GordonYuanyc · Oct 20, 2026 · Oct 20, 2026 · Oct 20, 2026 · Oct 20, 2026 · Oct 20, 2026
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -111,31 +111,3 @@ jobs:
 
       - name: Run tests
         run: cargo test --features experimental --locked --verbose
-
-  proto-vendored-up-to-date:
-    name: Vendored proto code is up to date
-    runs-on: ubuntu-latest
-
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Set up Rust
-        uses: dtolnay/rust-toolchain@stable
-
-      - name: Cache Cargo dependencies and build artifacts
-        uses: Swatinem/rust-cache@v2
-        with:
-          workspaces: tools/gen-proto -> tools/gen-proto/target
-
-      - name: Regenerate vendored proto code
-        run: cargo run --manifest-path tools/gen-proto/Cargo.toml --locked
-
-      - name: Verify src/proto/generated/ has no drift
-        run: |
-          if ! git diff --exit-code -- src/proto/generated; then
-            echo "::error::src/proto/generated/ is out of date. Run:" >&2
-            echo "  cargo run --manifest-path tools/gen-proto/Cargo.toml" >&2
-            echo "and commit the result." >&2
-            exit 1
-          fi
diff --git a/.github/workflows/docs.yml b/.github/workflows/docs.yml
@@ -5,8 +5,6 @@ on:
     branches: [ "main" ]
     paths:
       - "src/**"
-      - "proto/**"
-      - "tools/gen-proto/**"
       - "Cargo.toml"
       - "Cargo.lock"
       - "README.md"
@@ -16,8 +14,6 @@ on:
     branches: [ "main" ]
     paths:
       - "src/**"
-      - "proto/**"
-      - "tools/gen-proto/**"
       - "Cargo.toml"
       - "Cargo.lock"
       - "README.md"

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -331,6 +331,21 @@ signals a backwards-compatible change.
 
 ### Removed
 
+- **Protobuf** (breaking). Module `proto`, the `.proto` sources,
+  `tools/gen-proto` and the `prost`/`bytes` dependencies; with them the KLL
+  value-offset sample encoding.
+- **Legacy MessagePack** (breaking). Module `message_pack_format` with
+  `portable`, `native` and `MessagePackCodec`: the wrapper types
+  `CountMinSketch`, `CountSketch`, `CountMinSketchWithHeap`,
+  `CountSketchWithHeap`, `KllSketch`, `HllSketch`, `DdSketch`,
+  `HydraKllSketch`, their deltas and helpers, and the `xxhash-rust`
+  dependency. `SetAggregator` and `SetAggregatorDelta` move to
+  `sketches::set_aggregator`; the ASAPv1 framing is private.
+- **serde on sketch types** (breaking). Sketches, their storage and their
+  inputs no longer implement `Serialize` / `Deserialize`; ASAPv1 is the one
+  serialized form. `CountMinHll` and `MicroCm` lose
+  `serialize_to_bytes` / `deserialize_from_bytes`, and the
+  `serde-big-array` dependency goes.
 - **Dead and duplicate APIs** (breaking). `Kll::wire_k` (use `k()`);
   `PRECOMPUTED_SAMPLE_RATE_1PERCENT` and module `common::precompute_sample2`;
   module `common::hashspec` (`HashSpec`, `derive_index`, `derive_sign`,

diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -19,14 +19,10 @@ all-features = true
 [dependencies]
 rand = "0.9.3"
 serde = { version = "1.0", features = ["derive"] }
-serde-big-array = "0.5"
 serde_bytes = "0.11"
 rmp-serde = "1.3.0"
 twox-hash = "2.1.2"
 smallvec = "1.13.2"
-prost = "0.13"
-bytes = "1"
-xxhash-rust = { version = "0.8", features = ["xxh32"] }
 core_affinity = { version = "0.8", optional = true }
 crossbeam-channel = { version = "0.5", optional = true }
 
@@ -39,11 +35,6 @@ default = []
 experimental = []
 octo-runtime = ["core_affinity", "crossbeam-channel"]
 
-# Note: there are intentionally no `[build-dependencies]`. The prost output
-# under `src/proto/generated/` is checked in and refreshed by the maintainer
-# tool at `tools/gen-proto/`. This keeps the published crate pure-Rust with
-# no requirement on `protoc` for downstream consumers.
-
 [lib]
 crate-type = ["rlib"]
 name = "asap_sketchlib"

diff --git a/README.md b/README.md
@@ -220,31 +220,6 @@ cargo build --features experimental
 cargo test --features "experimental octo-runtime"
 ```
 
-## Protobuf code generation
-
-`asap_sketchlib` is a pure-Rust crate: building it does **not** require
-`protoc` or any build script. The Rust types generated from
-`proto/**/*.proto` are vendored into `src/proto/generated/` and refreshed
-manually by maintainers using the in-repo tool at `tools/gen-proto/`.
-
-Downstream users can therefore simply add the crate to their `Cargo.toml` and
-build it like any other pure-Rust dependency.
-
-### For maintainers
-
-After editing any `.proto` file, regenerate the vendored output and commit
-the result:
-
-```bash
-cargo run --manifest-path tools/gen-proto/Cargo.toml
-git add src/proto/generated
-```
-
-The tool uses `prost-build` together with the `protoc-bin-vendored` binary,
-so no system `protoc` is required to regenerate either. CI rejects any pull
-request whose committed `src/proto/generated/` does not match the result of
-running this command on its current `.proto` sources.
-
 ## FAQ
 
 ### When is Apache DataSketches a better fit?

diff --git a/docs/api/api_bloom.md b/docs/api/api_bloom.md
@@ -120,14 +120,6 @@ word count that disagrees with the declared dimensions, and any bit set in a
 row's trailing padding past `cols` — unreachable by `contains`, but counted by
 `count_ones`, so it would skew `fill_ratio` and `estimated_fpp` alone.
 
-Independently of ASAPv1, `Bloom` is plain `Serialize`/`Deserialize`; that form is
-`{ bits, inserted, mode }`, where `mode` is `"regular"` or `"fast"`. Decoding
-into the other hash path fails rather than producing a filter that reports its
-own members absent, and so does a row count past `BLOOM_MAX_SLICES`. `BitMatrix`
-carries `{ words, rows, cols }` only — the word stride and the fold masks are
-recomputed on decode, and a payload whose word count disagrees with its
-dimensions is rejected there rather than panicking later.
-
 ## Examples
 
 ```rust

diff --git a/docs/api/api_coco_sketch.md b/docs/api/api_coco_sketch.md
@@ -113,7 +113,7 @@ that holds nothing. Both methods require `H: HashProfile`.
 
 A zero dimension, a table whose bucket count disagrees with `rows`/`cols`, a payload
 length that disagrees with the declared geometry, and a `nil` key carrying a
-non-zero value are all rejected on both sides. `CocoSketch` also derives serde.
+non-zero value are all rejected on both sides.
 
 Row `i` hashes at seed index `i`, so the wire covers `1 <= d <= 20`
 (`MATRIX_MAX_ROWS`, the seed list length): past that, row `i` and row `i + 20`

diff --git a/docs/api/api_common_heap.md b/docs/api/api_common_heap.md
@@ -52,8 +52,8 @@ No dedicated merge method on `TopKHeap`; reconciliation is sketch-specific.
 
 ## Serialization
 
-No dedicated byte API helpers. Derived serde carries the heap and its capacity;
-the key index is derived data and is rebuilt on load rather than stored.
+None of its own; a sketch holding a `TopKHeap` carries its entries in its
+ASAPv1 payload.
 
 ## Examples
 

diff --git a/docs/api/api_common_input.md b/docs/api/api_common_input.md
@@ -61,8 +61,8 @@ fn merge(&mut self, other: &HydraCounter) -> Result<(), String>
 
 ## Serialization
 
-- `DataInput`, `OwnedDataInput`, `HeapEntry`, and `HydraCounter` derive serde traits.
-- `HydraQuery` is query-only and does not derive serde in current code.
+None of these types implements serde's `Serialize` / `Deserialize`; a sketch
+holding them serializes through its own `to_asapv1`.
 
 ## Variants (Core)
 

diff --git a/docs/api/api_count_sketch.md b/docs/api/api_count_sketch.md
@@ -111,9 +111,7 @@ as the state serialization: the changed cells as strictly ascending row-major
 indices `row * cols + col`, below `rows * cols`, and the non-zero signed value
 added to each. `new` and the decoder refuse anything else. `merge_delta` adds
 the delta to the counters; a delta of other dimensions or admission rate, or an
-integer add that overflows, fails and leaves the sketch unchanged. The crate
-root's `CountSketchDelta` is the legacy portable type; this one is
-`asap_sketchlib::sketches::CountSketchDelta`.
+integer add that overflows, fails and leaves the sketch unchanged.
 
 ## Admission rate
 

diff --git a/docs/api/api_countminsketch_hll.md b/docs/api/api_countminsketch_hll.md
@@ -79,13 +79,7 @@ returns an error otherwise, leaving `self` untouched.
 
 ## Serialization
 
-```rust
-fn serialize_to_bytes(&self) -> Result<Vec<u8>, RmpEncodeError>
-fn deserialize_from_bytes(bytes: &[u8]) -> Result<Self, RmpDecodeError>
-```
-
-Plain MessagePack, **not** an ASAPv1 envelope: no `kind_id` is allocated for
-this sketch while it is experimental.
+None: no `kind_id` is allocated for this sketch while it is experimental.
 
 ## Accuracy
 

diff --git a/docs/api/api_ddsketch.md b/docs/api/api_ddsketch.md
@@ -70,10 +70,6 @@ past `i32`, a non-zero offset on an empty store, counts that overflow the total,
 and scalars inconsistent with that total; `to_asapv1` enforces the same
 rules, so the format never emits bytes it would refuse to read back.
 
-Independently of ASAPv1, `DdSketch` is plain `Serialize`/`Deserialize` for use
-where it is nested in another type (`EhBucketSketch`); that form skips the four
-running scalars.
-
 ## Deltas
 
 ```rust
@@ -139,17 +135,8 @@ assert!(p50 >= 1.0);
 `rank = ceil(q * n)` (1-based), so it answers the order statistic
 `sorted[ceil(q*n) - 1]`. `q <= 0` and `q >= 1` short-circuit to the exactly
 retained minimum and maximum rather than to a bucket representative, so the
-endpoints carry zero error.
-
-This differs from the portable wire twin
-`message_pack_format::portable::ddsketch::DdSketch::quantile`, which uses the
-lower-quantile convention `floor(q * (n - 1))` of the DDSketch paper and of the
-Go reference implementation, and which retains no min/max scalars — its
-endpoints are ordinary bucket representatives, accurate to `alpha` like any
-other rank. The two agree at most `q` and diverge at small `n` or ragged `q`
-(at `n = 3, q = 0.4` this type answers `sorted[1]` and the portable one
-`sorted[0]`). Both conventions are deliberate and are pinned by
-`ddsketch_core_and_portable_answer_different_order_statistics`.
+endpoints carry zero error. The convention is pinned by
+`ddsketch_answers_the_ceil_nearest_rank_order_statistic`.
 
 ## Caveats
 

diff --git a/docs/api/api_eh_bucket_sketch.md b/docs/api/api_eh_bucket_sketch.md
@@ -83,7 +83,7 @@ and `0x01 0x03` (HIP) are rejected, and `Kll` is compact, so `0x06 0x01`
 (dynamic) is rejected. The dispatch is the same in every build: a decoder built
 without `experimental` rejects `0x0d 0x00` with an error naming the variant and
 the feature, and its encoder can never emit it.
-An unrecognized kind_id is rejected. `EhBucketSketch` also derives serde.
+An unrecognized kind_id is rejected.
 
 ## Examples
 

diff --git a/docs/api/api_elastic_sketch.md b/docs/api/api_elastic_sketch.md
@@ -343,7 +343,7 @@ both sides past it.
 A zero dimension, a `heavy_buckets` disagreeing with the heavy table, a free bucket
 that still names a flow, a payload length that disagrees with the declared
 geometry, and a `nil` flow id whose `vote_pos` is non-zero are all rejected on
-both sides. `ElasticSketch` also derives serde.
+both sides.
 
 ## Examples
 

diff --git a/docs/api/api_hydra.md b/docs/api/api_hydra.md
@@ -75,9 +75,7 @@ A grid mixing counter variants, UnivMon cells mixing key variants, a cell whose
 geometry differs from the prototype's, a `prototype` holding data, a
 declared grid the storage does not match, a row count past `MATRIX_MAX_ROWS`,
 and a payload length that disagrees with the declared geometry are all rejected
-on both sides. `Hydra` also derives
-serde, which is the nested codec used when a Hydra is embedded in a larger
-serde value.
+on both sides.
 
 ## Examples