microsoft · martintmk · Sep 11, 2026 · Sep 1, 2026 · Sep 1, 2026 · Sep 1, 2026
diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
@@ -389,8 +389,10 @@ jobs:
         run: cargo +${{ env.RUST_NIGHTLY }} careful nextest run --all-features --workspace ${{ needs.delta.outputs.exclude_not_affected }} --color always --target ${{ matrix.target }} --no-tests=pass
       - name: Miri (stacked borrows)
         if: success() || failure()
-        # Most excluded crates generate code rather than run it, so Miri has no
-        # aliasing or provenance violations to find in them:
+        # Excluded crates fall into two groups.
+        #
+        # Crates in the first group generate code rather than run it, so Miri has
+        # no aliasing or provenance violations to find in them:
         #
         #   * the `*_macros_impl` crates are proc-macro implementations whose
         #     tests are snapshot comparisons of generated token streams;
@@ -408,11 +410,24 @@ jobs:
         # Interpreting them anyway cost ~75 min of the ~3h16m step
         # (`multitude_macros_impl` alone ran 30 tests in 45 min).
         #
+        # The second group is `compressors`, whose backends Miri cannot all run:
+        # `zstd-safe` binds the native zstd library and Miri cannot call foreign
+        # functions, while `flate2`'s `zlib-rs` backend trips Stacked Borrows
+        # whenever a deflate or inflate stream is dropped
+        # (https://github.com/trifectatechfoundation/zlib-rs/issues/491). Brotli
+        # is pure Rust and would run; excluding the whole crate rather than
+        # maintaining a brotli-only Miri configuration is a separate decision,
+        # taken because gating every other format's tests on `cfg(miri)` costs
+        # more than the remaining coverage is worth. The crate also carries
+        # `package.metadata.anvil.miri.exclude`, which is what the `anvil-miri`
+        # recipe honours; this job builds its own command line, so it needs the
+        # exclusion spelled out here too.
+        #
         # Note `rest_over_grpc` is deliberately NOT excluded: its build half is
         # already `#[cfg_attr(miri, ignore)]`, and the tests that do run cover the
         # runtime transcoding path over `bytes`/`hyper`, which is exactly the kind
         # of buffer handling Miri needs to check.
-        run: cargo +${{ env.RUST_NIGHTLY }} miri test --all-features --workspace ${{ needs.delta.outputs.exclude_not_affected }} --exclude data_privacy_macros_impl --exclude fundle_macros_impl --exclude internity_macros_impl --exclude metabench --exclude metabench_macros_impl --exclude multitude_macros_impl --exclude observed_macros_impl --exclude ohno_macros_impl --exclude routerama_build --exclude templated_uri_macros_impl --exclude thread_aware_macros_impl --lib --tests
+        run: cargo +${{ env.RUST_NIGHTLY }} miri test --all-features --workspace ${{ needs.delta.outputs.exclude_not_affected }} --exclude compressors --exclude data_privacy_macros_impl --exclude fundle_macros_impl --exclude internity_macros_impl --exclude metabench --exclude metabench_macros_impl --exclude multitude_macros_impl --exclude observed_macros_impl --exclude ohno_macros_impl --exclude routerama_build --exclude templated_uri_macros_impl --exclude thread_aware_macros_impl --lib --tests
 
   model-checking:
     if: (github.event_name == 'pull_request' || github.event_name == 'merge_group') && needs.delta.outputs.skip != 'true'

diff --git a/.spelling b/.spelling
@@ -716,6 +716,21 @@ userland
 WebIO
 WebSocket
 wiremock
+Zlib
+Zstandard
+Zstd's
+brotli
+brotli's
+checksum
+codec's
+decompressor
+decompressors
+flate2
+PNG
+resumable
+selectable
+semver
+WOFF
 reentrantly
 rebalancing
 unsizing
@@ -867,7 +882,13 @@ hardcoding
 Idempotence
 symlinks
 junctions
+zlib
+zstd's
+unkeyed
+incompressible
+decompressor's
 recurse
+flate's
 Cancelling
 TUI
 mutex's

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -12,6 +12,7 @@ Please see each crate's change log below:
 - [`cachet_memory`](./crates/cachet_memory/CHANGELOG.md)
 - [`cachet_service`](./crates/cachet_service/CHANGELOG.md)
 - [`cachet_tier`](./crates/cachet_tier/CHANGELOG.md)
+- [`compressors`](./crates/compressors/CHANGELOG.md)
 - [`data_privacy`](./crates/data_privacy/CHANGELOG.md)
 - [`data_privacy_macros`](./crates/data_privacy_macros/CHANGELOG.md)
 - [`data_privacy_macros_impl`](./crates/data_privacy_macros_impl/CHANGELOG.md)

diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Cargo.toml b/Cargo.toml
@@ -79,6 +79,7 @@ azure_storage_blob = { version = "1.0.0", default-features = false }
 backtrace = { version = "0.3.76", default-features = false }
 base64 = { version = "0.23.0", default-features = false, features = ["alloc"] }
 bolero = { version = "0.13.4", default-features = false }
+brotli = { version = "8.0.4", default-features = false }
 bumpalo = { version = "3.20.2", default-features = false }
 bytemuck = { version = "1.25.0", default-features = false }
 bytes = { version = "1.11.1", default-features = false }
@@ -89,12 +90,14 @@ cachet_memory = { path = "crates/cachet_memory", default-features = false, versi
 cachet_service = { path = "crates/cachet_service", default-features = false, version = "0.5.0" }
 cachet_tier = { path = "crates/cachet_tier", default-features = false, version = "0.5.0" }
 cc = { version = "1.2", default-features = false }
+cfg_aliases = { version = "0.2.2", default-features = false }
 chrono = { version = "0.4.40", default-features = false }
 chrono-tz = { version = "0.10.4", default-features = false }
 chumsky = { version = "0.13.0", default-features = false }
 clap = { version = "4.6.4", default-features = false }
 # The latest command-group release still uses nix 0.27 on Unix; accept the duplicate until upstream updates.
 command-group = { version = "5.0.1", default-features = false }
+compressors = { path = "crates/compressors", default-features = false, version = "0.1.0" }
 const-hex = { version = "1.15.0", default-features = false }
 criterion = { version = "0.8.2", default-features = false }
 crossterm = { version = "0.29.0", default-features = false }
@@ -118,6 +121,7 @@ fetch_options = { path = "crates/fetch_options", default-features = false, versi
 fetch_tls = { path = "crates/fetch_tls", default-features = false, version = "0.4.0" }
 fetch_winhttp = { path = "crates/fetch_winhttp", default-features = false, version = "0.1.1" }
 fetch_winhttp_impl = { path = "crates/fetch_winhttp_impl", default-features = false, version = "0.1.1" }
+flate2 = { version = "1.1.10", default-features = false }
 foldhash = { version = "0.2.0", default-features = false }
 fundle = { path = "crates/fundle", default-features = false, version = "0.3.5" }
 fundle_macros = { path = "crates/fundle_macros", default-features = false, version = "0.3.5" }
@@ -268,6 +272,7 @@ time = { version = "0.3.47", default-features = false }
 tokio = { version = "1.48.0", default-features = false }
 tokio-native-tls = { version = "0.3.1", default-features = false }
 tokio-rustls = { version = "0.26.4", default-features = false }
+tokio-stream = { version = "0.1.19", default-features = false }
 tonic = { version = "0.14.2", default-features = false }
 tonic-prost = { version = "0.14.2", default-features = false }
 tonic-prost-build = { version = "0.14.2", default-features = false }
@@ -291,6 +296,7 @@ windows-sys = { version = "0.61.2", default-features = false }
 wiremock = { version = "0.6.5", default-features = false }
 xxhash-rust = { version = "0.8.15", default-features = false }
 zerocopy = { version = "0.8.26", default-features = false }
+zstd-safe = { version = "7.2.4", default-features = false }
 
 # >>> anvil-managed: anvil-workspace-lints
 [workspace.lints]

diff --git a/README.md b/README.md
@@ -36,6 +36,7 @@ These are the primary crates built out of this repo:
 - [`cachet_memory`](./crates/cachet_memory/README.md) - In-memory cache tier backed by Moka for the cachet caching library.
 - [`cachet_service`](./crates/cachet_service/README.md) - Layered service integration for the cachet caching library.
 - [`cachet_tier`](./crates/cachet_tier/README.md) - Core cache tier trait and abstractions for building cache backends.
+- [`compressors`](./crates/compressors/README.md) - Streaming compression and decompression over bytesbuf byte sequences.
 - [`data_privacy`](./crates/data_privacy/README.md) - Mechanisms to classify, manipulate, and redact sensitive data.
 - [`fetch`](./crates/fetch/README.md) - "Universal, composable and resilient HTTP client."
 - [`fetch_azure`](./crates/fetch_azure/README.md) - Azure SDK HTTP transport backed by the fetch HTTP client.

diff --git a/crates/compressors/CHANGELOG.md b/crates/compressors/CHANGELOG.md
@@ -0,0 +1 @@
+# Changelog
diff --git a/crates/compressors/Cargo.toml b/crates/compressors/Cargo.toml
@@ -0,0 +1,104 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+[package]
+name = "compressors"
+version = "0.1.0"
+description = "Streaming compression and decompression over bytesbuf byte sequences"
+readme = "README.md"
+keywords = ["compression", "gzip", "streaming", "zero-copy"]
+categories = ["compression", "network-programming"]
+edition = { workspace = true }
+rust-version = { workspace = true }
+authors = { workspace = true }
+license = { workspace = true }
+homepage = { workspace = true }
+include = { workspace = true }
+repository = "https://github.com/microsoft/oxidizer/tree/main/crates/compressors"
+
+# Miri cannot run all of this crate's backends. `zstd-safe` binds the native zstd library, and Miri
+# cannot call foreign functions; `flate2`'s `zlib-rs` backend has an open soundness bug that trips
+# Stacked Borrows whenever a Deflate or Inflate is dropped
+# (https://github.com/trifectatechfoundation/zlib-rs/issues/491). Brotli is pure Rust and would run;
+# excluding the whole crate rather than maintaining a brotli-only Miri configuration is a separate
+# decision, taken because the per-format `cfg(miri)` scaffolding the rest would need costs more than
+# the remaining coverage is worth.
+[package.metadata.anvil.miri]
+exclude = true
+
+[package.metadata.docs.rs]
+all-features = true
+
+# `cfg_aliases` is used only by `build.rs`. cargo-machete does not scan build
+# scripts, so it reports a false positive; ignore it here.
+[package.metadata.cargo-machete]
+ignored = ["cfg_aliases"]
+
+[package.metadata.cargo_check_external_types]
+allowed_external_types = [
+    "bytesbuf::*",
+    "futures_core::stream::Stream",
+    "recoverable::*",
+    "thread_aware_core::thread_aware::ThreadAware",
+]
+
+[features]
+brotli = ["dep:brotli"]
+# No format is enabled by default: a dependent names the ones it actually speaks, so a build that
+# only needs brotli never compiles flate2, and vice versa. A build with no format at all still gets
+# the shared contract and builders, which is what a crate that only re-exports them needs.
+default = []
+deflate = ["dep:flate2"]
+futures-stream = ["dep:futures-core", "dep:pin-project-lite"]
+gzip = ["dep:flate2"]
+zlib = ["dep:flate2"]
+zstd = ["dep:zstd-safe"]
+
+[dependencies]
+brotli = { workspace = true, features = ["std"], optional = true }
+bytesbuf = { workspace = true, features = ["std"] }
+# `zlib-rs` rather than flate2's default `miniz_oxide`: it is the fastest of flate2's backends and,
+# unlike the `zlib`/`zlib-ng` bindings, is pure Rust, so the deflate family stays free of C on a path
+# that parses untrusted input. The cost is recorded above -- it is what keeps this crate out of Miri.
+flate2 = { workspace = true, features = ["zlib-rs"], optional = true }
+futures-core = { workspace = true, optional = true }
+pin-project-lite = { workspace = true, optional = true }
+recoverable = { workspace = true }
+thread_aware = { workspace = true, features = ["std"] }
+zstd-safe = { workspace = true, features = ["std"], optional = true }
+
+[build-dependencies]
+cfg_aliases = { workspace = true }
+
+[dev-dependencies]
+bolero = { workspace = true, features = ["std"] }
+brotli = { workspace = true, features = ["std"] }
+bytesbuf = { workspace = true, features = ["std", "test-util"] }
+criterion = { workspace = true }
+flate2 = { workspace = true, features = ["zlib-rs"] }
+futures = { workspace = true, features = ["executor"] }
+futures-core = { workspace = true }
+metabench = { workspace = true }
+mutants = { workspace = true }
+pin-project-lite = { workspace = true }
+thread_aware = { path = "../thread_aware", default-features = false, features = ["std", "test-utils"] }
+tick = { path = "../tick", features = ["tokio", "test-util"] }
+tokio = { workspace = true, features = ["rt-multi-thread", "macros", "time", "sync"] }
+tokio-stream = { workspace = true }
+zstd-safe = { workspace = true, features = ["std"] }
+
+[[example]]
+name = "round_trip"
+required-features = ["brotli", "deflate", "gzip", "zlib", "zstd"]
+
+[[example]]
+name = "tokio_stream"
+required-features = ["futures-stream", "gzip"]
+
+[[bench]]
+harness = false
+name = "compressors_codec"
+required-features = ["brotli", "deflate", "futures-stream", "gzip", "zlib", "zstd"]
+
+[lints]
+workspace = true