art/ismap
1
0
Fork 0
Code Issues 56 Pull requests 1 Projects Releases Packages Wiki Activity Actions

# #110: Finland airports + airport_points from MTK-muut lentokenttaalue #147

Closed
art wants to merge 0 commits from issue-110-fi-airports into main
pull from: issue-110-fi-airports
merge into: art:main
Conversation 1 Commits - Files changed -
art commented 2026-08-17 07:02:10 +00:00
Owner
Copy link

Closes #110

#110: Finland airports + airport_points from MTK-muut lentokenttaalue

What changed

Finland's airports (+ derived airport_points) land from the already-fetched
lentokenttaalue layer in MTK-muut — no fetch change, only staging + two layers,
exactly as the issue scoped it. fi becomes the third per-country tile source to come
back for this pair (se #103, pl #106).

conf/fimap.yaml

  • fi_lentokentta staging — the 835 airport-grounds polygons. kohdeluokka is the
    area-type ladder, per MML's kohdemalli (read from the official 2023-01 kohdemalli
    workbook): 32411/32412 paved/unpaved runways, 32415–32418 other airport / other
    air-traffic areas, paved/unpaved; the delivered counts are in the staging comment.
    paallyste is redundant with the ladder (measured, no NULLs).
  • fi_paikannimi_cased gains 32401 — the traffic-area names ("Liikennealueen nimi",
    airport AND harbour names share the class) — so airport labels go through the shared
    R94 display-casing pass, the one copy of the rule. The hyphen-rule's known limit
    gains the airport genitive case, documented: HELSINKI-VANTAAN LENTOASEMA →
    Helsinki-vantaan lentoasema (the register's own cased convention, cf. its cased twin
    Kiuruveden lentokenttä). The pass also marks regnames AS MATERIALIZED so the
    hyphen-rule probes a materialised set instead of re-scanning the register per part.
  • airports layer (z10–13, filter_only): all 835 areas — runways plus grounds, the
    whole-complex shape Poland's KU* areas give. Names via the register join (R92):
    32401 label point within 500 units, Finnish preferred, biggest font first, longer
    name on a font tie (Kokkola-Pietarsaari's two same-font rows — the truncated
    KRUUNUPYYN LENTOASEMA ELI fragment loses to the full name). airport_class is the
    register's own airport tier     — LENTOASEMA → 1, LENTOKENTTÄ/KEVYTLENTOPAIKKA → 2 —
    the Finnish analogue of Poland's TYPLOTNISKA port-lotniczy ladder, because MTK's
    kohdeluokka does not carry a scheduled/other tier (a paved runway is a kohdeluokka,
    not a traffic class; measured: every named airport has exactly one runway area).
    26 lentoasema airports = Finland's 21 scheduled-service airports + the military
    bases (Halli, Utti) + Malmi, Mikkeli, Varkaus; 62 airfields class 2.
  • airport_points (z8–13, filter_only, z9 keeps class 1): one centroid per named
    airport (R140, reads fi_airports_13), 88 labels.
  • sources: [fi] — the config's first own source (it never had one; every Finland
    layer was shared until now).
  • Acceptance tests: R6 populated, R119 per-zoom no-empty, R120 no-NULL-class /
    no-invalid / class-1-exactly-lentoasema, on both layers, plus an R94 casing test
    on airport_points. R105 staging list grows to 17.

importer/staging.go — ANALYZE after staging builds. The cased rebuild exposed a
real landmine: a query-kind staging table reading a just-loaded load-kind table plans
on NO statistics (bulk-loaded tables have none until autoanalyze fires a minute
later), and the planner's no-stats guess for fi_paikannimi_cased was a nested loop
over the 1.17M-row register — 551k × 308k lower() comparisons, measured: the build
ran over an hour; the same query runs ~5 s once ANALYZEd. Both staging build paths
now end with ANALYZE, so every downstream query plans on real stats. This would have
hit any full import-fi (and #113's original build); documented in the code comment
and docs/architecture.md.

conf/mapsrv.yaml — sources: [map, is, fi]. html/style.json — fi source +
airports_fi + airport_labels_fi (the same two blocks se/pl got). Docs:
country-finland.md (catalogue line, known gaps), style.md, workstream-a.md,
country-requirements.md R13, unification.md, architecture.md.

Verification

  • make build, make test — ok.
  • importer/importer import --layer airports,airport_points conf/fimap.yaml — 22 s
    (cased staging rebuild now ~6 s), all layer acceptance tests pass.
  • make acceptance — every config rc 0. fimap tally:
    acceptance: 40 test(s): 40 passed, 0 failed, 0 skipped, 0 not run
  • Live numbers: fi_airports_13 835 rows (107 class-1 / 728 class-2, 345 named /
    490 unnamed); fi_airport_points_13 88 rows (26 class-1 / 62 class-2); z8/z9 hold
    the 26 lentoasema airports. Every register airport name is on the map except
    GENBÖLEN (13 km from any grounds — the survey names an airfield it does not map as
    airport area) and KRUUNUPYYN LENTOASEMA ELI (a second row for Kokkola-Pietarsaari,
    merged by the tie-break).
  • importer/importer tiles --check conf/fimap.yaml — R121: airports z10/z13 PASS,
    airport_points z8/z13 PASS, keys [name airport_class country].
  • Real-browser CDP render (never virtual time): z10 over Helsinki-Vantaa renders 77
    airports_fi polygons + the label "Helsinki-vantaan lentoasema"; z8/z9 render only
    class-1 labels (Jyväskylän lentoasema at the cursor), class-2 labels never appear
    below z10. Qwen2.5-VL corroborates the z10 screenshot (grey grounds + label).
    The only console error is the page's pre-existing /favicon.ico 404.

Known limits / left undone

  • Helsinki-vantaan lentoasema (and the other hyphenated town-genitive airport
    names: Tampere-pirkkalan, Kemi-tornion, …) lose the second town's capital — the
    register stores these in caps with no cased twin, and the shared R94 rule's
    first-letter fallback is the register's own cased convention (Kiuruveden
    lentokenttä). Documented on the casing pass; a register-membership test against
    inflected genitives would need a stemmer and is deliberately out of scope.
  • airports is a standalone layer drawn over the cover's transport fill, exactly like
    #106 — folding it into cover itself is #118's overlay-carve staging kind (still
    open).
  • The airport_class tier is the register's LENTOASEMA/LENTOKENTTÄ vocabulary, not a
    scheduled-service ledger: Halli, Utti, Malmi, Mikkeli and Varkaus are class 1 though
    they have no scheduled traffic (the same over-inclusion Sweden's iata and Poland's
    port-lotniczy signals carry). Traficom's airport register would be the exact signal,
    and is not on disk.

Merge notes

  • Branch is off current main; the R13/style.md/unification.md edits overlap #103's and
    #106's open PRs (each added the same sentences for its own country) — the final text
    names all three and is what the merged tree should say; git will need a small
    manual resolution there.
  • The DB now holds the rebuilt fi_paikannimi_cased (with 32401) and the new layer
    tables; nothing else was touched (places/stations/water_bodies read the cased pass
    filtered by their own kohdeluokka and are unchanged — their acceptance tests pass).
Closes #110 --- # #110: Finland airports + airport_points from MTK-muut lentokenttaalue ## What changed Finland's `airports` (+ derived `airport_points`) land from the already-fetched `lentokenttaalue` layer in `MTK-muut` — no fetch change, only staging + two layers, exactly as the issue scoped it. `fi` becomes the third per-country tile source to come back for this pair (`se` #103, `pl` #106). **conf/fimap.yaml** - `fi_lentokentta` staging — the 835 airport-grounds polygons. kohdeluokka is the area-type ladder, per MML's kohdemalli (read from the official 2023-01 kohdemalli workbook): 32411/32412 paved/unpaved runways, 32415–32418 other airport / other air-traffic areas, paved/unpaved; the delivered counts are in the staging comment. `paallyste` is redundant with the ladder (measured, no NULLs). - `fi_paikannimi_cased` gains 32401 — the traffic-area names ("Liikennealueen nimi", airport AND harbour names share the class) — so airport labels go through the shared R94 display-casing pass, the one copy of the rule. The hyphen-rule's known limit gains the airport genitive case, documented: HELSINKI-VANTAAN LENTOASEMA → Helsinki-vantaan lentoasema (the register's own cased convention, cf. its cased twin Kiuruveden lentokenttä). The pass also marks `regnames AS MATERIALIZED` so the hyphen-rule probes a materialised set instead of re-scanning the register per part. - `airports` layer (z10–13, filter_only): all 835 areas — runways plus grounds, the whole-complex shape Poland's KU* areas give. Names via the register join (R92): 32401 label point within 500 units, Finnish preferred, biggest font first, longer name on a font tie (Kokkola-Pietarsaari's two same-font rows — the truncated KRUUNUPYYN LENTOASEMA ELI fragment loses to the full name). **airport_class is the register's own airport tier** — LENTOASEMA → 1, LENTOKENTTÄ/KEVYTLENTOPAIKKA → 2 — the Finnish analogue of Poland's TYPLOTNISKA port-lotniczy ladder, because MTK's kohdeluokka does not carry a scheduled/other tier (a paved runway is a kohdeluokka, not a traffic class; measured: every named airport has exactly one runway area). 26 lentoasema airports = Finland's 21 scheduled-service airports + the military bases (Halli, Utti) + Malmi, Mikkeli, Varkaus; 62 airfields class 2. - `airport_points` (z8–13, filter_only, z9 keeps class 1): one centroid per named airport (R140, reads fi_airports_13), 88 labels. - `sources: [fi]` — the config's first own source (it never had one; every Finland layer was shared until now). - Acceptance tests: R6 populated, R119 per-zoom no-empty, R120 no-NULL-class / no-invalid / class-1-exactly-lentoasema, on both layers, plus an R94 casing test on airport_points. R105 staging list grows to 17. **importer/staging.go — ANALYZE after staging builds.** The cased rebuild exposed a real landmine: a query-kind staging table reading a just-loaded load-kind table plans on NO statistics (bulk-loaded tables have none until autoanalyze fires a minute later), and the planner's no-stats guess for `fi_paikannimi_cased` was a nested loop over the 1.17M-row register — 551k × 308k lower() comparisons, measured: **the build ran over an hour; the same query runs ~5 s once ANALYZEd**. Both staging build paths now end with `ANALYZE`, so every downstream query plans on real stats. This would have hit any full `import-fi` (and #113's original build); documented in the code comment and docs/architecture.md. **conf/mapsrv.yaml** — `sources: [map, is, fi]`. **html/style.json** — `fi` source + `airports_fi` + `airport_labels_fi` (the same two blocks `se`/`pl` got). Docs: country-finland.md (catalogue line, known gaps), style.md, workstream-a.md, country-requirements.md R13, unification.md, architecture.md. ## Verification - `make build`, `make test` — ok. - `importer/importer import --layer airports,airport_points conf/fimap.yaml` — 22 s (cased staging rebuild now ~6 s), all layer acceptance tests pass. - `make acceptance` — every config rc 0. fimap tally: `acceptance: 40 test(s): 40 passed, 0 failed, 0 skipped, 0 not run` - Live numbers: `fi_airports_13` 835 rows (107 class-1 / 728 class-2, 345 named / 490 unnamed); `fi_airport_points_13` 88 rows (26 class-1 / 62 class-2); z8/z9 hold the 26 lentoasema airports. Every register airport name is on the map except GENBÖLEN (13 km from any grounds — the survey names an airfield it does not map as airport area) and KRUUNUPYYN LENTOASEMA ELI (a second row for Kokkola-Pietarsaari, merged by the tie-break). - `importer/importer tiles --check conf/fimap.yaml` — R121: `airports` z10/z13 PASS, `airport_points` z8/z13 PASS, keys [name airport_class country]. - Real-browser CDP render (never virtual time): z10 over Helsinki-Vantaa renders 77 `airports_fi` polygons + the label "Helsinki-vantaan lentoasema"; z8/z9 render only class-1 labels (Jyväskylän lentoasema at the cursor), class-2 labels never appear below z10. Qwen2.5-VL corroborates the z10 screenshot (grey grounds + label). The only console error is the page's pre-existing `/favicon.ico` 404. ## Known limits / left undone - **Helsinki-vantaan lentoasema** (and the other hyphenated town-genitive airport names: Tampere-pirkkalan, Kemi-tornion, …) lose the second town's capital — the register stores these in caps with no cased twin, and the shared R94 rule's first-letter fallback is the register's own cased convention (Kiuruveden lentokenttä). Documented on the casing pass; a register-membership test against inflected genitives would need a stemmer and is deliberately out of scope. - `airports` is a standalone layer drawn over the cover's transport fill, exactly like #106 — folding it into `cover` itself is #118's overlay-carve staging kind (still open). - The `airport_class` tier is the register's LENTOASEMA/LENTOKENTTÄ vocabulary, not a scheduled-service ledger: Halli, Utti, Malmi, Mikkeli and Varkaus are class 1 though they have no scheduled traffic (the same over-inclusion Sweden's iata and Poland's port-lotniczy signals carry). Traficom's airport register would be the exact signal, and is not on disk. ## Merge notes - Branch is off current main; the R13/style.md/unification.md edits overlap #103's and #106's open PRs (each added the same sentences for its own country) — the final text names all three and is what the merged tree should say; git will need a small manual resolution there. - The DB now holds the rebuilt `fi_paikannimi_cased` (with 32401) and the new layer tables; nothing else was touched (places/stations/water_bodies read the cased pass filtered by their own kohdeluokka and are unchanged — their acceptance tests pass). 
Finland's airports from the already-fetched `lentokenttaalue` layer in MTK-muut
(835 grounds polygons, no fetch change). The `fi` tile source appears for the
first time — every Finland layer was shared until now — with the same pair that
brought `se` back in #103 and `pl` in #106.

The airports layer draws all 835 areas (runways + grounds, the whole-complex
shape Poland's KU* gives). kohdeluokka is the area-type ladder (32411/32412
paved/unpaved runways, 32415-32418 other airport / other air-traffic areas), but
it does not carry a scheduled/other tier — a paved runway is a kohdeluokka, not
a traffic class, and every named airport has exactly one runway area (measured;
Joensuu has both). The tier comes from the register's 32401 traffic-area names:
the survey prints its airports as LENTOASEMA and its airfields as LENTOKENTTÄ /
KEVYTLENTOPAIKKA — the Finnish analogue of Poland's port-lotniczy ladder — so
airport_class 1 = the register names it lentoasema (26 airports: all 21
scheduled-service airports plus Halli, Utti, Malmi, Mikkeli, Varkaus), else 2.

Names are the register join (R92): 32401 label within 500 units (the terminal
sits on a building footprint or a grounds hole — measured 0-127 m outside the
nearest area, so strict containment would leave Helsinki-Vantaa and the big
airports unnamed), Finnish preferred, biggest font first, longer name on a font
tie (Kokkola-Pietarsaari's KRUUNUPYYN LENTOASEMA ELI fragment loses to the full
name). The 32401 class also carries harbour names; the join filters the name
text to the airport vocabulary. 32401 joins the shared R94 casing pass so the
labels come out cased (one copy of the rule); the register has no cased twin for
the big airports, so the hyphenated genitives render first-letter-cased
(Helsinki-vantaan lentoasema) — the register's own convention, documented.
GENBÖLEN LENTOKENTTÄ is the one register name 13 km from any grounds (source
gap). airport_points derives per R140: one centroid per named airport, z9 keeps
class 1, 88 labels.

Importer fix this exposed: staging builds now end with ANALYZE. A query-kind
staging table reading a just-loaded load-kind table plans on no statistics
(autoanalyze fires a minute after a bulk load), and the planner's no-stats guess
for the fi_paikannimi_cased rebuild was a nested loop over the 1.17M-row
register — 551k x 308k lower() comparisons, an hour-long build vs ~5 s once
analyzed. This would have hit any full import-fi, not just this layer's rebuild.

Verification: make build/test; --layer import 22 s with all layer acceptance
tests green; make acceptance rc 0 across all configs (fimap 40/40); tile-check
R121 PASS on both layers; CDP render over Helsinki-Vantaa shows 77 polygons +
"Helsinki-vantaan lentoasema" at z10 and only class-1 labels at z8/z9, Qwen2.5-VL
corroborates. Acceptance tests added: R6 populated, R119 per-zoom, R120 no-NULL
class / no-invalid / class-1-exactly-lentoasema, R94 casing.
No change to Finland's output (measured: 88 points, 26 class 1, before and
after) — the derivation is made right for the sources that will exercise it,
and the comment that got the reasoning wrong is corrected.

* `min(airport_class)`, not `max`. The ladder is 1 = scheduled, 2 = everything
  else, so max() demotes an airport to the tier of its weakest polygon. The old
  comment claimed the derivation "takes max(airport_class), so the label tier
  comes out right" — that is not why it came out right. Finland's class is a
  pure function of `name` and the grouping is by name, so every row in a group
  already agrees and the aggregate cannot change the answer. Poland's BDOT10k
  does disagree with itself, and there max() drops Wrocław-Strachowice and
  Lublin off the z9 filter.

* Group by (name, spatial cluster), the R91 idiom places uses, not by name
  alone. eps 5 000 web-Mercator units; Finland's grounds areas touch (23 units
  apart at closest approach, the tightest of the four countries), so nothing
  splits here.
art commented 2026-08-17 22:10:33 +00:00
Author
Owner
Copy link

Merged through !154

Merged through !154
art closed this pull request 2026-08-17 22:10:33 +00:00

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels   
blocked
issue is stuck: missing decision, credential, or data (work-issue blocked path)

  
claude-wip
night-shift

  
map-report
map data feedback (mapsrv)

  
map:is
map data feedback (mapsrv)

  
map:se
map data feedback (mapsrv)

  
needs-review
night-shift

No labels
blocked
claude-wip
map-report
map:is
map:se
needs-review
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
art/ismap!147
No description provided.