vortex-data
vortex
Blog
Docs
Changelog
Blog
Docs
Changelog
Overview
Branches
Benchmarks
Runs
Performance History
Latest Results
Fix dtype mismatch in FileStatsLayoutReader for stat scalars Use the stat's own dtype (e.g. u64 for NullCount) rather than the field dtype when constructing stat scalars in stats_ref. This fixes IS NULL pruning on nullable timestamp columns which previously failed with a dtype mismatch. Add regression test for is_null pruning on a nullable timestamp column. Signed-off-by: Brian Hart <brian@brainhart.dev>
brainhart:fix-stat-dtype-mismatch
6 hours ago
benchmarks-website-v3: name the server crate in planning docs The component plan and decisions log previously left the server crate name to the agent. The actual crate is `vortex-bench-server` at `benchmarks-website/server/`; record that so the planning docs match the workspace. Signed-off-by: Claude <noreply@anthropic.com>
claude/benchmarks-v3-server
9 hours ago
benchmarks-website-v3: relax dep versions, prefer workspace, axum 0.8 - Drop the `=` exact pins in favour of caret minor-version constraints so the resolver can pick patches alongside the rest of the workspace. - Switch dev-dep `reqwest` and `tracing-subscriber` to `{ workspace = true }` (matching every other workspace consumer); the previous pinned `tempfile = "=3.14.0"` had already been switched. - Drop redundant `default-features = false` / `features = [...]` where the requested features are part of the upstream defaults. - Bump axum 0.7 -> 0.8 (route paths use the new `/{slug}` syntax) and maud 0.26 -> 0.27 (drops the unmaintained `proc-macro-error` advisory in favour of `proc-macro2-diagnostics`). Server unit + integration tests, clippy `-D warnings`, and nightly fmt all pass locally. Signed-off-by: Claude <noreply@anthropic.com>
claude/benchmarks-v3-server
9 hours ago
fix(bench): align Public-BI dims; tighten v3 snapshot tests - compress-bench now emits Public-BI sub-datasets as `dataset = "public-bi"`, `dataset_variant = "<sub-dataset>"`, matching the SQL query path and the per-suite dim values table in `benchmarks-website/planning/benchmark-mapping.md`. Adds a `Dataset::v3_dataset_dims` default method (`(name(), None)`) and overrides it on `PBIBenchmark` to return `("public-bi", Some(name))`. - Snapshot tests in `vortex-bench/src/v3.rs` now return `anyhow::Result<()>` and use `?` instead of `.expect`, matching the CLAUDE.md test convention. Adds one new snapshot covering the Public-BI variant case. Signed-off-by: Claude <noreply@anthropic.com>
claude/benchmarks-v3-emitter
10 hours ago
benchmarks-website-v3: drop tempfile pin so workspace stays consistent The dev-dep `tempfile = "=3.14.0"` in the new server crate forced the workspace lockfile to 3.14, but `lance-index 4.0.0` (transitive via lance-bench) calls `TempDir::keep()` which was added in 3.15. CI on linux-arm64 was failing with: error[E0599]: no method named `keep` found for struct `tempfile::TempDir` in the current scope --> lance-index-4.0.0/src/vector/ivf/shuffler.rs:58:41 The pinned-stack rule in the task brief covers axum / maud / duckdb-rs and direct deps; tempfile is a test utility and matches every other workspace consumer (`{ workspace = true }`). Signed-off-by: Claude <noreply@anthropic.com>
claude/benchmarks-v3-server
10 hours ago
benchmarks-website-v3: exclude vortex-bench-server from windows tests Mirrors the existing exclusion of vortex-duckdb on windows-x64. The new bench server crate also depends on duckdb-rs with the bundled feature, which does not build on Windows runners. Signed-off-by: Claude <noreply@anthropic.com>
claude/benchmarks-v3-server
10 hours ago
chore(bench): satisfy ruff and REUSE on emitter PR - ruff format + UP017 (`datetime.UTC` alias) on `scripts/post-ingest.py` - annotate `vortex-bench/src/snapshots/**.snap` in `vortex-bench/REUSE.toml` (insta `.snap` files do not allow leading comment lines for SPDX tags) Signed-off-by: Claude <noreply@anthropic.com>
claude/benchmarks-v3-emitter
11 hours ago
feat(bench): add `--gh-json-v3` emitter and post-ingest script Adds a v3-shape JSON emitter alongside the existing `-d gh-json -o ...` form. Each benchmark binary gains a `--gh-json-v3 <PATH>` flag that writes one JSONL record per line, discriminated by `kind`. The five record shapes match the fact tables in `benchmarks-website/planning/01-schema.md` and the wire contract in `02-contracts.md`. The legacy gh-json path is untouched -- both forms coexist at alpha. The post-ingest script in `scripts/post-ingest.py` (Python 3 stdlib only) reads the JSONL, fills the `commit` envelope by shelling out to `git show`, and POSTs to `<server>/api/ingest` with a bearer token. No retries, no spool, no outbox -- those are deferred per the alpha plan in `benchmarks-website/planning/components/emitter.md`. Snapshot tests (insta) cover one record per `kind`, scrubbing `commit_sha` and `env_triple`. Signed-off-by: Claude <noreply@anthropic.com>
claude/benchmarks-v3-emitter
11 hours ago
Latest Branches
CodSpeed Performance Gauge
0%
Fix dtype mismatch in FileStatsLayoutReader for stat scalars
#7639
9 hours ago
889b25f
brainhart:fix-stat-dtype-mismatch
CodSpeed Performance Gauge
0%
[claude] benchmarks-website-v3: alpha server (axum + maud + duckdb)
#7637
9 hours ago
89d1468
claude/benchmarks-v3-server
CodSpeed Performance Gauge
0%
[claude] feat(bench): add `--gh-json-v3` emitter and post-ingest script
#7638
10 hours ago
d991fec
claude/benchmarks-v3-emitter
© 2026 CodSpeed Technology
Home
Terms
Privacy
Docs