Summary
In Hong Kong the server guesses every street lamp. OpenStreetMap has 8 lamp nodes in 95 road geotiles there; the Highways Department publishes 3,032 lamp posts for the same tiles. Cities whose lamps are mapped in OpenStreetMap, such as Prague and Utrecht (two of the cities furniture.LAMP_SPACING was measured from), are given each lamp at its mapped point by #981 (worldprops._street_lamps, 44d05fc4). This PR brings Hong Kong on par with them: the department's points go through the same path.
Two commits. The first puts back the mapped_lamps argument that 97d0bcf2 dropped from the roads.build_tile call in blockworker._tile_parts, two hours after #981 added it. Without it a road tile draws no mapped lamp, from OpenStreetMap or from here. That commit's message does not mention lamps, so I read the removal as unintended; if it was deliberate, this PR needs another way in. That commit is also #1493 on its own, with the history behind this reading. The second commit is the Hong Kong source.
How it follows the existing code
- Same record, same path.
hklamps.addappends the records_street_lampsmakes (the ordinary post,STREET_LAMP_H, idhk:hyd:<lamp number>)._aim_lampsturns them to the road andfurniture.placedraws them, unchanged. - Fetched like France's buildings. Per z14 cell from the department's layer on the CSDI Portal, under the existing
portal.csdi.gov.hkhost policy, kept in<cache>/hklamps/v1. Prefetch, a bounded wait (4 s) and a revisit fault when the records are not here yet, asbdtopo.pydoes. - A government point set, as
radars.py,frmotorwaymarkers.pyandnyctanks.pyare. - Switch:
BNE_HK_LAMPS=off. Only cells that overlaphklidar.BOUNDSare asked for.
Which points become lamps
The layer gives a position and a number, no type and no height. The rules are in hklamps.py; counts are for the 3,032 points of the test tiles.
- An OpenStreetMap lamp within 5 m wins (1). Points within 1 m of each other are one lamp.
- A point on a bridge or tunnel footprint is not drawn (538 and 40). A mapped lamp stands on the ground, so one that really stands on a deck would be dropped by
furniture.under_decks. - A point with no drivable road within 40 m is not drawn (63).
- A point inside a ground carriageway is moved out of it. The model's lanes are wider than Hong Kong's, so 658 points are. The width is the one
roads.road_specsgives the road. In order:- to the near kerb, where a guessed lamp stands (half the width +
LAMP_EDGE_M): 434, moved 2.0 m at the median; - else into the median, if it leaves
LAMP_CURB_Mon both sides: 23; - else to the far kerb: 110, moved 10.2 m at the median;
- else not drawn: 78. A post moved into the median or to the far kerb that ends within 6 m of another is not drawn: 13.
- to the near kerb, where a guessed lamp stands (half the width +
Results
12 places all over Hong Kong: city streets, new towns and village roads, 95 road geotiles, 135 km of ground road. Road length is sampled every 10 m; a sample is lit when a drawn lamp of any kind is within 35 m.
| dev | this PR | |
|---|---|---|
| Mapped lamps drawn | 0 | 2,293 |
| Road with a lamp within 35 m | 66.5 % | 79.7 % |
| The same, main roads only | 82.7 % | 93.6 % |
| Lit on dev, dark here | 3.5 km | |
| Dark on dev, lit here | 21.3 km |
- Lamps in a carriageway: 1 post, 0.29 m in (the builder's lane shift,
placement.ring, whichhklampsdoes not apply). - Time:
hklamps.addtakes 5 ms a tile at the median from cache; the worst tile takes 1.36 s, the first read of a big cell's roads. - First visit: 14 cells, 18 requests and 2.1 MB for the 12 places. With the default wait all 95 tiles were built with their lamps.
Known limits
- Dark stretches: 3.5 km of 135 km. 131 stretches; 2 are 100 m or longer, both under flyovers at Hong Kong (22.30625 N 114.16097 E). A road with two mapped lamps anywhere along it guesses none (
furniture.roads_with_points), and rule 2 leaves out the points beside decks. I did not change that rule. - Two of the 12 places do not gain. Hong Kong (22.31583 N 114.15939 E) is level. Hong Kong (22.30669 N 114.16119 E) loses 0.9 km of lit road and gains 1.8 km.
- Flyover decks have no lamps, as on dev: none of dev's 1,161 guessed lamps in the test tiles stands at deck level.
- Far-kerb posts stand about 10 m from their surveyed place, on the right road at the right station along it.
- One post for every lamp. The layer has no type, so no high masts.
- Seen in the game on a 1.10.1-beta tree with this change, not on dev.
Who this affects
- The first commit: every place with OpenStreetMap street lamps. Their lamps stand at their mapped points again, and a road with two of them along it guesses none, as #981 describes. I have not run it outside Hong Kong: my lab could not get map tiles for other cities. In the 95 Hong Kong tiles it draws 6 mapped lamps and changes the rows of 4 tiles.
BNE_MAPPED_LAMPS=offstill gives guessed lamps only. - The second commit: only cells that overlap Hong Kong. With
BNE_HK_LAMPS=offthe rows of all 95 tiles are identical to dev with the first commit. - Other regions:
worldprops.tile_propstakes an optionalspec_argsand works out one more list of cells per road tile. Nothing is asked of the network. - Cache keys: nothing new; the tiles' code key changes with
worldprops.py. - Credits: rows in
CREDITS.md,SOURCES.mdandREADME.md, under the DATA.GOV.HK termsCREDITS.mdalready cites. - #1466: both add lines at the same places in
CREDITS.md,credits.py,mapdata.py,test_env_keys.pyandquality_gate.py, so they conflict as text. I will rebase whichever is left when one lands.
Validation
test_blockworkerhas a new test,MappedLampsTest.test_road_tile_gets_the_mapped_lamps, that fails without the first commit's line.test_hklamps(37 tests, made-up points only) passes.test_worldprops,test_furniture,test_blockworker,test_env_keys,test_mapdataandtest_creditspass.- Two runs of the 95 tiles give identical rows.
test_determinism.platform_mathsfinds nothing in the files changed. I have not run the whole suite.
🤖 Generated with Claude Code on Fable 5.2