Repository navigation
fix: resolve AU streets written before their number to the right street - #223
Conversation
…kups matching longer street names
Merging this PR will not alter performance
Comparing Footnotes
|
Speed, full index: no slowdown over 10%DetailsSearchAddressTime in ms over 1995 cases, each the fastest of 3 requests.
SuggestAddressTime in ms over 1996 checks, each the fastest of 3 requests.
ReverseGeocodeTime in ms over 1994 cases, each the fastest of 3 requests.
Fails when the total, p50, p90 or p99 of any table is more than 10% slower. Runs of one binary differ by up to 5%; the max is one request and only reported. Base |
Gold check, full index: no regressions
Pass rate by countryPassing checks out of those checked, baseline → candidate, for the 2 expected countries where any RPC changed. A case that expects several countries has its own row. Counts are checks, not independent locations.
Runs
SearchAddress: changes by cohort
Only cohorts with a change and kinds that occur are shown. gained/lost: a result appeared/disappeared; finer/coarser: resolution level; moved: >1 km; nudged: <1 km but visible at 3 decimals. Newly passing (verify before relabelling) (5)
Better failure class (1)
SearchAddress: changed cases (6)finer (1)
coarser (1)
moved (2)
nudged (2)
Landmark checks (head)
|
Summary
Australian queries that write the street before its number ("Panorama Place 3, Maleny", "Railway Terrace 9, Mount Pleasant") now resolve to that street's house instead of a house on a street that merely resembles the town name. On the local full index the AU panel goes from 58 to 60/100 searches and stays at 60/100 suggestions. Refs #222 (not closing it: 40 AU queries still fail, triaged in the new doc).
Changes
docs/evaluation/triage/global-au.md: the remaining AU failures, mostly G-NAF points that differ from the OSM node.Running the venue rule before the village rule was tried and left out. It also fixed NO "Sinneslisundet 28, Tjørhom", GB "Market Place 2, Wymondham" and some PL, ZA and ES Ceuta cases. But in CI it turned the JP case "帯屋町 2丁目1−27" from abstain to wrong_component, because the panel expects the chōme inside the house number and the index keeps it in the street. The village rule stays first; the reorder belongs with the Japan triage.
Review focus
The village-name guard (
fold(name) == fold(town) || !name.contains(' ')) and the street-type check on the venue tail insrc/index/mod.rs.Local full-index gold diff, main vs head on the same index and models: SearchAddress has 5 newly passing (AU Railway Terrace 9, Panorama Place 3; ES Urbanizacion Pedraza Beach 5 x2, Barrio Colmenar 23), 1 better class (ES 993663692-0, wrong house in Grávalos to locality Muro de Aguas), 0 newly failing, 0 worse. Suggest, Reverse and ResolveRegion are unchanged. On the first CI run gold-speed passed and quality failed only on the JP case, which the second commit addresses.
cargo fmt,clippy -D warningsandcargo test(including synthetic and synthetic--ignored) pass locally. CI gold needs a re-run on the new head.