blake3_ffi 1.1.1

blake3_ffi: ^1.1.1 copied to clipboard

Documentation and packaging only. The API, the native code and the digests are untouched.

• The benchmark tables in the README are written by a tool now, not copied by hand. The section claimed "nothing here is transcribed by hand" while both tables were hand-copied out of a benchmark run, and the test guarding them looked at one of the seven rows. tool/readme_tables.dart writes them from doc/benchmark.json, tool/benchmark_svg.dart draws the chart through the same helpers so the two cannot round one measurement differently, and test/published_numbers_test.dart now compares the whole generated block and looks for the headline figures where they are still typed: the README prose, the pubspec description and caption, and the chart's SVG. Six deliberate edits, one published figure each and two of them in rows the old test never read, all fail it. The words around those figures are not checked, and neither is doc/benchmark.png.
• Input sizes are labelled in binary units. A 1,048,576-byte buffer was called "1 MB" one line below "MB means 1,000,000 bytes"; the rows are 1, 16 and 64 MiB, and throughput stays in decimal MB/s.
• The method line said each figure was "the fastest of five batches in each of three orderings". It is the fastest batch of the fastest placement, which is a stronger claim than the sentence made; the README and the chart both say that now.
• "Gave up web and mobile" understated the cost of the native path: there is no Flutter support on any platform yet, desktop included. The opening paragraph and the advice in "Which BLAKE3 package" say so.
• A build/ directory would have been published. The root .pubignore shadowed the root .gitignore for the tree, so nothing that file excludes was being kept out of the archive; a dry-run with a build/ directory present listed its files. 1.1.0 shipped clean only because no such directory existed when it was published. The rule moved to doc/.pubignore, which keeps the blog images out without disabling anything else.
• description is a literal block rather than a folded one, matching the screenshot caption. A folded block joins its lines with a space, which is how a wrapped "One-shot" once reached pub.dev as "One- shot".

• Add blake3Hash, BLAKE3 as a package:crypto Hash. blake3(bytes) was already the shortest way to hash here, so this is not for that. It is for the call sites you do not own: a checksum helper, a content-addressed cache or an Hmac that takes a Hash parameter can now be handed BLAKE3 without changing shape. blockSize is 64, matching SHA-256, so Hmac pads correctly; startChunkedConversion streams through a Blake3Hasher and releases it on close; and one-shot convert is overridden to skip the sink. A test pins the two paths against blake3() — a Hash view that drifted from the direct call would quietly be a different algorithm. Nine tests cover it, and each of five deliberate defects in the new code turns the suite red.
• crypto moves from a dev dependency to a dependency, since the Hash type is now part of this package's surface. It is pure Dart and already resolved in most trees.

• Add example/README.md for pub.dev's Example tab (it was empty). It walks through the four things the package does — one-shot hash, streaming file hash, keyed MAC, key derivation — with the example's real output. Docs only.

• Fix the bulk benchmark's 1 MB row, which was a measurement artifact rather than a real number. bench/bench.dart hashed a flat four iterations per size, so the 1 MB case timed only a few megabytes — too little to average out scheduling jitter, and its throughput swung roughly 2x from run to run while 16 MB and 64 MB were steady. The benchmark now scales the iteration count so every size hashes a comparable total volume, and 1 MB lands near 14x like the larger sizes, not the 7.2x the README used to show. Benchmark and docs only; no library change.

First stable release. The public API is now committed to semantic versioning: the hashing functions, Blake3Hasher, and the blake3KeyLength / blake3OutLength constants will not change in a breaking way without a 2.0.0. The package covers all three BLAKE3 modes (hash, keyed, and key derivation), extendable output through finalize(outputLength:, seek:), and one-shot, incremental, and streaming hashing, all over the official BLAKE3 C sources compiled from a build hook.

• blake3HexStream remains as a deprecated alias for blake3StreamHex (the 0.4.0 rename) and now points at removal in 2.0.0 rather than 1.0.0, so code written against 0.4.0 keeps working across the 1.0.0 boundary.

• Rename blake3HexStream to blake3StreamHex so the hex variant is a Hex suffix on its base name, matching every other pair in the API: blake3 and blake3Hex, blake3Keyed and blake3KeyedHex, blake3DeriveKey and blake3DeriveKeyHex. blake3HexStream still works as a deprecated alias that forwards to blake3StreamHex, and will be removed in 1.0.0.

• Declare platforms: {linux, macos, windows} in pubspec.yaml. Flutter's build-hooks support isn't stable yet and this package has no Android/iOS build today; pub.dev had inferred support for all five platforms from static analysis alone with no declaration to override it.

• Blake3Hasher.update now takes a List<int> instead of a Uint8List. Chunks from a Stream<List<int>> (a file's openRead(), a socket) arrive as plain List<int>, so the documented manual streaming loop did not compile without wrapping every chunk in Uint8List.fromList. A Uint8List still passes through with no copy; any other list is copied once, the same coercion blake3Stream already does. Source-compatible: Uint8List is a List<int>.

• Widen the native-toolchain constraints so the package can be installed in a Flutter app at all. hooks 2.1.0 and native_toolchain_c 0.19.3 raised their meta floor to ^1.19.0, and Flutter's SDK pins meta to 1.17.0, so flutter pub add failed at version solving with "flutter from sdk is incompatible". Allowing hooks >=2.0.2 and native_toolchain_c >=0.19.2 lets the solver pick a version that works with the pinned meta, while a pure-Dart project still resolves to the newest. No API or behaviour change.

• Declare the benchmark chart in pubspec.yaml so pub.dev renders it on the package page. The chart was already in the repository and the README, but pub.dev shows only what the screenshots: field points at, so the page a reader lands on from search opened with text where the measurement should have been.

• Add blake3Stream and blake3HexStream: one call to hash a Stream<List<int>> (a file's openRead(), an upload, any byte stream) as it arrives, without holding the whole input in memory. Each drives a Blake3Hasher internally and disposes it when the stream ends. This is the memory-safe way to hash something too large to load at once, and unlike a SHA-256 stream from package:crypto it runs at BLAKE3's throughput. The outputLength (XOF) argument carries through.

• Add hex variants for the raw-output paths: blake3KeyedHex, blake3DeriveKeyHex, and Blake3Hasher.finalizeHex. Keyed, derived, and streamed digests now format to hex the way blake3Hex already does, so callers stop re-implementing it (the example used to).

• Docs: sharpen the pub.dev description to lead with the value and the terms people search.

Initial release, vendoring the official BLAKE3 C implementation 1.8.5.

• One-shot hashing: blake3, blake3Hex.
• Streaming hashing: Blake3Hasher with update, finalize, reset, and dispose.
• Keyed hashing (MAC/PRF): blake3Keyed, Blake3Hasher.keyed.
• Key derivation (KDF): blake3DeriveKey, Blake3Hasher.deriveKey.
• Extendable output (XOF) via outputLength and seek.
• Native code builds automatically via Dart build hooks (Dart 3.10+); no manual native setup. arm64 uses the NEON kernel; other architectures use the portable C kernel.
• Verified against the official BLAKE3 test vectors (default, keyed, and derive-key modes, extended output, and streaming).