freeeve / roaringrange / 27297741075
65%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 7
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: 85.137%. Remained the same
27297741075

push

github

freeeve 
feat(rust,openalex): range-probed global term Bloom for split sets (task 028)

The 29 KB summary-stripped manifest gave up absent-term pruning: a
rare/typo query descended through all 389 splits. Instead of putting
per-split Blooms back in the manifest (727 MB resident at 10 bits/key),
one GLOBAL Bloom over the whole vocabulary lives as a sidecar that is
never downloaded - a Bloom filter is range-probeable by construction:

- RemoteBloom reads the 8-byte header once, then probes k one-byte
  positions per key in one concurrent wave; false stays definitive.
- SplitFetcher::global_bloom_name (default None) names the sidecar;
  the tiered search paths consult it LAZILY - only after the top tier
  yields nothing (the rare/absent-term signal) - so present-term
  queries never pay, and a definitively-absent term ends the descent
  after one split + ~k tiny reads instead of 389 split opens. A
  missing/unreadable sidecar never breaks search.
- build_global_bloom example unions every local split's dictionary
  keys (header + dict reads only) into the standard bloom_build layout
  (now pub); wasm RrssIndex.setGlobalBloom wires it, demo split
  configs name openalex-global.bloom.

Tests: remote probe round-trip; absent term opens 1 split + 1 probe vs
all splits without the sidecar; rare-but-present and common terms
return identical results; a configured-but-missing sidecar degrades.

653 of 767 relevant lines covered (85.14%)

11.68 hits per line

Jobs
ID Job ID Ran Files Coverage
1 27297741075.1 7
85.14
 GitHub Action Run
Source Files on build 27297741075