joaoh82 / rust_sqlite / 25521223817
65%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 51
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 64.728%. Remained the same
25521223817

push

github

web-flow 
feat(bench): Group C differentiators — W10..W12 (SQLR-16, sub-phase 9.4) (#105)

Lands the Phase 7 / Phase 8 differentiator workloads on top of the 9.3
Group B baseline:

  W10 vector top-10  10k 384-dim vectors, cosine top-10. Two
                     variants: brute-force (no index) and HNSW
                     (CREATE INDEX ... USING hnsw). SQLRite-only —
                     sqlite-vec extension wiring is a follow-up,
                     rusqlite[bundled] doesn't ship it.
  W11 BM25 top-10    SQLRite USING fts vs SQLite FTS5 virtual
                     table. Engine-asymmetric setup + query.
                     1000-doc corpus.
  W12 hybrid         SQLRite-only — 50/50 BM25 + cosine fusion
                     over a 1000-doc corpus with paired text +
                     vector. Mirrors examples/hybrid-retrieval/.

Two notable plan deviations land alongside the workloads:

W11 / W12 corpus scaled 10k → 1000 docs. Plan target was 10k.
SQLRite's FTS doc-lengths sidecar is a single cell that scales
with corpus size (~3 bytes per doc); 10k docs blew through the
4 KiB page cap with `posting cell 1 of 31754 bytes exceeds
empty-page capacity 4085 (overflow chaining is Phase 8.1)`. The
practical cap is ~1,360 docs; v1 ships at 1000 with a comfort
margin. SQLR-21 tracks Phase 8.1.

W10 sqlite-vec comparator deferred. Plan said "sqlite-vec if
installable, else SQLRite-only baseline." rusqlite[bundled]
doesn't ship the extension; loading a pre-compiled .dylib at
runtime is non-trivial and out of scope for v1. Both W10
variants ship as SQLRite-only.

Smoke run (M1 Pro, criterion --warm-up-time 1 --measurement-time 1
--sample-size 10):

  W10 brute-force  sqlrite ~122 ms                  — parser cost
                                                     dominates per-iter
  W10 hnsw         sqlrite ~132 ms                  — same band as brute-
                                                     force at 10k vectors
  W11 BM25 top-10  sqlrite ~533 us / s... (continued)

8704 of 13447 relevant lines covered (64.73%)

1.2 hits per line

Jobs
ID Job ID Ran Files Coverage
1 25521223817.1 51
64.73
 GitHub Action Run
Source Files on build 25521223817