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:generateContent",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
#[test]
fn resolves_opaque_full_url_without_appending_gemini_models_path() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/generate-content",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/generate-content?alt=sse");
}
#[test]
fn preserves_opaque_full_url_containing_models_segment() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/models/invoke",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/models/invoke?alt=sse");
}
/// Regression: a relay whose fixed path starts with `/v1/models/` is an
/// opaque route, not a Gemini-structured endpoint. The previous
/// heuristic matched any `contains("/v1/models/")` and rewrote it to
/// `/v1beta/models/{model}:generateContent`, dropping the relay's own
/// route component (`/invoke`) and sending traffic to the wrong place.
#[test]
fn preserves_opaque_full_url_with_v1_models_prefix() {
let url = resolve_gemini_native_url(
"https://relay.example/v1/models/invoke",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/v1/models/invoke?alt=sse");
}
/// Same regression, `/v1beta/models/` variant — relays may use Google's
/// path layout defensively for routing while still being opaque.
#[test]
fn preserves_opaque_full_url_with_v1beta_models_prefix() {
let url = resolve_gemini_native_url(
"https://relay.example/v1beta/models/route",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/v1beta/models/route?alt=sse");
}
/// Counter-case: a full URL that *does* carry a genuine Gemini method
/// segment (`:generateContent`) under `/v1/models/` should still be
/// recognized as structured and normalized to the requested model.
#[test]
fn normalizes_structured_v1_models_path_with_method_segment() {
let url = resolve_gemini_native_url(
"https://relay.example/v1/models/gemini-2.5-pro:generateContent",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(
url,
"https://relay.example/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
// ------------------------------------------------------------------
// Google-host whitelist tests (generic REST suffix handling)
//
// Generic REST conventions like `/v1`, `/models`, `/openai` legitimately
// appear on opaque relays. `should_normalize_gemini_full_url` only
// treats these as structured Gemini endpoints when the host itself is
// Google's Gemini or Vertex AI endpoint.
// ------------------------------------------------------------------
/// Regression: a relay whose fixed path ends with `/v1` (a ubiquitous
/// REST convention) used to be rewritten to
/// `/v1beta/models/{model}:generateContent`, dropping the relay's own
/// `/v1` endpoint.
#[test]
fn preserves_opaque_full_url_with_v1_suffix() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/v1",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/v1?alt=sse");
}
/// Companion case: bare `/models` suffix on a non-Google host is a
/// generic REST path, not a Gemini-structured endpoint.
#[test]
fn preserves_opaque_full_url_with_models_suffix() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/models",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/models?alt=sse");
}
/// Companion case: `/v1/models` — same ambiguity as `/models`, with the
/// version prefix. Must stay as-is on non-Google hosts.
#[test]
fn preserves_opaque_full_url_with_v1_models_suffix() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/v1/models",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/v1/models?alt=sse");
}
/// Companion case: a relay that exposes an `/openai` compatibility
/// surface without the deep `/openai/chat/completions` path. Must stay
/// as-is on non-Google hosts.
#[test]
fn preserves_opaque_full_url_with_openai_suffix_on_non_google_host() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/openai",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/openai?alt=sse");
}
/// Counter-case: `/v1` on the official Gemini host must still be
/// normalized to the full `/v1beta/models/...` endpoint — users who
/// paste `https://generativelanguage.googleapis.com/v1` as their base
/// URL expect the proxy to resolve the method path.
#[test]
fn normalizes_google_host_with_v1_suffix() {
let url = resolve_gemini_native_url(
"https://generativelanguage.googleapis.com/v1",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
/// Counter-case: `/models` on the official Gemini host is recognized
/// and normalized.
#[test]
fn normalizes_google_host_with_models_suffix() {
let url = resolve_gemini_native_url(
"https://generativelanguage.googleapis.com/models",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
/// Counter-case: Vertex AI regional endpoints live under
/// `*-aiplatform.googleapis.com`. Those should also be treated as
/// Google-host for the whitelist.
#[test]
fn normalizes_vertex_aiplatform_host_with_v1_suffix() {
let url = resolve_gemini_native_url(
"https://us-central1-aiplatform.googleapis.com/v1",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(
url,
"https://us-central1-aiplatform.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
/// Safety: the Google-host whitelist must do an exact/suffix match, not
/// a `contains`. A lookalike host like `aiplatform.example.com` must
/// NOT be treated as Google.
#[test]
fn preserves_non_google_aiplatform_lookalike_host() {
let url = resolve_gemini_native_url(
"https://aiplatform.example.com/v1",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://aiplatform.example.com/v1?alt=sse");
}
/// Regression: `/v1beta` by itself is Google-conventional but not
/// literally impossible on other hosts. An opaque relay fronting a
/// non-Gemini service at `https://relay.example/custom/v1beta` would
/// be silently rewritten if `/v1beta` were classified as unconditional
/// structured Gemini. Require the Google-host whitelist instead.
#[test]
fn preserves_opaque_full_url_with_bare_v1beta_suffix() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/v1beta",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/v1beta?alt=sse");
}
/// Companion case: `/v1beta/models` (no method segment) on a non-Google
/// host stays as-is too.
#[test]
fn preserves_opaque_full_url_with_v1beta_models_suffix_no_method() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/v1beta/models",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(url, "https://relay.example/custom/v1beta/models?alt=sse");
}
/// Counter-case: `/v1beta` on the official Gemini host must still
/// normalize — this is the canonical base URL shape users paste from
/// AI Studio documentation.
#[test]
fn normalizes_google_host_with_v1beta_suffix() {
let url = resolve_gemini_native_url(
"https://generativelanguage.googleapis.com/v1beta",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
true,
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
/// Regression guard: in non-full-URL mode, a versioned third-party
/// relay base must have its `/v1beta` suffix **stripped** so the
/// appended standard endpoint (`/v1beta/models/{model}:method`) does
/// not produce a doubled `/v1beta/v1beta/models/...` path. Non-full
/// mode's contract is "base URL + cc-switch appends the canonical
/// Gemini endpoint" — a user who wants a relay's custom namespace
/// (e.g. `/v1/models/...`) must use full-URL mode instead.
///
/// This test pins the intentional asymmetry with
/// `preserves_opaque_full_url_with_bare_v1beta_suffix` (full-URL
/// preserves, non-full strips) so nobody "fixes" one side into
/// breaking the other.
#[test]
fn strips_versioned_relay_base_suffix_in_non_full_url_mode() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/v1beta",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
false,
);
assert_eq!(
url,
"https://relay.example/custom/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
/// Companion case: `/v1` base suffix also stripped in non-full-URL
/// mode regardless of host.
#[test]
fn strips_v1_relay_base_suffix_in_non_full_url_mode() {
let url = resolve_gemini_native_url(
"https://relay.example/custom/v1",
"/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse",
false,
);
assert_eq!(
url,
"https://relay.example/custom/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
// ------------------------------------------------------------------
// Model ID normalization tests.
//
// Gemini SDKs and documentation commonly surface model identifiers as
// `models/gemini-2.5-pro` (resource-name form). If that value flows
// straight into our URL builder, the format string
// `/v1beta/models/{model}:generateContent` produces a doubled prefix
// `/v1beta/models/models/gemini-2.5-pro:generateContent`, which is
// rejected upstream. `normalize_gemini_model_id` is the single source
// of truth callers should run the model through first.
// ------------------------------------------------------------------
#[test]
fn normalize_model_id_strips_models_prefix() {
assert_eq!(
normalize_gemini_model_id("models/gemini-2.5-pro"),
"gemini-2.5-pro"
);
}
#[test]
fn normalize_model_id_leaves_bare_id_unchanged() {
assert_eq!(
normalize_gemini_model_id("gemini-2.5-pro"),
"gemini-2.5-pro"
);
}
#[test]
fn normalize_model_id_preserves_nested_slashes_after_prefix() {
// e.g. tuned model resource like `models/gemini-2.5-pro/tunedModels/xxx`
// — the caller asked for a specific tuned model resource, keep its
// identity intact after stripping only the single leading prefix.
assert_eq!(
normalize_gemini_model_id("models/tunedModels/my-tuned"),
"tunedModels/my-tuned"
);
}
#[test]
fn normalize_model_id_tolerates_leading_slash() {
assert_eq!(
normalize_gemini_model_id("/models/gemini-2.5-flash"),
"gemini-2.5-flash"
);
}
#[test]
fn normalize_model_id_preserves_empty_input() {
// Edge: caller has no model at all. Pass through so the URL error
// surfaces at the request layer rather than producing a misleading
// empty segment here.
assert_eq!(normalize_gemini_model_id(""), "");
}
}
+2
View File
@@ -218,9 +218,11 @@ impl RequestContext {
non_streaming_timeout,
state.status.clone(),
state.current_providers.clone(),
state.gemini_shadow.clone(),
state.failover_manager.clone(),
state.app_handle.clone(),
self.current_provider_id.clone(),
self.session_id.clone(),
first_byte_timeout,
idle_timeout,
self.rectifier_config.clone(),
+21 -2
View File
@@ -15,8 +15,9 @@ use super::{
handler_context::RequestContext,
providers::{
get_adapter, get_claude_api_format, streaming::create_anthropic_sse_stream,
streaming_gemini::create_anthropic_sse_stream_from_gemini,
streaming_responses::create_anthropic_sse_stream_from_responses, transform,
transform_responses,
transform_gemini, transform_responses,
},
response_processor::{
create_logged_passthrough_stream, process_response, read_decoded_body,
@@ -145,11 +146,13 @@ async fn handle_claude_transform(
response: super::hyper_client::ProxyResponse,
ctx: &RequestContext,
state: &ProxyState,
_original_body: &Value,
original_body: &Value,
is_stream: bool,
api_format: &str,
) -> Result<axum::response::Response, ProxyError> {
let status = response.status();
let tool_schema_hints = transform_gemini::extract_anthropic_tool_schema_hints(original_body);
let tool_schema_hints = (!tool_schema_hints.is_empty()).then_some(tool_schema_hints);
if is_stream {
// 根据 api_format 选择流式转换器
@@ -158,6 +161,14 @@ async fn handle_claude_transform(
dyn futures::Stream<Item = Result<Bytes, std::io::Error>> + Send + Unpin,
> = if api_format == "openai_responses" {
Box::new(Box::pin(create_anthropic_sse_stream_from_responses(stream)))
} else if api_format == "gemini_native" {
Box::new(Box::pin(create_anthropic_sse_stream_from_gemini(
stream,
Some(state.gemini_shadow.clone()),
Some(ctx.provider.id.clone()),
Some(ctx.session_id.clone()),
tool_schema_hints.clone(),
)))
} else {
Box::new(Box::pin(create_anthropic_sse_stream(stream)))
};
@@ -242,6 +253,14 @@ async fn handle_claude_transform(
// 根据 api_format 选择非流式转换器
let anthropic_response = if api_format == "openai_responses" {
transform_responses::responses_to_anthropic(upstream_response)
} else if api_format == "gemini_native" {
transform_gemini::gemini_to_anthropic_with_shadow_and_hints(
upstream_response,
Some(state.gemini_shadow.as_ref()),
Some(&ctx.provider.id),
Some(&ctx.session_id),
tool_schema_hints.as_ref(),
)
} else {
transform::openai_to_anthropic(upstream_response)
}
+1
View File
@@ -10,6 +10,7 @@ pub mod error;
pub mod error_mapper;
pub(crate) mod failover_switch;
mod forwarder;
pub mod gemini_url;
pub mod handler_config;
pub mod handler_context;
mod handlers;
+361 -15
View File
@@ -6,6 +6,7 @@
//! - **anthropic** (默认): Anthropic Messages API 格式,直接透传
//! - **openai_chat**: OpenAI Chat Completions 格式,需要 Anthropic ↔ OpenAI 转换
//! - **openai_responses**: OpenAI Responses API 格式,需要 Anthropic ↔ Responses 转换
//! - **gemini_native**: Google Gemini Native generateContent 格式,需要 Anthropic ↔ Gemini 转换
//!
//! ## 认证模式
//! - **Claude**: Anthropic 官方 API (x-api-key + anthropic-version)
@@ -35,6 +36,7 @@ pub fn get_claude_api_format(provider: &Provider) -> &'static str {
return match api_format {
"openai_chat" => "openai_chat",
"openai_responses" => "openai_responses",
"gemini_native" => "gemini_native",
_ => "anthropic",
};
}
@@ -49,6 +51,7 @@ pub fn get_claude_api_format(provider: &Provider) -> &'static str {
return match api_format {
"openai_chat" => "openai_chat",
"openai_responses" => "openai_responses",
"gemini_native" => "gemini_native",
_ => "anthropic",
};
}
@@ -73,13 +76,18 @@ pub fn get_claude_api_format(provider: &Provider) -> &'static str {
}
pub fn claude_api_format_needs_transform(api_format: &str) -> bool {
matches!(api_format, "openai_chat" | "openai_responses")
matches!(
api_format,
"openai_chat" | "openai_responses" | "gemini_native"
)
}
pub fn transform_claude_request_for_api_format(
body: serde_json::Value,
provider: &Provider,
api_format: &str,
session_id: Option<&str>,
shadow_store: Option<&super::gemini_shadow::GeminiShadowStore>,
) -> Result<serde_json::Value, ProxyError> {
// Copilot 场景:优先从 metadata.user_id 提取 session ID 作为 cache key
// 格式: "uuid_sessionId" → 提取 "_" 后面的部分作为 session 标识
@@ -138,7 +146,24 @@ pub fn transform_claude_request_for_api_format(
is_codex_oauth,
)
}
"openai_chat" => super::transform::anthropic_to_openai(body),
"openai_chat" => {
let mut result = super::transform::anthropic_to_openai(body)?;
// Inject prompt_cache_key only if explicitly configured in meta
if let Some(key) = provider
.meta
.as_ref()
.and_then(|m| m.prompt_cache_key.as_deref())
{
result["prompt_cache_key"] = serde_json::json!(key);
}
Ok(result)
}
"gemini_native" => super::transform_gemini::anthropic_to_gemini_with_shadow(
body,
shadow_store,
Some(&provider.id),
session_id,
),
_ => Ok(body),
}
}
@@ -160,6 +185,16 @@ impl ClaudeAdapter {
/// - ClaudeAuth: auth_mode 为 bearer_only
/// - Claude: 默认 Anthropic 官方
pub fn provider_type(&self, provider: &Provider) -> ProviderType {
// 检测 Gemini Native 格式
if self.get_api_format(provider) == "gemini_native" {
return match self.extract_key(provider) {
Some(key) if key.starts_with("ya29.") || key.starts_with('{') => {
ProviderType::GeminiCli
}
_ => ProviderType::Gemini,
};
}
// 检测 Codex OAuth (ChatGPT Plus/Pro)
if self.is_codex_oauth(provider) {
return ProviderType::CodexOAuth;
@@ -262,6 +297,7 @@ impl ClaudeAdapter {
if let Some(key) = env
.get("ANTHROPIC_AUTH_TOKEN")
.and_then(|v| v.as_str())
.map(str::trim)
.filter(|s| !s.is_empty())
{
log::debug!("[Claude] 使用 ANTHROPIC_AUTH_TOKEN");
@@ -270,6 +306,7 @@ impl ClaudeAdapter {
if let Some(key) = env
.get("ANTHROPIC_API_KEY")
.and_then(|v| v.as_str())
.map(str::trim)
.filter(|s| !s.is_empty())
{
log::debug!("[Claude] 使用 ANTHROPIC_API_KEY");
@@ -279,6 +316,7 @@ impl ClaudeAdapter {
if let Some(key) = env
.get("OPENROUTER_API_KEY")
.and_then(|v| v.as_str())
.map(str::trim)
.filter(|s| !s.is_empty())
{
log::debug!("[Claude] 使用 OPENROUTER_API_KEY");
@@ -288,6 +326,7 @@ impl ClaudeAdapter {
if let Some(key) = env
.get("OPENAI_API_KEY")
.and_then(|v| v.as_str())
.map(str::trim)
.filter(|s| !s.is_empty())
{
log::debug!("[Claude] 使用 OPENAI_API_KEY");
@@ -301,6 +340,7 @@ impl ClaudeAdapter {
.get("apiKey")
.or_else(|| provider.settings_config.get("api_key"))
.and_then(|v| v.as_str())
.map(str::trim)
.filter(|s| !s.is_empty())
{
log::debug!("[Claude] 使用 apiKey/api_key");
@@ -388,14 +428,43 @@ impl ProviderAdapter for ClaudeAdapter {
));
}
let strategy = match provider_type {
ProviderType::OpenRouter => AuthStrategy::Bearer,
ProviderType::ClaudeAuth => AuthStrategy::ClaudeAuth,
_ => AuthStrategy::Anthropic,
};
let key = self.extract_key(provider)?;
self.extract_key(provider)
.map(|key| AuthInfo::new(key, strategy))
match provider_type {
ProviderType::GeminiCli => {
// Parse stored OAuth JSON and only attach access_token when
// it's actually usable. `parse_oauth_credentials` accepts
// refresh-token-only JSON (which is legitimate before the
// first refresh) and also surfaces `{"access_token": "", ...}`
// for expired credentials. In both cases we would otherwise
// send `Authorization: Bearer ` to upstream and get a 401.
//
// CC Switch does not currently exchange the refresh_token for
// a fresh access_token. Until that path exists, degrade to
// plain GoogleOAuth strategy (which still sends the raw key
// as a fallback) and log loudly so users know to refresh
// their `~/.gemini/oauth_creds.json`.
match super::gemini::GeminiAdapter::new().parse_oauth_credentials(&key) {
Some(creds) if !creds.access_token.is_empty() => {
Some(AuthInfo::with_access_token(key, creds.access_token))
}
Some(_) => {
log::warn!(
"[Gemini OAuth] access_token missing or empty for provider `{}`; \
bearer auth will likely fail with 401. Refresh \
~/.gemini/oauth_creds.json via the gemini CLI to obtain a new token.",
provider.id
);
Some(AuthInfo::new(key, AuthStrategy::GoogleOAuth))
}
None => Some(AuthInfo::new(key, AuthStrategy::GoogleOAuth)),
}
}
ProviderType::Gemini => Some(AuthInfo::new(key, AuthStrategy::Google)),
ProviderType::OpenRouter => Some(AuthInfo::new(key, AuthStrategy::Bearer)),
ProviderType::ClaudeAuth => Some(AuthInfo::new(key, AuthStrategy::ClaudeAuth)),
_ => Some(AuthInfo::new(key, AuthStrategy::Anthropic)),
}
}
fn build_url(&self, base_url: &str, endpoint: &str) -> String {
@@ -437,6 +506,23 @@ impl ProviderAdapter for ClaudeAdapter {
HeaderValue::from_str(&bearer).unwrap(),
)]
}
AuthStrategy::Google => vec![(
HeaderName::from_static("x-goog-api-key"),
HeaderValue::from_str(&auth.api_key).unwrap(),
)],
AuthStrategy::GoogleOAuth => {
let token = auth.access_token.as_ref().unwrap_or(&auth.api_key);
vec![
(
HeaderName::from_static("authorization"),
HeaderValue::from_str(&format!("Bearer {token}")).unwrap(),
),
(
HeaderName::from_static("x-goog-api-client"),
HeaderValue::from_static("GeminiCLI/1.0"),
),
]
}
AuthStrategy::CodexOAuth => {
// 注意:bearer token 由 forwarder 动态注入到 auth.api_key
// ChatGPT-Account-Id 由 forwarder 注入额外 header
@@ -507,7 +593,6 @@ impl ProviderAdapter for ClaudeAdapter {
),
]
}
_ => vec![],
}
}
@@ -528,7 +613,7 @@ impl ProviderAdapter for ClaudeAdapter {
// - "openai_responses": 需要 Anthropic ↔ OpenAI Responses API 格式转换
matches!(
self.get_api_format(provider),
"openai_chat" | "openai_responses"
"openai_chat" | "openai_responses" | "gemini_native"
)
}
@@ -537,7 +622,13 @@ impl ProviderAdapter for ClaudeAdapter {
body: serde_json::Value,
provider: &Provider,
) -> Result<serde_json::Value, ProxyError> {
transform_claude_request_for_api_format(body, provider, self.get_api_format(provider))
transform_claude_request_for_api_format(
body,
provider,
self.get_api_format(provider),
None,
None,
)
}
fn transform_response(&self, body: serde_json::Value) -> Result<serde_json::Value, ProxyError> {
@@ -546,7 +637,9 @@ impl ProviderAdapter for ClaudeAdapter {
// config, so we can't check api_format here. Instead we rely on the fact that
// Responses API always returns "output" while Chat Completions returns "choices".
// This is safe because the two formats are structurally disjoint.
if body.get("output").is_some() {
if body.get("candidates").is_some() || body.get("promptFeedback").is_some() {
super::transform_gemini::gemini_to_anthropic(body)
} else if body.get("output").is_some() {
super::transform_responses::responses_to_anthropic(body)
} else {
super::transform::openai_to_anthropic(body)
@@ -684,6 +777,146 @@ mod tests {
assert_eq!(auth.strategy, AuthStrategy::ClaudeAuth);
}
/// Regression: a Gemini OAuth credential JSON that carries only a
/// refresh_token (no active access_token) must not be surfaced as an
/// `AuthInfo` whose bearer would be empty. Without the guard, downstream
/// header injection produces `Authorization: Bearer ` and a deterministic
/// 401 from upstream.
#[test]
fn test_extract_auth_gemini_cli_refresh_only_json_does_not_expose_empty_bearer() {
let adapter = ClaudeAdapter::new();
let refresh_only_json =
r#"{"refresh_token":"rt-abc","client_id":"cid","client_secret":"cs"}"#;
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://generativelanguage.googleapis.com",
"ANTHROPIC_API_KEY": refresh_only_json
}
}),
ProviderMeta {
api_format: Some("gemini_native".to_string()),
..Default::default()
},
);
let auth = adapter.extract_auth(&provider).unwrap();
// access_token must not be surfaced as `Some("")` — the OAuth header
// builder uses `access_token.as_ref().unwrap_or(&api_key)`, so a
// `Some("")` would win over the raw key and emit `Bearer `.
assert!(
auth.access_token.as_deref().is_none_or(|t| !t.is_empty()),
"empty access_token leaked into AuthInfo"
);
assert_eq!(auth.strategy, AuthStrategy::GoogleOAuth);
}
/// Companion case: a JSON credential with an empty-string `access_token`
/// field (the shape an expired credential can take after partial writes)
/// must degrade the same way.
#[test]
fn test_extract_auth_gemini_cli_empty_access_token_degrades_to_raw_key() {
let adapter = ClaudeAdapter::new();
let expired_json = r#"{"access_token":"","refresh_token":"rt-abc","client_id":"cid","client_secret":"cs"}"#;
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://generativelanguage.googleapis.com",
"ANTHROPIC_API_KEY": expired_json
}
}),
ProviderMeta {
api_format: Some("gemini_native".to_string()),
..Default::default()
},
);
let auth = adapter.extract_auth(&provider).unwrap();
assert!(
auth.access_token.as_deref().is_none_or(|t| !t.is_empty()),
"empty access_token leaked into AuthInfo"
);
assert_eq!(auth.strategy, AuthStrategy::GoogleOAuth);
}
/// Counter-case: a well-formed JSON credential with a non-empty
/// access_token must still flow through the OAuth path unchanged.
#[test]
fn test_extract_auth_gemini_cli_valid_json_keeps_access_token() {
let adapter = ClaudeAdapter::new();
let valid_json = r#"{"access_token":"ya29.valid","refresh_token":"rt"}"#;
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://generativelanguage.googleapis.com",
"ANTHROPIC_API_KEY": valid_json
}
}),
ProviderMeta {
api_format: Some("gemini_native".to_string()),
..Default::default()
},
);
let auth = adapter.extract_auth(&provider).unwrap();
assert_eq!(auth.access_token.as_deref(), Some("ya29.valid"));
assert_eq!(auth.strategy, AuthStrategy::GoogleOAuth);
}
/// 回归:从 oauth_creds.json 复制时常带前导换行/空格。未 trim 时
/// `starts_with('{')` 会落空,导致误分类为 `ProviderType::Gemini`,再
/// 以 raw JSON 当 `x-goog-api-key` 发出去触发 401。trim 应在 provider
/// 类型判定和 OAuth 解析前统一生效。
#[test]
fn test_extract_auth_gemini_cli_json_with_leading_whitespace_classifies_correctly() {
let adapter = ClaudeAdapter::new();
let valid_json = r#"{"access_token":"ya29.valid","refresh_token":"rt"}"#;
let key_with_whitespace = format!("\n {valid_json}\n");
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://generativelanguage.googleapis.com",
"ANTHROPIC_API_KEY": key_with_whitespace
}
}),
ProviderMeta {
api_format: Some("gemini_native".to_string()),
..Default::default()
},
);
assert_eq!(adapter.provider_type(&provider), ProviderType::GeminiCli);
let auth = adapter.extract_auth(&provider).unwrap();
assert_eq!(auth.access_token.as_deref(), Some("ya29.valid"));
assert_eq!(auth.strategy, AuthStrategy::GoogleOAuth);
}
/// 回归:裸 `ya29.` access_token 若带前导换行,也应被 trim 后识别为
/// Gemini CLI OAuth,避免前导空白把 `starts_with("ya29.")` 检查顶穿。
#[test]
fn test_extract_auth_gemini_cli_access_token_with_leading_newline_classifies_correctly() {
let adapter = ClaudeAdapter::new();
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://generativelanguage.googleapis.com",
"ANTHROPIC_API_KEY": "\nya29.raw-token-value\n"
}
}),
ProviderMeta {
api_format: Some("gemini_native".to_string()),
..Default::default()
},
);
assert_eq!(adapter.provider_type(&provider), ProviderType::GeminiCli);
let auth = adapter.extract_auth(&provider).unwrap();
assert_eq!(auth.access_token.as_deref(), Some("ya29.raw-token-value"));
assert_eq!(auth.strategy, AuthStrategy::GoogleOAuth);
}
#[test]
fn test_provider_type_detection() {
let adapter = ClaudeAdapter::new();
@@ -850,6 +1083,24 @@ mod tests {
);
assert!(adapter.needs_transform(&openai_responses_provider));
let gemini_native_provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://generativelanguage.googleapis.com",
"ANTHROPIC_API_KEY": "test-key"
}
}),
ProviderMeta {
api_format: Some("gemini_native".to_string()),
..Default::default()
},
);
assert!(adapter.needs_transform(&gemini_native_provider));
assert_eq!(
adapter.provider_type(&gemini_native_provider),
ProviderType::Gemini
);
// meta takes precedence over legacy settings_config fields
let meta_precedence_over_settings = create_provider_with_meta(
json!({
@@ -957,11 +1208,106 @@ mod tests {
"max_tokens": 128
});
let transformed =
transform_claude_request_for_api_format(body, &provider, "openai_responses").unwrap();
let transformed = transform_claude_request_for_api_format(
body,
&provider,
"openai_responses",
None,
None,
)
.unwrap();
assert_eq!(transformed["model"], "gpt-5.4");
assert!(transformed.get("input").is_some());
assert!(transformed.get("max_output_tokens").is_some());
}
#[test]
fn test_transform_claude_request_for_api_format_gemini_native() {
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://generativelanguage.googleapis.com",
"ANTHROPIC_API_KEY": "test-key"
}
}),
ProviderMeta {
api_format: Some("gemini_native".to_string()),
..Default::default()
},
);
let body = json!({
"model": "gemini-2.5-pro",
"system": "You are helpful.",
"messages": [{ "role": "user", "content": "hello" }],
"max_tokens": 64
});
let transformed =
transform_claude_request_for_api_format(body, &provider, "gemini_native", None, None)
.unwrap();
assert!(transformed.get("contents").is_some());
assert_eq!(
transformed["systemInstruction"]["parts"][0]["text"],
"You are helpful."
);
assert_eq!(transformed["generationConfig"]["maxOutputTokens"], 64);
}
#[test]
fn test_transform_claude_request_for_api_format_openai_chat_skips_prompt_cache_key_by_default()
{
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://api.example.com",
"ANTHROPIC_API_KEY": "test-key"
}
}),
ProviderMeta {
api_format: Some("openai_chat".to_string()),
..Default::default()
},
);
let body = json!({
"model": "gpt-5.4",
"messages": [{ "role": "user", "content": "hello" }],
"max_tokens": 64
});
let transformed =
transform_claude_request_for_api_format(body, &provider, "openai_chat", None, None)
.unwrap();
assert!(transformed.get("prompt_cache_key").is_none());
}
#[test]
fn test_transform_claude_request_for_api_format_openai_chat_keeps_explicit_prompt_cache_key() {
let provider = create_provider_with_meta(
json!({
"env": {
"ANTHROPIC_BASE_URL": "https://api.example.com",
"ANTHROPIC_API_KEY": "test-key"
}
}),
ProviderMeta {
api_format: Some("openai_chat".to_string()),
prompt_cache_key: Some("claude-cache-route".to_string()),
..Default::default()
},
);
let body = json!({
"model": "gpt-5.4",
"messages": [{ "role": "user", "content": "hello" }],
"max_tokens": 64
});
let transformed =
transform_claude_request_for_api_format(body, &provider, "openai_chat", None, None)
.unwrap();
assert_eq!(transformed["prompt_cache_key"], "claude-cache-route");
}
}
+13 -1
View File
@@ -70,6 +70,11 @@ impl GeminiAdapter {
/// 解析 OAuth 凭证
pub fn parse_oauth_credentials(&self, key: &str) -> Option<OAuthCredentials> {
// 防御性 trim:前端在 input 事件中会 trim,但 JSON 编辑器 / deeplink
// 导入 / live 回填等路径会绕过。带前导换行的 oauth_creds.json 粘贴
// 是常见场景,此处统一兜底。
let key = key.trim();
// 直接是 access_token
if key.starts_with("ya29.") {
return Some(OAuthCredentials {
@@ -120,7 +125,12 @@ impl GeminiAdapter {
fn extract_key_raw(&self, provider: &Provider) -> Option<String> {
if let Some(env) = provider.settings_config.get("env") {
// 使用 GEMINI_API_KEY
if let Some(key) = env.get("GEMINI_API_KEY").and_then(|v| v.as_str()) {
if let Some(key) = env
.get("GEMINI_API_KEY")
.and_then(|v| v.as_str())
.map(str::trim)
.filter(|s| !s.is_empty())
{
return Some(key.to_string());
}
}
@@ -131,6 +141,8 @@ impl GeminiAdapter {
.get("apiKey")
.or_else(|| provider.settings_config.get("api_key"))
.and_then(|v| v.as_str())
.map(str::trim)
.filter(|s| !s.is_empty())
{
return Some(key.to_string());
}
@@ -0,0 +1,338 @@
//! Gemini tool schema helpers.
//!
//! Gemini `FunctionDeclaration` supports two schema channels:
//! - `parameters`: a restricted `Schema` subset
//! - `parametersJsonSchema`: richer JSON Schema via arbitrary JSON `Value`
//!
//! Anthropic tool schemas are closer to JSON Schema, so we choose the richer
//! channel when unsupported `Schema` fields are present.
use serde_json::{json, Map, Value};
#[derive(Debug, Clone, PartialEq)]
pub enum GeminiFunctionParameters {
Schema(Value),
JsonSchema(Value),
}
pub fn build_gemini_function_parameters(input_schema: Value) -> GeminiFunctionParameters {
let schema = ensure_object_schema(normalize_json_schema(input_schema));
if requires_parameters_json_schema(&schema) {
GeminiFunctionParameters::JsonSchema(schema)
} else {
GeminiFunctionParameters::Schema(to_gemini_schema(schema))
}
}
/// Vertex AI rejects FunctionDeclarations whose `parameters` schema lacks an
/// explicit `type: "object"`, returning:
///
/// > functionDeclaration parameters schema should be of type OBJECT.
///
/// Anthropic tools sometimes arrive with empty or type-less `input_schema`
/// (e.g. no-argument tools like Claude Code's `TodoRead`). Normalize those to
/// `{type: "object", properties: {}}` so the Gemini upstream accepts them.
///
/// References: google-gemini/generative-ai-python#423, BerriAI/litellm#5055.
fn ensure_object_schema(schema: Value) -> Value {
match schema {
Value::Object(mut obj) => {
obj.entry("type".to_string())
.or_insert_with(|| json!("object"));
if obj.get("type").and_then(|v| v.as_str()) == Some("object") {
obj.entry("properties".to_string())
.or_insert_with(|| json!({}));
}
Value::Object(obj)
}
other => other,
}
}
fn normalize_json_schema(schema: Value) -> Value {
match schema {
Value::Object(mut obj) => {
obj.remove("$schema");
obj.remove("$id");
if let Some(properties) = obj
.get_mut("properties")
.and_then(|value| value.as_object_mut())
{
for value in properties.values_mut() {
*value = normalize_json_schema(value.clone());
}
}
if let Some(items) = obj.get_mut("items") {
*items = normalize_json_schema(items.clone());
}
for key in ["anyOf", "oneOf", "allOf", "prefixItems"] {
if let Some(values) = obj.get_mut(key).and_then(|value| value.as_array_mut()) {
for value in values.iter_mut() {
*value = normalize_json_schema(value.clone());
}
}
}
for key in ["not", "if", "then", "else", "additionalProperties"] {
if let Some(value) = obj.get_mut(key) {
*value = normalize_json_schema(value.clone());
}
}
Value::Object(obj)
}
Value::Array(values) => {
Value::Array(values.into_iter().map(normalize_json_schema).collect())
}
other => other,
}
}
fn requires_parameters_json_schema(schema: &Value) -> bool {
match schema {
Value::Object(obj) => object_requires_parameters_json_schema(obj),
Value::Array(values) => values.iter().any(requires_parameters_json_schema),
_ => false,
}
}
fn object_requires_parameters_json_schema(obj: &Map<String, Value>) -> bool {
for (key, value) in obj {
match key.as_str() {
"type" => {
if value.is_array() {
return true;
}
}
"format" | "title" | "description" | "nullable" | "enum" | "maxItems" | "minItems"
| "required" | "minProperties" | "maxProperties" | "minLength" | "maxLength"
| "pattern" | "example" | "propertyOrdering" | "default" | "minimum" | "maximum" => {}
"properties" => {
let Some(properties) = value.as_object() else {
return true;
};
if properties.values().any(requires_parameters_json_schema) {
return true;
}
}
"items" => {
if !value.is_object() || requires_parameters_json_schema(value) {
return true;
}
}
"anyOf" => {
let Some(values) = value.as_array() else {
return true;
};
if values.iter().any(requires_parameters_json_schema) {
return true;
}
}
// JSON Schema keywords that Gemini `parameters` does not accept.
"$ref"
| "$defs"
| "definitions"
| "additionalProperties"
| "unevaluatedProperties"
| "patternProperties"
| "oneOf"
| "allOf"
| "const"
| "not"
| "if"
| "then"
| "else"
| "dependentRequired"
| "dependentSchemas"
| "contains"
| "minContains"
| "maxContains"
| "prefixItems"
| "exclusiveMinimum"
| "exclusiveMaximum"
| "multipleOf"
| "examples" => return true,
// Be conservative for unknown keywords.
_ => return true,
}
}
false
}
fn to_gemini_schema(schema: Value) -> Value {
match schema {
Value::Object(obj) => {
let mut result = Map::new();
for (key, value) in obj {
match key.as_str() {
"type" | "format" | "title" | "description" | "nullable" | "enum"
| "maxItems" | "minItems" | "required" | "minProperties" | "maxProperties"
| "minLength" | "maxLength" | "pattern" | "example" | "propertyOrdering"
| "default" | "minimum" | "maximum" => {
result.insert(key, value);
}
"properties" => {
if let Some(properties) = value.as_object() {
let converted = properties
.iter()
.map(|(name, property_schema)| {
(name.clone(), to_gemini_schema(property_schema.clone()))
})
.collect();
result.insert("properties".to_string(), Value::Object(converted));
}
}
"items" if value.is_object() => {
result.insert("items".to_string(), to_gemini_schema(value));
}
"anyOf" => {
if let Some(values) = value.as_array() {
result.insert(
"anyOf".to_string(),
Value::Array(
values
.iter()
.map(|value| to_gemini_schema(value.clone()))
.collect(),
),
);
}
}
_ => {}
}
}
Value::Object(result)
}
other => other,
}
}
pub fn build_gemini_function_declaration(
name: &str,
description: Option<&str>,
input_schema: Value,
) -> Value {
let mut declaration = Map::new();
declaration.insert("name".to_string(), json!(name));
declaration.insert("description".to_string(), json!(description.unwrap_or("")));
match build_gemini_function_parameters(input_schema) {
GeminiFunctionParameters::Schema(schema) => {
declaration.insert("parameters".to_string(), schema);
}
GeminiFunctionParameters::JsonSchema(schema) => {
declaration.insert("parametersJsonSchema".to_string(), schema);
}
}
Value::Object(declaration)
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn uses_schema_for_simple_openapi_subset() {
let schema = json!({
"type": "object",
"properties": {
"city": { "type": "string", "description": "Target city" }
},
"required": ["city"]
});
let result = build_gemini_function_declaration("weather", Some("Weather lookup"), schema);
assert!(result.get("parameters").is_some());
assert!(result.get("parametersJsonSchema").is_none());
assert_eq!(result["parameters"]["properties"]["city"]["type"], "string");
}
#[test]
fn uses_parameters_json_schema_for_additional_properties() {
let schema = json!({
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": {
"city": { "type": "string" }
},
"required": ["city"],
"additionalProperties": false
});
let result = build_gemini_function_declaration("weather", Some("Weather lookup"), schema);
assert!(result.get("parameters").is_none());
assert!(result.get("parametersJsonSchema").is_some());
assert!(result["parametersJsonSchema"].get("$schema").is_none());
assert_eq!(
result["parametersJsonSchema"]["additionalProperties"],
false
);
}
#[test]
fn uses_parameters_json_schema_for_one_of() {
let schema = json!({
"type": "object",
"properties": {
"target": {
"oneOf": [
{ "type": "string" },
{ "type": "integer" }
]
}
}
});
let result = build_gemini_function_declaration("search", Some("Search"), schema);
assert!(result.get("parameters").is_none());
assert!(result.get("parametersJsonSchema").is_some());
}
/// Regression for P2 (Vertex AI rejecting empty schemas): zero-argument
/// Anthropic tools (no `input_schema`) must produce `parameters` with an
/// explicit `type: "object"` and an empty `properties` map so the Gemini
/// upstream does not return `schema should be of type OBJECT`.
#[test]
fn empty_input_schema_produces_explicit_object_type() {
let result = build_gemini_function_declaration("ping", Some("no-arg"), json!({}));
assert_eq!(result["parameters"]["type"], "object");
assert!(result["parameters"]["properties"].is_object());
}
/// A schema that carries descriptive fields but no `type` is still a
/// zero-arg object for Gemini purposes — promote it explicitly.
#[test]
fn input_schema_missing_type_is_promoted_to_object() {
let result = build_gemini_function_declaration(
"noop",
None,
json!({ "description": "does nothing" }),
);
assert_eq!(result["parameters"]["type"], "object");
assert!(result["parameters"]["properties"].is_object());
}
/// Defensive: an atomic (non-object) schema is left untouched, because
/// forcing `type: "object"` here would corrupt primitive parameter types
/// that happen to flow through this path.
#[test]
fn non_object_schema_is_not_mutated() {
let result = build_gemini_function_declaration("bare", None, json!({ "type": "string" }));
assert_eq!(result["parameters"]["type"], "string");
assert!(result["parameters"].get("properties").is_none());
}
}
@@ -0,0 +1,389 @@
//! Gemini Native shadow state
//!
//! Keeps provider/session-scoped assistant content snapshots and tool call metadata
//! so Gemini thought signatures and tool turns can be replayed without bloating
//! the main proxy files.
use serde_json::Value;
use std::collections::{HashMap, VecDeque};
use std::sync::RwLock;
/// Composite key for a Gemini shadow session.
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub struct GeminiShadowKey {
pub provider_id: String,
pub session_id: String,
}
impl GeminiShadowKey {
pub fn new(provider_id: impl Into<String>, session_id: impl Into<String>) -> Self {
Self {
provider_id: provider_id.into(),
session_id: session_id.into(),
}
}
}
/// Gemini function call metadata captured from an assistant turn.
#[derive(Debug, Clone, PartialEq)]
pub struct GeminiToolCallMeta {
pub id: Option<String>,
pub name: String,
pub args: Value,
pub thought_signature: Option<String>,
}
impl GeminiToolCallMeta {
pub fn new(
id: Option<impl Into<String>>,
name: impl Into<String>,
args: Value,
thought_signature: Option<impl Into<String>>,
) -> Self {
Self {
id: id.map(Into::into),
name: name.into(),
args,
thought_signature: thought_signature.map(Into::into),
}
}
}
/// Stored assistant turn snapshot.
#[derive(Debug, Clone, PartialEq)]
pub struct GeminiAssistantTurn {
pub assistant_content: Value,
pub tool_calls: Vec<GeminiToolCallMeta>,
}
impl GeminiAssistantTurn {
pub fn new(assistant_content: Value, tool_calls: Vec<GeminiToolCallMeta>) -> Self {
Self {
assistant_content,
tool_calls,
}
}
}
/// Session snapshot returned by read APIs.
#[derive(Debug, Clone, PartialEq)]
pub struct GeminiShadowSessionSnapshot {
pub provider_id: String,
pub session_id: String,
pub turns: Vec<GeminiAssistantTurn>,
}
#[derive(Debug, Clone)]
struct GeminiShadowSession {
turns: VecDeque<GeminiAssistantTurn>,
}
impl GeminiShadowSession {
fn new() -> Self {
Self {
turns: VecDeque::new(),
}
}
}
#[derive(Debug, Clone)]
struct GeminiShadowInner {
sessions: HashMap<GeminiShadowKey, GeminiShadowSession>,
session_order: VecDeque<GeminiShadowKey>,
}
impl GeminiShadowInner {
fn new() -> Self {
Self {
sessions: HashMap::new(),
session_order: VecDeque::new(),
}
}
}
/// Thread-safe shadow store for Gemini Native replay state.
///
/// The store is intentionally small and explicit:
/// - sessions are keyed by `(provider_id, session_id)`
/// - each session keeps only a bounded number of recent assistant turns
/// - the oldest session is evicted first when the store is full
#[derive(Debug)]
pub struct GeminiShadowStore {
max_sessions: usize,
max_turns_per_session: usize,
inner: RwLock<GeminiShadowInner>,
}
impl Default for GeminiShadowStore {
fn default() -> Self {
Self::with_limits(200, 64)
}
}
impl GeminiShadowStore {
#[allow(dead_code)]
pub fn new() -> Self {
Self::default()
}
pub fn with_limits(max_sessions: usize, max_turns_per_session: usize) -> Self {
Self {
max_sessions: max_sessions.max(1),
max_turns_per_session: max_turns_per_session.max(1),
inner: RwLock::new(GeminiShadowInner::new()),
}
}
/// Record a Gemini assistant turn for later replay.
pub fn record_assistant_turn(
&self,
provider_id: impl Into<String>,
session_id: impl Into<String>,
assistant_content: Value,
tool_calls: Vec<GeminiToolCallMeta>,
) -> GeminiShadowSessionSnapshot {
let key = GeminiShadowKey::new(provider_id, session_id);
let turn = GeminiAssistantTurn::new(assistant_content, tool_calls);
let mut inner = self.inner.write().expect("gemini shadow lock poisoned");
Self::touch_session_order(&mut inner.session_order, &key);
let snapshot = {
let session = inner
.sessions
.entry(key.clone())
.or_insert_with(GeminiShadowSession::new);
session.turns.push_back(turn);
while session.turns.len() > self.max_turns_per_session {
session.turns.pop_front();
}
Self::snapshot_session(&key, session)
};
Self::prune_sessions(&mut inner, self.max_sessions);
snapshot
}
/// Get the latest assistant content for a provider/session pair.
#[allow(dead_code)]
pub fn latest_assistant_content(&self, provider_id: &str, session_id: &str) -> Option<Value> {
self.get_session(provider_id, session_id)
.and_then(|snapshot| {
snapshot
.turns
.last()
.map(|turn| turn.assistant_content.clone())
})
}
/// Get the latest tool calls for a provider/session pair.
#[allow(dead_code)]
pub fn latest_tool_calls(
&self,
provider_id: &str,
session_id: &str,
) -> Option<Vec<GeminiToolCallMeta>> {
self.get_session(provider_id, session_id)
.and_then(|snapshot| snapshot.turns.last().map(|turn| turn.tool_calls.clone()))
}
/// Read a full session snapshot.
pub fn get_session(
&self,
provider_id: &str,
session_id: &str,
) -> Option<GeminiShadowSessionSnapshot> {
let key = GeminiShadowKey::new(provider_id, session_id);
let mut inner = self.inner.write().expect("gemini shadow lock poisoned");
let snapshot = inner
.sessions
.get(&key)
.map(|session| Self::snapshot_session(&key, session));
if snapshot.is_some() {
Self::touch_session_order(&mut inner.session_order, &key);
}
snapshot
}
/// Remove a single session from the store.
#[allow(dead_code)]
pub fn clear_session(&self, provider_id: &str, session_id: &str) -> bool {
let key = GeminiShadowKey::new(provider_id, session_id);
let mut inner = self.inner.write().expect("gemini shadow lock poisoned");
let removed = inner.sessions.remove(&key).is_some();
if removed {
Self::remove_key_from_order(&mut inner.session_order, &key);
}
removed
}
/// Remove all sessions for a provider.
#[allow(dead_code)]
pub fn clear_provider(&self, provider_id: &str) -> usize {
let mut inner = self.inner.write().expect("gemini shadow lock poisoned");
let keys: Vec<_> = inner
.sessions
.keys()
.filter(|key| key.provider_id == provider_id)
.cloned()
.collect();
for key in &keys {
inner.sessions.remove(key);
Self::remove_key_from_order(&mut inner.session_order, key);
}
keys.len()
}
/// Number of tracked sessions.
#[allow(dead_code)]
pub fn session_count(&self) -> usize {
self.inner
.read()
.expect("gemini shadow lock poisoned")
.sessions
.len()
}
fn snapshot_session(
key: &GeminiShadowKey,
session: &GeminiShadowSession,
) -> GeminiShadowSessionSnapshot {
GeminiShadowSessionSnapshot {
provider_id: key.provider_id.clone(),
session_id: key.session_id.clone(),
turns: session.turns.iter().cloned().collect(),
}
}
fn touch_session_order(order: &mut VecDeque<GeminiShadowKey>, key: &GeminiShadowKey) {
if let Some(pos) = order.iter().position(|existing| existing == key) {
order.remove(pos);
}
order.push_back(key.clone());
}
#[allow(dead_code)]
fn remove_key_from_order(order: &mut VecDeque<GeminiShadowKey>, key: &GeminiShadowKey) {
if let Some(pos) = order.iter().position(|existing| existing == key) {
order.remove(pos);
}
}
fn prune_sessions(inner: &mut GeminiShadowInner, max_sessions: usize) {
while inner.sessions.len() > max_sessions {
let Some(evicted_key) = inner.session_order.pop_front() else {
break;
};
inner.sessions.remove(&evicted_key);
}
}
}
#[cfg(test)]
mod tests {
use super::*;
use serde_json::json;
#[test]
fn record_and_read_latest_turn() {
let store = GeminiShadowStore::with_limits(8, 4);
let snapshot = store.record_assistant_turn(
"provider-a",
"session-1",
json!({"parts": [{"text": "hello", "thoughtSignature": "sig-1"}]}),
vec![GeminiToolCallMeta::new(
Some("call-1"),
"get_weather",
json!({"location": "Tokyo"}),
Some("sig-1"),
)],
);
assert_eq!(snapshot.provider_id, "provider-a");
assert_eq!(snapshot.session_id, "session-1");
assert_eq!(snapshot.turns.len(), 1);
let content = store
.latest_assistant_content("provider-a", "session-1")
.expect("content");
assert_eq!(content["parts"][0]["text"], "hello");
assert_eq!(content["parts"][0]["thoughtSignature"], "sig-1");
let tool_calls = store
.latest_tool_calls("provider-a", "session-1")
.expect("tool calls");
assert_eq!(tool_calls.len(), 1);
assert_eq!(tool_calls[0].id.as_deref(), Some("call-1"));
assert_eq!(tool_calls[0].name, "get_weather");
assert_eq!(tool_calls[0].args["location"], "Tokyo");
assert_eq!(tool_calls[0].thought_signature.as_deref(), Some("sig-1"));
}
#[test]
fn sessions_are_isolated_by_provider_and_session_id() {
let store = GeminiShadowStore::with_limits(8, 4);
store.record_assistant_turn("provider-a", "session-1", json!({"text": "a"}), vec![]);
store.record_assistant_turn("provider-b", "session-1", json!({"text": "b"}), vec![]);
store.record_assistant_turn("provider-a", "session-2", json!({"text": "c"}), vec![]);
assert_eq!(store.session_count(), 3);
assert_eq!(
store.latest_assistant_content("provider-a", "session-1"),
Some(json!({"text": "a"}))
);
assert_eq!(
store.latest_assistant_content("provider-b", "session-1"),
Some(json!({"text": "b"}))
);
assert_eq!(
store.latest_assistant_content("provider-a", "session-2"),
Some(json!({"text": "c"}))
);
}
#[test]
fn retains_only_latest_turns_per_session() {
let store = GeminiShadowStore::with_limits(8, 2);
store.record_assistant_turn("provider-a", "session-1", json!({"idx": 1}), vec![]);
store.record_assistant_turn("provider-a", "session-1", json!({"idx": 2}), vec![]);
store.record_assistant_turn("provider-a", "session-1", json!({"idx": 3}), vec![]);
let snapshot = store
.get_session("provider-a", "session-1")
.expect("snapshot");
assert_eq!(snapshot.turns.len(), 2);
assert_eq!(snapshot.turns[0].assistant_content, json!({"idx": 2}));
assert_eq!(snapshot.turns[1].assistant_content, json!({"idx": 3}));
}
#[test]
fn evicts_oldest_session_when_capacity_is_exceeded() {
let store = GeminiShadowStore::with_limits(2, 2);
store.record_assistant_turn("provider-a", "session-1", json!({"idx": 1}), vec![]);
store.record_assistant_turn("provider-a", "session-2", json!({"idx": 2}), vec![]);
store.record_assistant_turn("provider-a", "session-3", json!({"idx": 3}), vec![]);
assert!(store.get_session("provider-a", "session-1").is_none());
assert!(store.get_session("provider-a", "session-2").is_some());
assert!(store.get_session("provider-a", "session-3").is_some());
}
#[test]
fn clear_session_and_provider_work() {
let store = GeminiShadowStore::with_limits(8, 4);
store.record_assistant_turn("provider-a", "session-1", json!({"idx": 1}), vec![]);
store.record_assistant_turn("provider-a", "session-2", json!({"idx": 2}), vec![]);
store.record_assistant_turn("provider-b", "session-3", json!({"idx": 3}), vec![]);
assert!(store.clear_session("provider-a", "session-1"));
assert!(store.get_session("provider-a", "session-1").is_none());
let removed = store.clear_provider("provider-a");
assert_eq!(removed, 1);
assert!(store.get_session("provider-a", "session-2").is_none());
assert!(store.get_session("provider-b", "session-3").is_some());
}
}
+12
View File
@@ -18,10 +18,14 @@ mod codex;
pub mod codex_oauth_auth;
pub mod copilot_auth;
mod gemini;
pub(crate) mod gemini_schema;
pub mod gemini_shadow;
pub mod models;
pub mod streaming;
pub mod streaming_gemini;
pub mod streaming_responses;
pub mod transform;
pub mod transform_gemini;
pub mod transform_responses;
use crate::app_config::AppType;
@@ -101,6 +105,14 @@ impl ProviderType {
pub fn from_app_type_and_config(app_type: &AppType, provider: &Provider) -> Self {
match app_type {
AppType::Claude => {
if get_claude_api_format(provider) == "gemini_native" {
let adapter = ClaudeAdapter::new();
return match adapter.extract_auth(provider).map(|auth| auth.strategy) {
Some(AuthStrategy::GoogleOAuth) => ProviderType::GeminiCli,
_ => ProviderType::Gemini,
};
}
// 检测是否为 GitHub Copilot
if let Some(meta) = provider.meta.as_ref() {
if meta.provider_type.as_deref() == Some("github_copilot") {
+2 -5
View File
@@ -2,7 +2,7 @@
//!
//! 实现 OpenAI SSE → Anthropic SSE 格式转换
use crate::proxy::sse::strip_sse_field;
use crate::proxy::sse::{strip_sse_field, take_sse_block};
use bytes::Bytes;
use futures::stream::{Stream, StreamExt};
use serde::{Deserialize, Serialize};
@@ -118,10 +118,7 @@ pub fn create_anthropic_sse_stream<E: std::error::Error + Send + 'static>(
Ok(bytes) => {
crate::proxy::sse::append_utf8_safe(&mut buffer, &mut utf8_remainder, &bytes);
while let Some(pos) = buffer.find("\n\n") {
let line = buffer[..pos].to_string();
buffer = buffer[pos + 2..].to_string();
while let Some(line) = take_sse_block(&mut buffer) {
if line.trim().is_empty() {
continue;
}
File diff suppressed because it is too large Load Diff
@@ -9,7 +9,7 @@
//! 与 Chat Completions 的 delta chunk 模型完全不同,需要独立的状态机处理。
use super::transform_responses::{build_anthropic_usage_from_responses, map_responses_stop_reason};
use crate::proxy::sse::strip_sse_field;
use crate::proxy::sse::{strip_sse_field, take_sse_block};
use bytes::Bytes;
use futures::stream::{Stream, StreamExt};
use serde_json::{json, Value};
@@ -122,10 +122,7 @@ pub fn create_anthropic_sse_stream_from_responses<E: std::error::Error + Send +
crate::proxy::sse::append_utf8_safe(&mut buffer, &mut utf8_remainder, &bytes);
// SSE 事件由 \n\n 分隔
while let Some(pos) = buffer.find("\n\n") {
let block = buffer[..pos].to_string();
buffer = buffer[pos + 2..].to_string();
while let Some(block) = take_sse_block(&mut buffer) {
if block.trim().is_empty() {
continue;
}
File diff suppressed because it is too large Load Diff
@@ -194,23 +194,14 @@ pub(crate) fn map_responses_stop_reason(
incomplete_reason: Option<&str>,
) -> Option<&'static str> {
status.map(|s| match s {
"completed" => {
if has_tool_use {
"tool_use"
} else {
"end_turn"
}
}
"incomplete" => {
"completed" if has_tool_use => "tool_use",
"incomplete"
if matches!(
incomplete_reason,
Some("max_output_tokens") | Some("max_tokens")
) || incomplete_reason.is_none()
{
"max_tokens"
} else {
"end_turn"
}
) || incomplete_reason.is_none() =>
{
"max_tokens"
}
_ => "end_turn",
})
+2 -5
View File
@@ -5,7 +5,7 @@
use super::session::ProxySession;
use super::usage::parser::TokenUsage;
use super::ProxyError;
use crate::proxy::sse::strip_sse_field;
use crate::proxy::sse::{strip_sse_field, take_sse_block};
use bytes::Bytes;
use futures::stream::{Stream, StreamExt};
use serde_json::Value;
@@ -86,10 +86,7 @@ impl StreamHandler {
crate::proxy::sse::append_utf8_safe(&mut buffer, &mut utf8_remainder, &bytes);
// 提取完整事件
while let Some(pos) = buffer.find("\n\n") {
let event_text = buffer[..pos].to_string();
buffer = buffer[pos + 2..].to_string();
while let Some(event_text) = take_sse_block(&mut buffer) {
for line in event_text.lines() {
if let Some(data) = strip_sse_field(line, "data") {
if data.trim() != "[DONE]" {
+4 -5
View File
@@ -7,7 +7,7 @@ use super::{
handler_context::{RequestContext, StreamingTimeoutConfig},
hyper_client::ProxyResponse,
server::ProxyState,
sse::strip_sse_field,
sse::{strip_sse_field, take_sse_block},
usage::parser::TokenUsage,
ProxyError,
};
@@ -662,10 +662,7 @@ pub fn create_logged_passthrough_stream(
crate::proxy::sse::append_utf8_safe(&mut buffer, &mut utf8_remainder, &bytes);
// 尝试解析并记录完整的 SSE 事件
while let Some(pos) = buffer.find("\n\n") {
let event_text = buffer[..pos].to_string();
buffer = buffer[pos + 2..].to_string();
while let Some(event_text) = take_sse_block(&mut buffer) {
if !event_text.trim().is_empty() {
// 提取 data 部分并尝试解析为 JSON
for line in event_text.lines() {
@@ -726,6 +723,7 @@ mod tests {
use crate::provider::ProviderMeta;
use crate::proxy::failover_switch::FailoverSwitchManager;
use crate::proxy::provider_router::ProviderRouter;
use crate::proxy::providers::gemini_shadow::GeminiShadowStore;
use crate::proxy::types::{ProxyConfig, ProxyStatus};
use rust_decimal::Decimal;
use std::collections::HashMap;
@@ -846,6 +844,7 @@ mod tests {
start_time: Arc::new(RwLock::new(None)),
current_providers: Arc::new(RwLock::new(HashMap::new())),
provider_router: Arc::new(ProviderRouter::new(db.clone())),
gemini_shadow: Arc::new(GeminiShadowStore::default()),
app_handle: None,
failover_manager: Arc::new(FailoverSwitchManager::new(db)),
}
+5 -1
View File
@@ -10,7 +10,8 @@
use super::{
failover_switch::FailoverSwitchManager, handlers, log_codes::srv as log_srv,
provider_router::ProviderRouter, types::*, ProxyError,
provider_router::ProviderRouter, providers::gemini_shadow::GeminiShadowStore, types::*,
ProxyError,
};
use crate::database::Database;
use axum::{
@@ -35,6 +36,8 @@ pub struct ProxyState {
pub current_providers: Arc<RwLock<std::collections::HashMap<String, (String, String)>>>,
/// 共享的 ProviderRouter(持有熔断器状态,跨请求保持)
pub provider_router: Arc<ProviderRouter>,
/// Gemini Native shadow state,用于 thoughtSignature / tool call 回放
pub gemini_shadow: Arc<GeminiShadowStore>,
/// AppHandle,用于发射事件和更新托盘菜单
pub app_handle: Option<tauri::AppHandle>,
/// 故障转移切换管理器
@@ -68,6 +71,7 @@ impl ProxyServer {
start_time: Arc::new(RwLock::new(None)),
current_providers: Arc::new(RwLock::new(std::collections::HashMap::new())),
provider_router,
gemini_shadow: Arc::new(GeminiShadowStore::default()),
app_handle,
failover_manager,
};
+69
View File
@@ -242,6 +242,12 @@ pub fn extract_session_id(
body: &serde_json::Value,
client_format: &str,
) -> SessionIdResult {
if client_format == "claude" {
if let Some(result) = extract_claude_session(headers, body) {
return result;
}
}
// Codex 请求特殊处理
if client_format == "codex" || client_format == "openai" {
if let Some(result) = extract_codex_session(headers, body) {
@@ -258,6 +264,28 @@ pub fn extract_session_id(
generate_new_session_id()
}
/// 提取 Claude Session ID
fn extract_claude_session(
headers: &HeaderMap,
body: &serde_json::Value,
) -> Option<SessionIdResult> {
for header_name in &["x-claude-code-session-id", "claude-code-session-id"] {
if let Some(value) = headers.get(*header_name) {
if let Ok(session_id) = value.to_str() {
if !session_id.is_empty() {
return Some(SessionIdResult {
session_id: session_id.to_string(),
source: SessionIdSource::Header,
client_provided: true,
});
}
}
}
}
extract_from_metadata(body)
}
/// 提取 Codex Session ID
fn extract_codex_session(headers: &HeaderMap, body: &serde_json::Value) -> Option<SessionIdResult> {
// 1. 从 headers 提取
@@ -515,6 +543,47 @@ mod tests {
assert!(result.client_provided);
}
#[test]
fn test_extract_session_from_claude_header() {
let mut headers = HeaderMap::new();
headers.insert(
"x-claude-code-session-id",
"d937243f-2702-4f20-97b6-c9682235ab81".parse().unwrap(),
);
let body = json!({
"model": "claude-3-5-sonnet",
"messages": [{"role": "user", "content": "Hello"}]
});
let result = extract_session_id(&headers, &body, "claude");
assert_eq!(result.session_id, "d937243f-2702-4f20-97b6-c9682235ab81");
assert_eq!(result.source, SessionIdSource::Header);
assert!(result.client_provided);
}
#[test]
fn test_extract_session_from_claude_header_precedes_metadata() {
let mut headers = HeaderMap::new();
headers.insert(
"x-claude-code-session-id",
"header-session-123".parse().unwrap(),
);
let body = json!({
"model": "claude-3-5-sonnet",
"messages": [{"role": "user", "content": "Hello"}],
"metadata": {
"session_id": "my-session-123"
}
});
let result = extract_session_id(&headers, &body, "claude");
assert_eq!(result.session_id, "header-session-123");
assert_eq!(result.source, SessionIdSource::Header);
assert!(result.client_provided);
}
#[test]
fn test_extract_session_from_codex_previous_response_id() {
let headers = HeaderMap::new();
+41 -1
View File
@@ -4,6 +4,24 @@ pub(crate) fn strip_sse_field<'a>(line: &'a str, field: &str) -> Option<&'a str>
.or_else(|| line.strip_prefix(&format!("{field}:")))
}
#[inline]
pub(crate) fn take_sse_block(buffer: &mut String) -> Option<String> {
let mut best: Option<(usize, usize)> = None;
for (delimiter, len) in [("\r\n\r\n", 4usize), ("\n\n", 2usize)] {
if let Some(pos) = buffer.find(delimiter) {
if best.is_none_or(|(best_pos, _)| pos < best_pos) {
best = Some((pos, len));
}
}
}
let (pos, len) = best?;
let block = buffer[..pos].to_string();
buffer.drain(..pos + len);
Some(block)
}
/// Append raw bytes to a UTF-8 `String` buffer, correctly handling multi-byte
/// characters that are split across chunk boundaries.
///
@@ -68,7 +86,7 @@ pub(crate) fn append_utf8_safe(buffer: &mut String, remainder: &mut Vec<u8>, new
#[cfg(test)]
mod tests {
use super::{append_utf8_safe, strip_sse_field};
use super::{append_utf8_safe, strip_sse_field, take_sse_block};
#[test]
fn strip_sse_field_accepts_optional_space() {
@@ -91,6 +109,28 @@ mod tests {
assert_eq!(strip_sse_field("id:1", "data"), None);
}
#[test]
fn take_sse_block_supports_lf_delimiters() {
let mut buffer = "data: {\"ok\":true}\n\nrest".to_string();
assert_eq!(
take_sse_block(&mut buffer),
Some("data: {\"ok\":true}".to_string())
);
assert_eq!(buffer, "rest");
}
#[test]
fn take_sse_block_supports_crlf_delimiters() {
let mut buffer = "data: {\"ok\":true}\r\n\r\nrest".to_string();
assert_eq!(
take_sse_block(&mut buffer),
Some("data: {\"ok\":true}".to_string())
);
assert_eq!(buffer, "rest");
}
// ------------------------------------------------------------------
// append_utf8_safe tests
// ------------------------------------------------------------------
+27
View File
@@ -52,6 +52,14 @@ pub fn should_rectify_thinking_signature(
return true;
}
// 场景1b: Gemini/第三方渠道返回 "Thought signature is not valid"
// 错误示例: "Unable to submit request because Thought signature is not valid"
if lower.contains("thought signature")
&& (lower.contains("not valid") || lower.contains("invalid"))
{
return true;
}
// 场景2: assistant 消息必须以 thinking block 开头
// 错误示例: "must start with a thinking block"
if lower.contains("must start with a thinking block") {
@@ -280,6 +288,16 @@ mod tests {
));
}
#[test]
fn test_detect_invalid_thought_signature_message() {
assert!(should_rectify_thinking_signature(
Some(
"Unable to submit request because Thought signature is not valid.. Learn more: https://example.com/help"
),
&enabled_config()
));
}
#[test]
fn test_detect_invalid_signature_nested_json() {
// 测试嵌套 JSON 格式的错误消息(第三方渠道常见格式)
@@ -290,6 +308,15 @@ mod tests {
));
}
#[test]
fn test_detect_invalid_thought_signature_nested_json() {
let nested_error = r#"{"error":{"message":"Unable to submit request because Thought signature is not valid.. Learn more: https://example.com/help","type":"upstream_error","param":"","code":400}}"#;
assert!(should_rectify_thinking_signature(
Some(nested_error),
&enabled_config()
));
}
#[test]
fn test_detect_thinking_expected() {
assert!(should_rectify_thinking_signature(
+1 -1
View File
@@ -27,7 +27,7 @@ pub fn get_custom_endpoints(
}
let mut result: Vec<_> = meta.custom_endpoints.values().cloned().collect();
result.sort_by(|a, b| b.added_at.cmp(&a.added_at));
result.sort_by_key(|ep| std::cmp::Reverse(ep.added_at));
Ok(result)
}
+10 -12
View File
@@ -297,18 +297,16 @@ fn sync_single_codex_file(db: &Database, file_path: &Path) -> Result<(u32, u32),
};
match event_type {
"session_meta" => {
if state.session_id.is_none() {
let payload = value.get("payload");
state.session_id = payload
.and_then(|p| {
p.get("session_id")
.or_else(|| p.get("sessionId"))
.or_else(|| p.get("id"))
})
.and_then(|v| v.as_str())
.map(|s| s.to_string());
}
"session_meta" if state.session_id.is_none() => {
let payload = value.get("payload");
state.session_id = payload
.and_then(|p| {
p.get("session_id")
.or_else(|| p.get("sessionId"))
.or_else(|| p.get("id"))
})
.and_then(|v| v.as_str())
.map(|s| s.to_string());
}
"turn_context" => {
if let Some(payload) = value.get("payload") {
+4 -4
View File
@@ -1268,7 +1268,7 @@ impl SkillService {
}
}
entries.sort_by(|a, b| b.created_at.cmp(&a.created_at));
entries.sort_by_key(|entry| std::cmp::Reverse(entry.created_at));
Ok(entries)
}
@@ -1772,7 +1772,7 @@ impl SkillService {
let results: Vec<Result<Vec<DiscoverableSkill>>> =
futures::future::join_all(fetch_tasks).await;
for (repo, result) in enabled_repos.into_iter().zip(results.into_iter()) {
for (repo, result) in enabled_repos.into_iter().zip(results) {
match result {
Ok(repo_skills) => skills.extend(repo_skills),
Err(e) => log::warn!("获取仓库 {}/{} 技能失败: {}", repo.owner, repo.name, e),
@@ -1781,7 +1781,7 @@ impl SkillService {
// 去重并排序
Self::deduplicate_discoverable_skills(&mut skills);
skills.sort_by(|a, b| a.name.to_lowercase().cmp(&b.name.to_lowercase()));
skills.sort_by_key(|skill| skill.name.to_lowercase());
Ok(skills)
}
@@ -1848,7 +1848,7 @@ impl SkillService {
}
}
skills.sort_by(|a, b| a.name.to_lowercase().cmp(&b.name.to_lowercase()));
skills.sort_by_key(|skill| skill.name.to_lowercase());
Ok(skills)
}
+121 -4
View File
@@ -12,8 +12,10 @@ use std::time::Instant;
use crate::app_config::AppType;
use crate::error::AppError;
use crate::provider::Provider;
use crate::proxy::gemini_url::{normalize_gemini_model_id, resolve_gemini_native_url};
use crate::proxy::providers::copilot_auth;
use crate::proxy::providers::transform::anthropic_to_openai;
use crate::proxy::providers::transform_gemini::anthropic_to_gemini;
use crate::proxy::providers::transform_responses::anthropic_to_responses;
use crate::proxy::providers::{get_adapter, AuthInfo, AuthStrategy};
@@ -288,6 +290,8 @@ impl StreamCheckService {
/// 根据供应商的 api_format 选择请求格式:
/// - "anthropic" (默认): Anthropic Messages API (/v1/messages)
/// - "openai_chat": OpenAI Chat Completions API (/v1/chat/completions)
/// - "openai_responses": OpenAI Responses API (/v1/responses)
/// - "gemini_native": Gemini Native streamGenerateContent
///
/// `extra_headers` 是一个可选的供应商级自定义 header 集合(从 OpenClaw
/// 的 `settings_config.headers` 或 OpenCode 的 `settings_config.options.headers`
@@ -329,8 +333,14 @@ impl StreamCheckService {
.unwrap_or(false);
let is_openai_chat = effective_api_format == "openai_chat";
let is_openai_responses = effective_api_format == "openai_responses";
let url =
Self::resolve_claude_stream_url(base, auth.strategy, effective_api_format, is_full_url);
let is_gemini_native = effective_api_format == "gemini_native";
let url = Self::resolve_claude_stream_url(
base,
auth.strategy,
effective_api_format,
is_full_url,
model,
);
let max_tokens = if is_openai_responses { 16 } else { 1 };
@@ -352,6 +362,9 @@ impl StreamCheckService {
let body = if is_openai_responses {
anthropic_to_responses(anthropic_body, Some(&provider.id), is_codex_oauth)
.map_err(|e| AppError::Message(format!("Failed to build test request: {e}")))?
} else if is_gemini_native {
anthropic_to_gemini(anthropic_body)
.map_err(|e| AppError::Message(format!("Failed to build test request: {e}")))?
} else if is_openai_chat {
anthropic_to_openai(anthropic_body)
.map_err(|e| AppError::Message(format!("Failed to build test request: {e}")))?
@@ -387,6 +400,23 @@ impl StreamCheckService {
.header("x-vscode-user-agent-library-version", "electron-fetch")
.header("x-request-id", &request_id)
.header("x-agent-task-id", &request_id);
} else if is_gemini_native {
request_builder = match auth.strategy {
AuthStrategy::GoogleOAuth => {
let token = auth.access_token.as_ref().unwrap_or(&auth.api_key);
request_builder
.header("authorization", format!("Bearer {token}"))
.header("x-goog-api-client", "GeminiCLI/1.0")
.header("content-type", "application/json")
.header("accept", "text/event-stream")
.header("accept-encoding", "identity")
}
_ => request_builder
.header("x-goog-api-key", &auth.api_key)
.header("content-type", "application/json")
.header("accept", "text/event-stream")
.header("accept-encoding", "identity"),
};
} else if is_openai_chat || is_openai_responses {
// OpenAI-compatible targets: Bearer auth + SSE headers only
request_builder = request_builder
@@ -568,13 +598,16 @@ impl StreamCheckService {
extra_headers: Option<&serde_json::Map<String, serde_json::Value>>,
) -> Result<(u16, String), AppError> {
let base = base_url.trim_end_matches('/');
// Strip `models/` resource-name prefix from the model id — see
// `normalize_gemini_model_id` for rationale.
let normalized_model = normalize_gemini_model_id(model);
// Gemini 原生 API: /v1beta/models/{model}:streamGenerateContent?alt=sse
// 智能处理 /v1beta 路径:如果 base_url 不包含版本路径,则添加 /v1beta
// alt=sse 参数使 API 返回 SSE 格式(text/event-stream)而非 JSON 数组
let url = if base.contains("/v1beta") || base.contains("/v1/") {
format!("{base}/models/{model}:streamGenerateContent?alt=sse")
format!("{base}/models/{normalized_model}:streamGenerateContent?alt=sse")
} else {
format!("{base}/v1beta/models/{model}:streamGenerateContent?alt=sse")
format!("{base}/v1beta/models/{normalized_model}:streamGenerateContent?alt=sse")
};
// Gemini 原生请求体格式
@@ -1292,7 +1325,19 @@ impl StreamCheckService {
auth_strategy: AuthStrategy,
api_format: &str,
is_full_url: bool,
model: &str,
) -> String {
if api_format == "gemini_native" {
// Strip an optional `models/` resource-name prefix so that model
// identifiers copied from Gemini SDK outputs (e.g.
// `models/gemini-2.5-pro`) don't produce a doubled
// `/v1beta/models/models/...` URL.
let normalized_model = normalize_gemini_model_id(model);
let endpoint =
format!("/v1beta/models/{normalized_model}:streamGenerateContent?alt=sse");
return resolve_gemini_native_url(base_url, &endpoint, is_full_url);
}
if is_full_url {
return base_url.to_string();
}
@@ -1621,6 +1666,7 @@ mod tests {
AuthStrategy::Bearer,
"openai_chat",
true,
"gpt-5.4",
);
assert_eq!(url, "https://relay.example/v1/chat/completions");
@@ -1633,6 +1679,7 @@ mod tests {
AuthStrategy::GitHubCopilot,
"openai_chat",
false,
"gpt-5.4",
);
assert_eq!(url, "https://api.githubcopilot.com/chat/completions");
@@ -1645,6 +1692,7 @@ mod tests {
AuthStrategy::GitHubCopilot,
"openai_responses",
false,
"gpt-5.4",
);
assert_eq!(url, "https://api.githubcopilot.com/v1/responses");
@@ -1657,6 +1705,7 @@ mod tests {
AuthStrategy::Bearer,
"openai_chat",
false,
"gpt-5.4",
);
assert_eq!(url, "https://example.com/v1/chat/completions");
@@ -1669,6 +1718,7 @@ mod tests {
AuthStrategy::Bearer,
"openai_responses",
false,
"gpt-5.4",
);
assert_eq!(url, "https://example.com/v1/responses");
@@ -1681,11 +1731,78 @@ mod tests {
AuthStrategy::Anthropic,
"anthropic",
false,
"claude-sonnet-4-6",
);
assert_eq!(url, "https://api.anthropic.com/v1/messages");
}
#[test]
fn test_resolve_claude_stream_url_for_gemini_native() {
let url = StreamCheckService::resolve_claude_stream_url(
"https://generativelanguage.googleapis.com",
AuthStrategy::Google,
"gemini_native",
false,
"gemini-2.5-flash",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
#[test]
fn test_resolve_claude_stream_url_for_gemini_native_full_url_openai_compat_base() {
let url = StreamCheckService::resolve_claude_stream_url(
"https://generativelanguage.googleapis.com/v1beta/openai/chat/completions",
AuthStrategy::Google,
"gemini_native",
true,
"gemini-2.5-flash",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"
);
}
#[test]
fn test_resolve_claude_stream_url_for_gemini_native_opaque_full_url() {
let url = StreamCheckService::resolve_claude_stream_url(
"https://relay.example/custom/generate-content",
AuthStrategy::Google,
"gemini_native",
true,
"gemini-2.5-flash",
);
assert_eq!(url, "https://relay.example/custom/generate-content?alt=sse");
}
/// Regression: Gemini SDK outputs commonly surface model ids as the
/// resource-name form `models/gemini-2.5-pro`. Interpolating that raw
/// value used to produce `/v1beta/models/models/gemini-2.5-pro:...`
/// which the upstream rejects and the health check records as a
/// false-negative for an otherwise valid provider.
#[test]
fn test_resolve_claude_stream_url_for_gemini_native_strips_models_prefix() {
let url = StreamCheckService::resolve_claude_stream_url(
"https://generativelanguage.googleapis.com",
AuthStrategy::Google,
"gemini_native",
false,
"models/gemini-2.5-pro",
);
assert_eq!(
url,
"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:streamGenerateContent?alt=sse"
);
}
#[test]
fn test_resolve_codex_stream_urls_for_full_url_mode() {
let urls = StreamCheckService::resolve_codex_stream_urls(
+1 -4
View File
@@ -173,10 +173,7 @@ async fn run_worker_loop(
let started_at = Instant::now();
let mut merged_count = 1usize;
loop {
let Some(wait_for) = auto_sync_wait_duration(started_at, Instant::now()) else {
break;
};
while let Some(wait_for) = auto_sync_wait_duration(started_at, Instant::now()) {
let timeout = tokio::time::timeout(wait_for, rx.recv()).await;
match timeout {
@@ -113,7 +113,7 @@ interface ClaudeFormFieldsProps {
// Speed Test Endpoints
speedTestEndpoints: EndpointCandidate[];
// API Format (for third-party providers that use OpenAI Chat Completions format)
// API Format (for Claude-compatible providers that need request/response conversion)
apiFormat: ClaudeApiFormat;
onApiFormatChange: (format: ClaudeApiFormat) => void;
@@ -436,7 +436,14 @@ export function ClaudeFormFields({
? t("providerForm.apiHintResponses")
: apiFormat === "openai_chat"
? t("providerForm.apiHintOAI")
: t("providerForm.apiHint")
: apiFormat === "gemini_native"
? t("providerForm.apiHintGeminiNative")
: t("providerForm.apiHint")
}
fullUrlHint={
apiFormat === "gemini_native"
? t("providerForm.fullUrlHintGeminiNative")
: undefined
}
onManageClick={() => onEndpointModalToggle(true)}
showFullUrlToggle={true}
@@ -511,6 +518,11 @@ export function ClaudeFormFields({
defaultValue: "OpenAI Responses API (需转换)",
})}
</SelectItem>
<SelectItem value="gemini_native">
{t("providerForm.apiFormatGeminiNative", {
defaultValue: "Gemini Native generateContent (需转换)",
})}
</SelectItem>
</SelectContent>
</Select>
<p className="text-xs text-muted-foreground">
@@ -11,6 +11,7 @@ interface EndpointFieldProps {
onChange: (value: string) => void;
placeholder: string;
hint?: string;
fullUrlHint?: string;
showManageButton?: boolean;
onManageClick?: () => void;
manageButtonLabel?: string;
@@ -26,6 +27,7 @@ export function EndpointField({
onChange,
placeholder,
hint,
fullUrlHint,
showManageButton = true,
onManageClick,
manageButtonLabel,
@@ -40,7 +42,8 @@ export function EndpointField({
});
const effectiveHint =
showFullUrlToggle && isFullUrl
? t("providerForm.fullUrlHint", {
? fullUrlHint ||
t("providerForm.fullUrlHint", {
defaultValue:
"💡 请填写完整请求 URL,并且必须开启代理后使用;代理将直接使用此 URL,不拼接路径",
})
+27 -1
View File
@@ -49,7 +49,12 @@ export interface ProviderPreset {
// - "anthropic" (默认): Anthropic Messages API 格式,直接透传
// - "openai_chat": OpenAI Chat Completions 格式,需要格式转换
// - "openai_responses": OpenAI Responses API 格式,需要格式转换
apiFormat?: "anthropic" | "openai_chat" | "openai_responses";
// - "gemini_native": Gemini Native generateContent API 格式,需要格式转换
apiFormat?:
| "anthropic"
| "openai_chat"
| "openai_responses"
| "gemini_native";
// 供应商类型标识(用于特殊供应商检测)
// - "github_copilot": GitHub Copilot 供应商(需要 OAuth 认证)
@@ -80,6 +85,27 @@ export const providerPresets: ProviderPreset[] = [
icon: "anthropic",
iconColor: "#D4915D",
},
{
name: "Gemini Native",
websiteUrl: "https://ai.google.dev/gemini-api",
apiKeyUrl: "https://aistudio.google.com/app/apikey",
apiKeyField: "ANTHROPIC_API_KEY",
settingsConfig: {
env: {
ANTHROPIC_BASE_URL: "https://generativelanguage.googleapis.com",
ANTHROPIC_API_KEY: "",
ANTHROPIC_MODEL: "gemini-2.5-pro",
ANTHROPIC_DEFAULT_HAIKU_MODEL: "gemini-2.5-flash",
ANTHROPIC_DEFAULT_SONNET_MODEL: "gemini-2.5-pro",
ANTHROPIC_DEFAULT_OPUS_MODEL: "gemini-2.5-pro",
},
},
category: "third_party",
apiFormat: "gemini_native",
endpointCandidates: ["https://generativelanguage.googleapis.com"],
icon: "gemini",
iconColor: "#4285F4",
},
{
name: "Shengsuanyun",
nameKey: "providerForm.presets.shengsuanyun",
+3
View File
@@ -801,6 +801,7 @@
"modelHint": "💡 Leave blank to use provider's default model",
"apiHint": "💡 Fill in Claude API compatible service endpoint, avoid trailing slash",
"apiHintOAI": "💡 Fill in OpenAI Chat Completions compatible service endpoint, avoid trailing slash",
"apiHintGeminiNative": "💡 Prefer a Gemini Native base URL such as https://generativelanguage.googleapis.com or https://generativelanguage.googleapis.com/v1beta; the proxy will append models/*:generateContent automatically",
"codexApiHint": "💡 Fill in service endpoint compatible with OpenAI Response format",
"fillSupplierName": "Please fill in provider name",
"fillConfigContent": "Please fill in configuration content",
@@ -823,9 +824,11 @@
"fullUrlEnabled": "Full URL Mode",
"fullUrlDisabled": "Mark as Full URL",
"fullUrlHint": "💡 Enter the full request URL. This mode requires routing to be enabled, and routing will use the URL as-is without appending a path",
"fullUrlHintGeminiNative": "💡 In Gemini Native full URL mode, two inputs are supported: 1. official/structured Gemini URLs, which will still be normalized to the requested model and streaming method; 2. opaque custom relay URLs, which will be used mostly as-is with only query parameters appended",
"apiFormatAnthropic": "Anthropic Messages (Native)",
"apiFormatOpenAIChat": "OpenAI Chat Completions (Requires routing)",
"apiFormatOpenAIResponses": "OpenAI Responses API (Requires routing)",
"apiFormatGeminiNative": "Gemini Native generateContent (Requires routing)",
"authField": "Auth Field",
"authFieldAuthToken": "ANTHROPIC_AUTH_TOKEN (Default)",
"authFieldApiKey": "ANTHROPIC_API_KEY",
+3
View File
@@ -801,6 +801,7 @@
"modelHint": "💡 空欄ならプロバイダーのデフォルトモデルを使用します",
"apiHint": "💡 Claude API 互換サービスのエンドポイントを入力してください。末尾にスラッシュを付けないでください",
"apiHintOAI": "💡 OpenAI Chat Completions 互換サービスのエンドポイントを入力してください。末尾にスラッシュを付けないでください",
"apiHintGeminiNative": "💡 Gemini Native では https://generativelanguage.googleapis.com または https://generativelanguage.googleapis.com/v1beta のような base URL を推奨します。プロキシが models/*:generateContent を自動補完します",
"codexApiHint": "💡 OpenAI Response 互換のサービスエンドポイントを入力してください",
"fillSupplierName": "プロバイダー名を入力してください",
"fillConfigContent": "設定内容を入力してください",
@@ -823,9 +824,11 @@
"fullUrlEnabled": "フル URL モード",
"fullUrlDisabled": "フル URL として設定",
"fullUrlHint": "💡 完全なリクエスト URL を入力してください。このモードはルーティングを有効にして使用する必要があり、ルーティングはこの URL をそのまま使用し、パスを追加しません",
"fullUrlHintGeminiNative": "💡 Gemini Native のフル URL モードでは 2 種類の入力を扱えます。1. 公式/構造化された Gemini URL は、要求されたモデルやストリーミング方式に合わせて正規化されます。2. カスタム relay の opaque な完全 URL は、主にそのまま使用され、必要なクエリだけ追加されます",
"apiFormatAnthropic": "Anthropic Messages(ネイティブ)",
"apiFormatOpenAIChat": "OpenAI Chat Completions(ルーティングが必要)",
"apiFormatOpenAIResponses": "OpenAI Responses API(ルーティングが必要)",
"apiFormatGeminiNative": "Gemini Native generateContent(ルーティングが必要)",
"authField": "認証フィールド",
"authFieldAuthToken": "ANTHROPIC_AUTH_TOKEN(デフォルト)",
"authFieldApiKey": "ANTHROPIC_API_KEY",
+3
View File
@@ -802,6 +802,7 @@
"modelHint": "💡 留空将使用供应商的默认模型",
"apiHint": "💡 填写兼容 Claude API 的服务端点地址,不要以斜杠结尾",
"apiHintOAI": "💡 填写兼容 OpenAI Chat Completions 的服务端点地址,不要以斜杠结尾",
"apiHintGeminiNative": "💡 建议填写 Gemini Native 的 base URL,例如 https://generativelanguage.googleapis.com 或 https://generativelanguage.googleapis.com/v1beta;代理会自动补全 models/*:generateContent",
"codexApiHint": "💡 填写兼容 OpenAI Response 格式的服务端点地址",
"fillSupplierName": "请填写供应商名称",
"fillConfigContent": "请填写配置内容",
@@ -824,9 +825,11 @@
"fullUrlEnabled": "完整 URL 模式",
"fullUrlDisabled": "标记为完整 URL",
"fullUrlHint": "💡 请填写完整请求 URL,并且必须开启路由后使用;路由将直接使用此 URL,不拼接路径",
"fullUrlHintGeminiNative": "💡 Gemini Native 下,完整 URL 模式同时兼容两类地址:1. 官方/标准 Gemini URL,代理会按模型和流式参数自动归一化;2. 自定义 relay 的完整 URL,代理会尽量原样使用,只补查询参数,不再强行追加 models 路径",
"apiFormatAnthropic": "Anthropic Messages (原生)",
"apiFormatOpenAIChat": "OpenAI Chat Completions (需开启路由)",
"apiFormatOpenAIResponses": "OpenAI Responses API (需开启路由)",
"apiFormatGeminiNative": "Gemini Native generateContent (需开启路由)",
"authField": "认证字段",
"authFieldAuthToken": "ANTHROPIC_AUTH_TOKEN(默认)",
"authFieldApiKey": "ANTHROPIC_API_KEY",
+12 -2
View File
@@ -141,7 +141,12 @@ export interface ProviderMeta {
// - "anthropic": 原生 Anthropic Messages API 格式,直接透传
// - "openai_chat": OpenAI Chat Completions 格式,需要格式转换
// - "openai_responses": OpenAI Responses API 格式,需要格式转换
apiFormat?: "anthropic" | "openai_chat" | "openai_responses";
// - "gemini_native": Gemini Native generateContent API 格式,需要格式转换
apiFormat?:
| "anthropic"
| "openai_chat"
| "openai_responses"
| "gemini_native";
// 通用认证绑定
authBinding?: AuthBinding;
// Claude 认证字段名
@@ -166,7 +171,12 @@ export type SkillStorageLocation = "cc_switch" | "unified";
// - "anthropic": 原生 Anthropic Messages API 格式,直接透传
// - "openai_chat": OpenAI Chat Completions 格式,需要格式转换
// - "openai_responses": OpenAI Responses API 格式,需要格式转换
export type ClaudeApiFormat = "anthropic" | "openai_chat" | "openai_responses";
// - "gemini_native": Gemini Native generateContent API 格式,需要格式转换
export type ClaudeApiFormat =
| "anthropic"
| "openai_chat"
| "openai_responses"
| "gemini_native";
// Claude 认证字段类型
export type ClaudeApiKeyField = "ANTHROPIC_AUTH_TOKEN" | "ANTHROPIC_API_KEY";