## Summary
An honest, tiered cross-library benchmark — and the optimization pass it triggered.
### Performance (wickra-core, outputs unchanged)
Profiling against the other Rust TA crates exposed real inefficiencies. Each
benchmarked indicator is now **5–79% faster** in both streaming and batch:
- **SMA, Bollinger**: flat `Box<[f64]>` ring buffers replace `VecDeque` (−69…79%).
- **RSI**: `100·ag/(ag+al)` collapses three divisions into one; Wilder smoothing
hoists `1/period` out of the hot path (−46%).
- **ATR**: reciprocal hoisted (−42%).
- **EMA/RSI/ATR**: per-tick `Option<f64>` hot state → bare `f64` + ready flag.
Net result vs `kand`: Wickra now wins **RSI, Bollinger and ATR** (streaming), and
ties `ta-rs` on SMA — up from losing every indicator 1.5–6× before.
### Benchmark harness
New `crates/wickra-bench` (publish=false): a Criterion benchmark comparing Wickra
against `kand`, `ta-rs` and `yata` on an identical BTCUSDT candle series, in
streaming and batch modes. Peer APIs were verified against their source, not
guessed. Wired into the nightly `cross-library-bench` workflow as a separate job.
### Honest README
The benchmark section is rewritten into three layered tables (Rust core vs Rust
crates; Python vs the Python ecosystem) that **show the losses as well as the
wins**. The "only library that combines…" claim is gone; the new framing is
breadth + multi-language reach + the deliberate safety trade-off that costs raw
speed. Added an origin/why-slower rationale and a star CTA.
### Python benchmark
Added `tulipy` runners and expanded per-tick streaming coverage to SMA/EMA/RSI/
MACD/Bollinger. `bench.in`/`bench.txt` now lock `TA-Lib` + `tulipy` (hash-pinned);
`pandas-ta` stays out (it requires Python ≥ 3.12, the bench runs on 3.11).
### Notes
- TA-Lib/tulipy numbers in the README Python table are marked ⧗ — they are
produced by the CI Linux job (C extensions don't build cleanly on every
desktop), not measured locally.
- The matching `wickra-docs` prose update is committed separately and will be
pushed with the release, per the docs-don't-lead-the-registries rule.
Verified locally: `cargo fmt`, `cargo test --workspace --all-features` (3413 core
+ bindings), `cargo clippy --workspace --all-targets --all-features -D warnings`,
Node build + 498 tests, and pytest all green.
* ci: pass ref context through env in release tag step
zizmor flagged the "Resolve target tag" step in release.yml for
template-injection: github.event_name / github.ref / github.ref_name
were interpolated directly into the shell script. On a tag push the tag
name is attacker-influenceable, so a crafted tag could inject commands.
Move all three context values into the step env and reference them as
shell variables instead. Verified with zizmor 1.16.3: template-injection
findings on release.yml drop from 2 to 0.
* ci: accept release.yml build caches via zizmor config
The release pipeline restores Swatinem/rust-cache and actions/setup-node
caches as a deliberate optimisation. zizmor flags all eight under
cache-poisoning because release.yml publishes to crates.io / PyPI / npm.
The caches are maintainer-controlled and the restore speedup is kept on
purpose, so accept the finding via a zizmor config ignore for release.yml
rather than running cache-free release builds. (Six of the eight are
actions/setup-node, reported at Low confidence.)
Adds .github/zizmor.yml; release.yml now reports 0 high findings.
* ci: drop persisted checkout credentials on read-only jobs
zizmor's artipacked audit flags every actions/checkout that keeps the
default persisted credential: the token is written to the runner's
.git/config, where it can leak if a later step packs .git into an
uploaded artifact, or be read by another step in the same job.
Set persist-credentials: false on the 20 checkouts whose jobs never push
or authenticate to git (build/test/clippy/msrv/coverage/supply-chain/
fuzz/python/wasm/node in ci.yml, plus bench.yml, codeql.yml, the seven
release.yml build/publish jobs, and sync-metadata.yml). The publish and
release jobs authenticate to crates.io / npm / PyPI / the GitHub API with
their own tokens, not persisted git credentials, so this is safe.
sync-about.yml genuinely pushes the indicator-count fix-up to the PR
branch, so it keeps its credential and is accepted via .github/zizmor.yml.
zizmor artipacked for the repo drops to 0 (0 high, 0 medium remaining).
* ci: use npm ci instead of npm install for reproducible installs
Pins the node binding dependency install to the committed package-lock.json
integrity hashes (OpenSSF Scorecard PinnedDependencies). npm ci installs
strictly from the lockfile; npm install could resolve newer patch versions.
Covers ci.yml and both release.yml node steps.
* ci: hash-pin Python dev tooling in ci.yml (Scorecard #19)
Replaces the unpinned 'pip install maturin pytest numpy hypothesis' with a
hash-locked '--require-hashes -r' install (OpenSSF Scorecard PinnedDependencies).
Two lock files are needed because numpy publishes no single release with wheels
for both cp39 and cp313 (<=2.0.2 has cp39 only, >=2.1 drops cp39):
ci-dev-py39.txt numpy 2.0.2 (Python 3.9, + tomli/exceptiongroup)
ci-dev-py3.txt numpy 2.4.6 (Python 3.10+)
The step selects the file by matrix.python-version under shell: bash. Both are
generated from ci-dev.in via uv (scripts/update-lockfiles.sh, added next).
* ci: hash-pin Python deps in bench.yml (Scorecard #16)
Replaces the unpinned 'pip install maturin numpy pandas talipp finta' with a
hash-locked '--require-hashes -r .github/requirements/bench.txt' install.
bench.yml runs on a single Python version (3.11), so one lock file (generated
from bench.in via uv) is sufficient.
* build: add scripts/update-lockfiles.sh to regenerate all lockfiles
One command refreshes every committed lockfile across languages: Cargo.lock and
fuzz/Cargo.lock (cargo update), the Node binding package-lock.json, and the
hash-pinned Python requirements under .github/requirements/ (uv pip compile
--generate-hashes). Uses uv for the Python locks so a target Python version's
hashed transitive closure can be resolved without that interpreter installed
(needed for the numpy cp39/cp313 split); bootstraps uv if absent.
.gitattributes pins *.sh to LF so the script stays runnable on Linux/macOS.
* ci: split ci-dev requirements per Python version + Dependabot rehash
Splits the single ci-dev.in into ci-dev-py39.in (numpy <2.1, the last series
with cp39 wheels) and ci-dev-py3.in (3.10+), giving a 1:1 .in->.txt layout.
The cap keeps Python 3.9 permanently installable and stops Dependabot from
proposing 3.9-breaking numpy bumps.
Adds a Dependabot pip entry on /.github/requirements so the hash-locked tooling
is kept current automatically; the canonical manual refresh stays
scripts/update-lockfiles.sh. Only the '# via -r' provenance lines in the .txt
change; no package versions or hashes move.
* ci: cache pip and npm downloads in the PR-loop jobs
Adds setup-python cache: pip (ci.yml python matrix + bench.yml, keyed on the
hash-locked requirements) and setup-node cache: npm (ci.yml node job, keyed on
bindings/node/package-lock.json), on both the primary and retry setup steps.
Scoped to jobs that actually install dependencies; the examples-smoke and
clippy-bindings jobs install nothing and are left uncached. release.yml is
intentionally left out: it runs only on tag push (not the PR loop) and is the
publish-critical path, so no caching is added there.
* docs: document hash-pinned requirements and update-lockfiles.sh
Updates the lockfile-policy table: the bindings/python row no longer claims CI
installs tooling unpinned, and a new .github/requirements row documents the
hash-locked CI/bench tooling and the per-Python-version ci-dev split. Adds a
paragraph pointing contributors at scripts/update-lockfiles.sh (uv-based,
self-bootstrapping) as the canonical lockfile refresh.
* docs: changelog entry for hash-pinned CI dependency installs
The v0.4.0-era CI failure was a runner network blip — `napi build` invokes cargo,
whose fetch of index.crates.io hit "Could not resolve host: index.crates.io" and
failed the Node-on-macOS job, forcing a manual re-run. The earlier flake-hardening
(setup-node/setup-python + rust-cache retries) only covered toolchain download and
cache restore, not the registry fetches inside the actual build/publish steps.
Set tool-level network retries as workflow env so every cargo/napi/maturin/
wasm-pack/npm/pip invocation in every job inherits them — including the nested
cargo calls inside napi/maturin/wasm-pack:
- CARGO_NET_RETRY=10 (default 3): cargo classes DNS-resolve / connect / timeout
errors as spurious and retries with backoff; 10 attempts ride out a transient
blip instead of failing the job.
- CARGO_NET_GIT_FETCH_WITH_CLI=true: more robust git-dep fetches.
- npm_config_fetch_retries=5 / maxtimeout=120s: npm ci/install registry retries.
- PIP_RETRIES=5 / PIP_DEFAULT_TIMEOUT=120: pip install resilience.
Applied to ci.yml, release.yml and bench.yml (the workflows that build). No more
manual re-runs for transient registry flakes.
The auto-injected GITHUB_TOKEN defaulted to write-all in ci.yml, bench.yml
and release.yml (no top-level permissions block), and codeql.yml declared
its scopes only at job level. Add a top-level `permissions: contents: read`
to all four so the token starts read-only and only the jobs that genuinely
write through it raise the scope:
- release.yml: github-release keeps contents: write; node-/wasm-publish and
attestations keep their id-token / attestations: write blocks. The
cargo/python/node publish jobs push to crates.io/PyPI/npm via their own
registry secrets, not the GITHUB_TOKEN, so read-only is correct for them.
- codeql.yml: analyze keeps security-events: write (job level).
- ci.yml / bench.yml: no job writes back to the repo (coverage uploads via
CODECOV_TOKEN; bench only uploads an artifact), so no job override is needed.
sync-about.yml already had a top-level block but at contents: write; demote
the top level to read and move contents: write down to the single `sync` job
(the PR-head counter push is the only GITHUB_TOKEN write). The cross-repo
About/docs/webpage/org writes are unaffected — they run through the
fine-grained ABOUT_SYNC_TOKEN, which the permissions key does not govern.
Raises OpenSSF Scorecard Token-Permissions from 0 toward 10.
R10 — `cross-library-bench` previously ran on every push and every PR
to `main`, adding 5–10 minutes of build + bench time per CI run with
no automated consumer of the artefact. It moves to a dedicated
workflow (`.github/workflows/bench.yml`) that fires nightly at 03:00
UTC and on-demand via `workflow_dispatch` (with optional `size` /
`iterations` inputs). The job in `ci.yml` is removed and a pointer
comment is left in its place so future readers find the new home.
R11 — `pyproject.toml`'s `requires-python = ">=3.9"` already allowed
Python 3.13 installs, but the classifier list stopped at 3.12 and CI
tested only 3.9 / 3.11 / 3.12. The Python CI matrix gains `"3.13"`
and the matching `Programming Language :: Python :: 3.13` classifier
is added so PyPI listings and version-search tooling reflect the
actually-tested range.
kingchenc