tipsy-ab-config 0.15.0__tar.gz

tipsy_ab_config-0.15.0/CHANGELOG.md

@@ -0,0 +1,424 @@
+# Changelog
+
+All notable changes to the `tipsy-ab-config` Python SDK are documented in
+this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [Unreleased]
+
+## [0.15.0] - 2026-08-27
+
+### Added
+- `Config.secret_key`：仅配置 secretKey（即 `TIPSY_SERVICE_SECRET` 本身）即可完成鉴权（issue #16，
+  需平台侧配套 verifier）。凭据优先级 **SecretKey > token_provider > token**（每请求求值）；
+  发送形态为 `Authorization: SecretKey <secret>`（gRPC metadata 与 HTTP header 双路径一致，
+  SDK 恒发精确字面量，服务端按 RFC 7235 对 scheme 大小写不敏感匹配）。Init 校验放宽为
+  `secret_key / token / token_provider` 至少其一（`channel_factory` 豁免语义不变：工厂通道
+  绕开 SDK auth 插桩，secret_key 同样不附着，无需凭据）。SDK 不读任何环境变量，secretKey 由
+  业务方显式传入。secretKey 校验通过 = 全量访问（等效 `internal_service` + `"*"`），信任边界
+  见 `docs/usage-and-integration.md` §3。
+- `get_config` / `get_config_default` 命中日志（`get_config hit (abtest)` / `(full)`，msg 文本不变）
+  新增结构化归因字段：恒定输出 `reason`（枚举 4 值：`full_release` / `experiment` /
+  `gray_whitelist` / `abtest_unattributed`）与 `trace_id`（Python 端本次补齐，取自
+  AbtestContext，对齐 Go/Java）；`reason=experiment` 时另输出 `experiment_id` / `group_id`，
+  `reason=gray_whitelist` 时另输出 `release_id`（条件字段缺席即 omit，不输出键）。既有字段
+  `ns` / `key` / `version` / `uid` 不变。`get_all_configs` 不打每-key 命中日志（其聚合 Debug
+  日志不变）；`get_config_static` 行为与日志不变（无 reason，不属于入组事件）。
+
+### Changed
+- 内部 per-ns 取数（`_fetch_config_version_flat_kv_for_ns`）请求的 display_type 由
+  `FLAT_KV` 切换为 `EACH_EXPERIMENT_GROUP`（experiment_type 仍为 `CONFIG_VERSION`），
+  SDK 按平台 flat 合并规则本地重建 key→versionId map（灰度无条件优先于实验；实验组间
+  后写覆盖；灰度间 first-writer-wins），值解析结果与原 FLAT_KV 语义等价，同时保留服务端
+  折叠前的归因信息供命中日志使用。公共 API（`get_experiment_result` 及含 "flat_kv" 的
+  方法名）签名与语义不变。
+- ab→full fallback 的 WARN 日志（`ab version missing in local cache`）补上 `trace_id`。
+
+## [0.14.2] - 2026-08-24
+
+### Changed
+- `get_config` / `get_config_default` / `get_config_static` 命中日志从 `DEBUG` 提升到 `INFO`
+  （字段 `ns`/`key`/`version`，动态路径另带 `uid`）。目的：在默认 `INFO` 级别下即可
+  从调用方日志观测业务实际请求的 config key 与命中版本，无需临时开 DEBUG。
+  出于泄漏风险，命中日志**不记录 config value**。
+
+## [0.14.1] - 2026-08-10
+
+### Fixed
+
+- **`ConfigService.Subscribe` was sent with no `authorization` header, so every
+  subscribe attempt was rejected with `Unauthenticated: missing authorization
+  metadata`** ([#8](https://github.com/Lightspeed-Intelligence/tipsy-ab-config-sdk/issues/8)).
+  The bearer interceptor was a single class inheriting both
+  `grpc.aio.UnaryUnaryClientInterceptor` and
+  `grpc.aio.UnaryStreamClientInterceptor`; grpcio sorts interceptors into
+  per-method-kind lists with an if/elif chain, so the one instance was only ever
+  registered as unary-unary and the unary-stream list stayed empty. Unary calls
+  (`PullAll`, `GetExperimentResult`) authenticated normally, which is why this
+  was not an outage: change detection silently degraded from Subscribe push to
+  `pull_interval` polling (default 10s staleness) while the log filled with one
+  `Subscribe stream error; reconnecting` ERROR per reconnect, indefinitely. Fixed
+  by splitting into two single-purpose interceptors registered separately.
+  Affects the plaintext and TLS paths equally. Go and Java are unaffected — they
+  use `PerRPCCredentials` / `CallCredentials`, which apply to all RPC kinds.
+
+  **Upgrade note.** If you run in gRPC mode and have been seeing repeated
+  `Subscribe stream error; reconnecting` ERROR lines, this release is the fix:
+  config-change push starts working again and that log noise stops. No code or
+  config change is needed on your side.
+
+- **Declared dependency floors were not installable.** `grpcio>=1.60` disagreed
+  with the checked-in generated stubs, which carry
+  `GRPC_GENERATED_VERSION = '1.66.2'` and raise `RuntimeError` at import on
+  anything lower — so a consumer resolving to the declared floor (an explicit
+  pin, an old lockfile, a conservative resolver) crashed on
+  `import tipsy_ab_config`. Raised to `grpcio>=1.66.2,<2`, which is the floor the
+  stubs actually enforce; wheels exist for every supported Python (3.10–3.13), so
+  no platform support is lost. Separately, `protobuf>=5.29` named a floor that
+  resolves to the **yanked** 5.29.0 when pinned exactly
+  ([protocolbuffers/protobuf#19430](https://github.com/protocolbuffers/protobuf/issues/19430));
+  raised to `>=5.29.1,<7`. No runtime behaviour change — this only corrects
+  metadata that was wrong, plus the matching README compatibility table.
+  A new `min-deps` CI job installs the declared floors and runs the full suite
+  against them (the existing matrix always resolves to the newest release, which
+  is why this escaped), and `tests/test_version_sync.py` now derives the expected
+  `grpcio` floor from the stubs so the two cannot silently diverge again.
+
+  **Upgrade note.** If your project pins `grpcio` or `protobuf` below these
+  floors, raise the pins when you upgrade. The floors were always what the
+  shipped stubs required — the declared metadata was simply wrong.
+
+## [0.14.0] - 2026-07-28
+
+### Added
+
+- `Client.get_all_configs(ctx, namespace)` and
+  `Client.get_all_configs_default(ctx)` — resolve **every** dynamic config under
+  a `(namespace, user)` into a fresh `dict[str, str]` mapping key → value, using
+  the exact same resolution logic as `get_config` (abtest whitelist / experiment
+  hit > full-release fallback), just without a specific `key`. A key that
+  resolves to neither an abtest hit nor a full-release value is **omitted** from
+  the map (no value = not present; a genuinely empty-string value is still
+  returned). A subscribed-but-not-yet-pulled namespace (no snapshot) returns an
+  empty `dict` with zero RPC. The per-ns abtest result is shared with
+  `get_config`, so the whole request link issues **at most one**
+  `GetExperimentResult` RPC per namespace; when every key in the namespace is
+  explicitly pure-full-release (`has_dynamic_resolution` present and `False`) the
+  abtest wait is skipped entirely (zero RPC). `get_all_configs_default` resolves
+  the project default namespace. Mirrors the Go and Java SDKs.
+
+### Changed
+
+- **A `user_id` of `""` or `"0"` is now treated as identity-less: no
+  `GetExperimentResult` RPC, static resolution only.** Both `get_config` and
+  `get_all_configs` for such a context skip the experiment / whitelist logic
+  entirely (no RPC) and resolve straight to the full-release value — for
+  `get_config` the default when no full release exists; for `get_all_configs`
+  the key is dropped. The short-circuit lives in the lazy-fetch layer, so
+  `prefetch_config_version_flat_kv_for_namespace` / `wait_for_abtest` also issue
+  no RPC for these uids (behaving like `empty_abtest_context`). A pre-seeded
+  `mock_abtest_context(user_id="", …)` result still wins. Any other uid is
+  unchanged. The `"0"` value is called out explicitly because it is a common
+  "no logged-in user" sentinel. Previously both `""` and `"0"` sent a
+  `GetExperimentResult` RPC with an empty/zero user_id, which has no meaningful
+  experiment bucketing. Mirrors the Go and Java SDKs.
+- `AbtestContext(user_id=None)` is now normalised to `""` at construction
+  (parity with the Java constructor), so a `None` uid takes the identity-less
+  short-circuit above instead of reaching proto encoding.
+
+## [0.13.0] - 2026-07-23
+
+### Changed
+
+- **env is now judged server-side; the SDK no longer sends an env request
+  field.** env matching moved to the abConfig server, which reads its own
+  `TIPSY_ENV` process environment variable and compares it against each
+  experiment's env set. Removed `Config.env`; the four request construction
+  points (`get_experiment_result`, the `config_version` flat_kv fetch behind
+  `get_config`, `_pull_once`, `_subscribe_once`) no longer set the request's
+  `env` field.
+
+### Compatibility
+
+- The `env` field remains in the vendored proto bindings (`abtest_pb2.py` /
+  `config_pb2.py`) and is left unset by this SDK — retained for wire
+  compatibility with the previously released 0.12.0 and now deprecated on the
+  request path. Newer servers ignore any request-supplied env and use
+  `TIPSY_ENV` instead, so both old (0.12.0, still sending env) and new SDK
+  builds interoperate with the new server. The proto bindings are unchanged (no
+  regen).
+
+## [0.12.0] - 2026-07-22
+
+### Added
+
+- `Config.env` (`str`, default `""`) — a single-value environment identifier
+  stamped onto **every** outbound request: `GetExperimentResult`, the
+  `config_version` flat_kv fetch behind `get_config`, the background `PullAll`,
+  and the `Subscribe` stream. It tells the server which environment this
+  process runs in so experiment env filtering can apply (an experiment with a
+  non-empty env set is only entered when this env is a member; `env=""` enters
+  only experiments that do not restrict their env). No `os.getenv` fallback —
+  the value is used verbatim. Regenerates the vendored proto bindings
+  (`abtest_pb2.py` / `config_pb2.py`) to include the `env` field on the five
+  SDK-facing request messages. Mirrors the Go and Java SDKs.
+
+### Compatibility
+
+- 100% backwards compatible. `env` defaults to `""`; the HTTP transport's
+  `MessageToJson` omits an empty scalar, so an unset `env` is byte-for-byte
+  wire-compatible with older servers. A configured env sent to an older server
+  that predates the field is safely ignored.
+
+## [0.11.0] - 2026-07-16
+
+### Added
+
+- Ignore Subscribe `Heartbeat` events (forward-compat): the server may emit
+  liveness-only `Heartbeat` frames on an otherwise-idle Subscribe stream. The
+  SDK now handles them as an explicit no-op — no cache mutation, no sequence
+  tracking, no metric — and regenerates the vendored proto bindings
+  (`config_pb2.py`) to include the `Heartbeat` message. Mirrors the Go SDK.
+
+### Changed
+
+- Reset the Subscribe reconnect backoff after a healthy connection drop: a
+  stream that stayed up at least 60s before dropping now reconnects at the
+  initial delay instead of continuing the exponential climb. A short-lived
+  connection still backs off exponentially. Mirrors the Go SDK.
+
+## [0.10.0] - 2026-07-03
+
+### Added
+
+- Debug-level per-call timing log for `GetExperimentResult`. Both call
+  sites (the public `get_experiment_result` and the `AbtestContext` lazy
+  per-namespace fetch) emit one Debug record per RPC with `ns`, `trace_id`
+  and float-millisecond `duration_ms` (an `err` field on failure). Fields
+  are written both into the log message and `extra=` so they are visible
+  under the default formatter and available as structured record
+  attributes. Info level stays silent; the existing fallback warnings are
+  unchanged. Mirrors the Go SDK v0.9.0 change.
+
+## [0.9.0] - 2026-06-30
+
+### Added
+
+- Typed config accessors, mirroring the Go SDK's v0.7.0 surface. Static
+  (sync, cache-only) and dynamic (async, abtest-resolved) variants for each
+  scalar type plus JSON:
+  - Static: `get_config_static_bool/long/double/string/json(namespace, key,
+    default=...)`.
+  - Dynamic: `async get_config_bool/long/double/string/json(ctx, namespace,
+    key, default=...)`.
+  - **Bool is lenient and never raises**: the stripped value casefold-equal to
+    `"true"` or equal to `"1"` ⇒ `True`, everything else ⇒ `False`. The console
+    writes canonical `true`/`false`, so the write-strict / read-lenient
+    asymmetry is deliberate.
+  - **Long uses `int(str)`**, which is arbitrary-precision in Python, so values
+    beyond 2^53 are lossless.
+  - Miss and value-parse failure both fall back to the supplied default; the
+    underlying `get_config` exceptions (`SDKClosed` / `NamespaceRequired` /
+    `NamespaceNotSubscribed` / abtest-context) still propagate — only
+    `ValueError` / `json.JSONDecodeError` from parsing are swallowed.
+  - The config value stays a canonical string end-to-end; the declared
+    `value_type` is a console-side write contract and is not carried to the SDK.
+
+## [0.8.0] - 2026-07-01
+
+### Changed (BREAKING)
+
+- **`GetExperimentResultResponse.gray_hits` is now grouped per gray release.**
+  `GrayReleaseHit` changed from the flat `{release_id, key, version_id}` (one
+  entry per `(release, key)`) to `{release_id, map<string,int64> key_versions}`
+  (one entry per hit `release_id`; `key_versions` maps each `config_key.key`
+  name → versionId). This aligns `gray_hits` with
+  `ExperimentGroupResult.params_versions`. Read a single key's target via
+  `gray_hits[i].key_versions[key_name]` instead of the removed
+  `gray_hits[i].version_id`. No backward compatibility — the old flat fields
+  are gone. The int64 values remain versionId (config_version PRIMARY KEY id,
+  globally unique), never the semantic version_no.
+
+## [0.7.0] - 2026-06-27
+
+### Added
+
+- `KeyState.has_dynamic_resolution` (`Optional[bool]`) — the SDK now
+  deserializes the new `optional bool has_dynamic_resolution` field on each
+  `KeyState` from both the `PullAll` and `Subscribe` snapshot paths. The value
+  preserves wire presence: `True`/`False` when the server explicitly set it,
+  `None` when the field was absent (an older server that predates it).
+- `ConfigCache.has_dynamic_resolution(namespace, key) -> tuple[bool, bool]` —
+  cache accessor returning `(value, present)`; `(False, False)` on any miss
+  (no snapshot, no such key, or the field absent on the wire).
+
+### Changed
+
+- `Client.get_config` fast path: when the server has explicitly reported that a
+  key carries no gray release / experiment (`has_dynamic_resolution` is present
+  **and** `False`), `get_config` now **skips** the `GetExperimentResult` abtest
+  wait and returns the full-release value directly. This removes a guaranteed-
+  wasted RPC for pure-full-release keys. All fallback/default semantics and
+  metrics are unchanged. When the field is absent (old server) or `True`, the
+  existing always-wait abtest path is preserved exactly, so gray release /
+  experiments never silently break.
+
+### Server compatibility (REQUIRED — upgrade server first)
+
+- This version expects the server to already emit the
+  `has_dynamic_resolution` field on `KeyState`, i.e. a server built against
+  **`api/gen/go` v0.3.0 or newer**. **Upgrade the server first, then this
+  SDK.** If this SDK is pointed at an older server that does not emit the
+  field, it safely falls back to the previous behaviour — it ALWAYS waits on
+  the abtest result (correct results; gray release / experiments keep working),
+  it just does not get the fast-path benefit. The fast path is only taken when
+  the field is explicitly `False`, never when it is absent, so a new SDK on an
+  old server never wrongly skips abtest.
+
+## [0.6.0] - 2026-06-25
+
+### Changed (BREAKING)
+
+- `Client.new_abtest_context` and `Client.abtest_scope` no longer accept a
+  `namespace` parameter. Construction is now a pure create and issues **no**
+  `GetExperimentResult` RPC. Every namespace is fetched lazily and memoised
+  at-most-once on first dynamic `get_config`. **Latency-shape change:** the
+  default namespace is no longer eagerly pre-fetched at construction, so the
+  first `get_config` for it now pays the GetExperimentResult RPC latency
+  inline (previously that RPC was overlapped with construction).
+- `AbtestMiddleware` no longer auto-prefetches on every request. It gained a
+  `prefetch_paths: Optional[Sequence[str]] = None` parameter (an exact-match
+  URL whitelist). The **default is no prefetch**; only requests whose
+  `request.url.path` exactly matches a whitelisted path warm the default
+  namespace. This removes the "every request through the middleware fires a
+  possibly-wasted experiment RPC" behaviour.
+
+### Added
+
+- `AbtestContext.prefetch_config_version_flat_kv_for_namespace(ns)` — explicit,
+  opt-in, single-namespace prefetch API. Non-blocking and idempotent; spawns
+  the single per-ns `GetExperimentResult` (config_version + flat_kv) task so a
+  later `get_config` for that namespace reuses it (at-most-once). Empty /
+  identity-less / mock contexts and unsubscribed namespaces short-circuit
+  without issuing any RPC. Requires a running asyncio event loop.
+
+### Internal
+
+- Renamed the internal `Client._get_experiment_result_for_ns` →
+  `Client._fetch_config_version_flat_kv_for_ns` (hardwired config_version +
+  flat_kv), to disambiguate it from the public general-purpose
+  `Client.get_experiment_result` (unchanged). The slot-creation logic moved
+  into a single shared `AbtestContext._ensure_fetch` primitive (synchronous,
+  lock-free) used by both `wait_for_abtest` and the new prefetch API, removing
+  the per-context `asyncio.Lock`.
+
+## [0.5.0] - 2026-06-19
+
+### Fixed
+
+- Restore version-string parity: prior `python-sdk/v0.4.0` shipped with
+  `__init__.py` still claiming 0.3.0; this release jumps to 0.5.0 to skip
+  the inconsistent state and re-establish a clean baseline.
+  (修复版本号一致性：上一版 `python-sdk/v0.4.0` 发布时 `__init__.py`
+  仍声明 0.3.0，存在历史不一致；本版直接跳到 0.5.0 重建清晰基线。)
+
+### Added
+
+- Auto-enable client-side `round_robin` load balancing when the channel
+  address uses the `dns:///` gRPC name resolver scheme (typically
+  `dns:///<service>.<ns>.svc.cluster.local:<port>` for K8S Headless
+  Service deployment). All other address forms (bare `host:port`,
+  `grpc://`, `grpcs://`, `passthrough:///`, `unix:`) keep grpcio's
+  default `pick_first` behavior. No SDK `Config` field changes; opt-in
+  via address string only. See `docs/usage-and-integration.md` §4.1.
+
+## [0.4.0] - 2026-06-18
+
+### Removed (BREAKING)
+
+- `Config.exposure_sink`, `Config.exposure_dedup_ttl`, class `ExposureSink`,
+  class `ExposureEvent`, internal `ExposureEmitter`. The Python SDK no
+  longer emits exposure events on `get_config`. Use the upstream
+  experiment-result data report channel instead.
+- `GetExperimentResultResponse.exposures` is retained on the proto wire
+  for backward compatibility but is never populated by the server.
+
+### Added
+
+- `GetExperimentResultResponse.gray_hits` (`repeated GrayReleaseHit`) —
+  populated when `display_type==EACH_EXPERIMENT_GROUP` and
+  `experiment_type ∈ {CONFIG_VERSION, ALL}`; otherwise an empty list.
+
+## [0.3.0] - 2026-06-18
+
+### Changed
+
+- **Repository move.** The Python SDK now lives in the public
+  `Lightspeed-Intelligence/tipsy-ab-config-sdk` repository. Install via
+  `pip install "git+https://github.com/Lightspeed-Intelligence/tipsy-ab-config-sdk.git@python-sdk/v0.3.0#subdirectory=sdk/python"`
+  — no `GH_PAT` needed.
+- **License → MIT.** The Python SDK package is now released under MIT,
+  aligned with the parent repository.
+- **`pyproject.toml` URLs** (`Homepage`, `Documentation`, `Source`,
+  `Issues`, `Changelog`) all point at the new repository.
+
+### Removed
+
+- Dead `_proto` stubs that had never been imported by SDK production code
+  (`tipsy/config/v1/internal_pb2*`, `tipsy/abtest/v1/internal_pb2*`,
+  `tipsy/audit/v1/audit_pb2*`). The 3 backend-private protos that
+  produced them stayed in the private `tipsy-ab-config` repo; only the
+  2 SDK-facing protos (`config.proto` + `abtest.proto`) ship here.
+
+### Migration
+
+- Old installs pinned to
+  `git+https://${GH_PAT}@github.com/Lightspeed-Intelligence/tipsy-ab-config.git@python-sdk/v0.2.0#subdirectory=sdk/python`
+  continue to resolve via the legacy tag preserved on the ab-config repo.
+  Starting with v0.3.0, install from the new public repo (no PAT).
+
+## [0.2.0] - 2025-11-21
+
+### Added
+
+- Optional `trace_id` keyword argument on `Client.get_experiment_result`,
+  `Client.new_abtest_context`, and `Client.abtest_scope`. When omitted or
+  empty the SDK generates a fresh UUID v4 (36-char with dashes) and forwards
+  it on the wire, so every request is trace-identifiable end-to-end without
+  caller changes.
+- `AbtestContext.trace_id` attribute, set at construction and reused by every
+  per-namespace `GetExperimentResult` RPC the context issues.
+- `AbtestMiddleware` now resolves the per-request `trace_id` from the
+  inbound `X-Trace-Id` header, falling back to `X-Request-Id`, and finally
+  to a freshly generated UUID. The resolved id is forwarded onto the
+  AbtestContext so it propagates to every downstream RPC.
+- Background `PullAll` and `Subscribe` RPCs now generate a fresh `trace_id`
+  per call and log it at `debug` level before issuing the RPC, mirroring the
+  Go SDK pattern.
+
+## [0.1.0] - 2026-06-16
+
+### Added
+
+- 首发 release of the Tipsy AB-config Python SDK.
+  - `Client` / `init` / `Config` — process-local config cache populated by a
+    startup `PullAll`, long-lived server-streaming `Subscribe`, and a
+    10-second fallback `PullAll` safety net.
+  - `Client.get_config_static` — pure cache read, no abtest, no exposure.
+  - `Client.get_config` — abtest-aware lookup with the per-request
+    `AbtestContext` (abtest hit > full release fallback); emits exposure
+    events asynchronously with a 5-minute per-process dedup window.
+  - `Client.new_abtest_context` — eagerly pre-fetches the prefetch
+    namespace; other namespaces lazy-loaded and memoised at-most-once on
+    first dynamic `get_config`.
+  - `Client.get_experiment_result` — exposes the full wire request
+    (`namespace`, `user_info`, `layer_ids`, `type`, `display`).
+  - `AbtestMiddleware` — FastAPI / Starlette middleware that binds
+    `AbtestContext` into `contextvars` per request.
+  - gRPC + HTTP transports (gRPC is the default; HTTP via `tipsy-ab-config[http]`).
+  - Python 3.10 / 3.11 / 3.12 / 3.13 CI matrix.
+  - Shipped proto bindings under `tipsy_ab_config/_proto/**` regenerated
+    with `grpcio-tools==1.66.2` (protobuf 5.x major emission; runtime
+    requires `protobuf>=5.29,<7`).

tipsy_ab_config-0.15.0/LICENSE

@@ -0,0 +1,21 @@
+The MIT License (MIT)
+
+Copyright (c) 2026 Lightspeed-Intelligence
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

tipsy_ab_config-0.15.0/MANIFEST.in

@@ -0,0 +1,12 @@
+include README.md
+include LICENSE
+include CHANGELOG.md
+include tipsy_ab_config/py.typed
+recursive-include tipsy_ab_config/_proto *.py
+global-exclude __pycache__
+global-exclude *.pyc
+global-exclude *.pyo
+prune .venv
+prune .pytest_cache
+prune tipsy_ab_config.egg-info
+prune tests/__pycache__