Repository navigation
merge ground-plan into develop: the TrailView road rebuild so far - #3491
Merged
Merged
Conversation
With the 2^11 cells of the town grid, the index of a Lucerne tile took 2.8–3.4 MB, most of it for the cells under a lake's box. With 2^13 cells it takes 1.1–1.3 MB. A grid now carries the size of its cells; without one it keeps the town grid's. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… still wished read At Lucerne the first jobs of cold tiles pushed the eye's tile out of the worker's tile memo before its last wishes came, and the eye's tile was decoded and shaped again just before the eye field was final. Each job now names the feature tiles of the nodes on their way and of the frame's queued wishes, and `trim` keeps them, as it keeps the newest. Among the others it evicts as before. The set of tiles whose body went to the worker keeps only the tiles that jobs read now: a tile asked again later sends its body again, which the worker ignores where it still holds the tile. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ecords per node The record's 16-bit line number only groups a line's segments in a cell. A tile with more than 65,536 lines now numbers the lines that get a record in each node, in tile order, so a late river or bridge stamps; a tile with fewer keeps the tile index and its bytes. A node with more than 2^16 lines with records passes STAMP_WORDS with their records alone, so it stamps nothing and counts a limit. The first indices of the outline groups of the tile index are named `seasFrom` and so on. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…hen the wishes go `trim` kept a wanted tile past the budget and ran only when a decode ended, so an excess stayed at idle. `wantTiles` now trims after it replaces the wanted set. The main thread names the tiles of the frame's wishes once when it sends them, not again for each job. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ach job Trimming at each job dropped the eye's tile at Lucerne between two waves of its wishes: in 3 of 3 cold loads it was decoded again at 11.7-12.3 s, and the eye field was final 1-2 s later. The next decode trims the tiles that no job wants, as before; a test holds that. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s memo at each wanted set The tiles of the jobs on their way and of the queued wishes did not keep the eye's tile at Lucerne: its own wishes stop for seconds between the waves of its levels, and the first jobs of cold tiles evicted it. In 2 of 2 loads of the earlier rule it was decoded again at 10.7-12.9 s. The wanted tiles now also hold the feature tiles of the near levels' nodes that the last frame read. `wantTiles` trims at once, so the memo falls back under its budget when the wishes go. The main thread names the tiles of the frame's wishes once per frame, not for each job. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
trailview: the stamp worker indexes a tile once, builds only the stages a job reads, and degrades per primitive
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tests its owner CellInput::new cloned every part of each zoom-13 tile and ran the owner test on it before the box test, so most of that work was thrown away. It now tests the box first, and copies a kept feature without its parts. The kept parts and their order do not change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The main thread sent every feature tile of a cell to its plan worker with every build, and the worker unzipped each tile for every cell, although the bridge keeps the parsed tile. Now the main thread sends a tile's body to a worker once and then only its name. The bridge checks the named tiles before the build. Where it no longer holds one (its cache let the tile go), the worker answers `missing`, and the job goes again with that body. A new worker gets every body again. The tile cache keeps no absent tile, so a held tile is always one that a caller gave. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…st runs only where it is read A records build encoded its request (`to_bytes`), decoded it again (`Request::decode`) and ran an FNV digest over all its bytes, and `add_tile` and `add_page` copied their bytes once more. Now the bridge takes the tile and page bytes that JavaScript gives it, and builds from the request in memory (`cell::build_request`). The request digest is optional (`CellBuild::request`): `build_cell` and `build_bytes` compute it from the canonical bytes as before, and the bridge's `build` computes it only with `digest`, which the debug draw gives. Then its plan digest is the native one. The unused bridge function `records` is gone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…, not a level A page that finds no room in its pool is left out: the finest level first, by cell in levels 17 and 18, the farthest first; a page left out frees its slot at once and never draws, so the coarser page draws there. While the camera moves, a resident page and a cell whose pages all fit rank as if 0.8 times as far. The quality falls only when a page of level 16 or coarser is left out. A cell's mesh rests while a page finer than 16 in it is left out; the other cells keep drawing. The stats and the HUD show the pages left out per level. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ver it keep drawing A left-out page's parent joins the wanted pages, so the page that draws in its place is one level coarser where it fits: before, a left-out level-18 page near the eye drew level 15, as no level-17 or level-16 page is wanted there. The pages rank by level and distance alone. A cell's mesh rests only where a page coarser than 17 draws in place of one of the mesh level; over a level-17 parent it draws as in a load. The quality falls only once the pages that fit are in: while a view loads, its selection is wider than the pool. When every page fits, the fit does not sort. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ity step Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…it answered The main thread counted a tile as held when it posted the tile's body. A later job that named the tile could then reach the bridge before the job with the body, and answered `missing`. Now a tile counts as held only when the job that carried its body answers with its build: until then every job for that worker carries the body too. A normal load has no `missing` answer. A job that answers `missing` goes again with every body, so it cannot miss twice. A `missing` answer for a job whose cell left frees the job, as every other answer does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… its build `add_page` returns an error for a page that does not hold 65,536 codes: `Request::decode` refused one before, and the direct path would trap on it. `build` takes the request by value, so a second build of one request throws instead of building without inputs. A constant check keeps the tile cache at four tiles or more, the most that a cell reads. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflicts: engine.ts keeps develop's GroundSource.open(config, stop.signal) and ground-plan's TrailView constructor and early request; planner.rs takes develop's snow format 2 and keeps ground-plan's stamp step note and trailview-features revision 9; catalog.rs, planner_runtime.py, planner_verify.py, obc-data.md and trailview-tiles.md list develop's three snow archives and ground-plan's stamp layer. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…lvert A way on the ground no longer lies 0.2 m under the water that it shares a vertex with or that it crosses as a ford. FORD_M, rows::fords, the ford and stepping-stone flags of the solver and the fords metric go. A way tagged as a ford takes the culvert cover of an untagged crossing like any other way. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… rests every mesh or none Review fixes of the page shortage (#3265): - The quality only makes the range smaller, never the finest level. It steps down only where the step lowers the demand of the short pool, and up only where the demand of the finer quality fits, both from a selection at that quality: for fixed pages and pools it comes to a fixed point, and a still view stops drawing. On a usual screen it stays 0; the pages left out alone answer a shortage. - If one cell in range would rest, every cell in range rests, with the CALM delay as before: no resting cell beside a drawing one inside the sink range. - A cell rests by the page that draws in place of a left-out page after the update. A left-out page draws until that page draws, and any allocation may evict it. - The synthesized pool is cut first: inputs of a left-out synthesis do not count. - The wanted pages and the ground's reads go to the pages sorted together again. - The short-pool test checks that no two pages share a slot. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
merge develop into ground-plan
…-shortage # Conflicts: # builder/web/src/lib/trailview/engine.ts
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An allocation takes first a page that no frame wants, then a page left out under the page that the slot is for (it draws in the frame of its slot), and a page left out that still stands for a page not yet in its place last. Without this, an allocation took such a page, the page under it drew level 16 or 15 for some frames, and every mesh rested. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
TrailView load: a plan worker gets each tile once, and the request crosses into the crate once (#3264)
… the ground; each deck arm ends on its mouth A junction where a deck ends among ground ways was no point of the storey, so each deck arm ended on a straight line through the junction and its outline lay over the ground arms; two height probes in the shell hid part of that and left shards (Acheregg's east deck end). - The graph of the storeys gives such a junction to the ground: each deck arm's stretch up to its mouth, the first sample of the deck where its stroke has left the strokes of the ground arms that leave across it (more than FLAT_TURN_DEG off), becomes a ground piece that meets the deck at a joint, as a stub meets its deck. A deck that runs through such a node ends there; a mouth that reaches another deck node takes that node to the ground too. - The shell's two probes (an open edge over a ground way within 0.1 m, a parapet's end within 1 m) are gone. - Tests: no edge, parapet, railing or post stands in the mouth of a way that meets its deck at a junction (storey 0's records); stubs at joints and ways beside a deck as measured budgets; open pieces need a paved face at the top's height or a stub's verge or slope past them (no water or bank); the seam budget at the fixture's 14. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ncrete only beside its ground stretch STOREY_TOP shared bit 29 with BRIDGE_SIDE, so a ground railway's wall side painted its rails at any distance. The 3 m near_deck rule had no source and reached other ways' walls; the stubs' own flag stays. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…and plain lists A B-tree of nodes and a new sort added 5 kB of gzip to the device. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Measured in CI run 43572252884: 806,847 B gzipped with the glue (+4,581 B), 2,087,262 B raw. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ground plan GP12.3b part B: the viewer draws the decks, the device links the storeys
Merges origin/ground-plan 46793ba (#3482, deck step part B) into the merge of develop. Both sides had trailview-features revision 19 for other contents (this branch: develop's 12 and ground-plan's 18 merged; #3482: the merged A lines), so the layer takes the next number above both: 20. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
merge develop into ground-plan
The budget was the CI size rounded up to the next whole KiB: 65 B gzipped and 674 B raw of headroom on a toolchain that follows stable. The guard now states the recorded CI size and the owner's margin apart, and prints both parts with the change against the recorded size. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
builder: give the ground-plan bridge budget the owner's 8 KiB margin
…r, and only a way on top hides it Two rules in the levels that the feature bake computes for tunnels: - Within open_max_m of a tag end, a tunnel station with too little cover is OPEN also where its floor lies over the terrain model: the way goes on as a ground way on its bank until the cover holds. - A paved stroke hides a tunnel station only where its surface lies more than the at-grade band (the lower snap_height_m of the two ways) over the floor. The lift on free ground still reads every paved stroke: a paved face never moves. The features take revision 21 and the fixture is baked again: one tunnel feature changes its levels tag. The spec says how a tunnel station gets its state. The device code does not change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tiles of an older bake hold a way through a building (`tunnel=building_passage`) as the kind `tunnel` with its tag. The viewer's `lineType` and the plan's `classify` now take such a way as a way of its class on the ground, as develop's viewer does, so no hill, portal or lifted house shows at a gateway. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ls rules The bridge's code does not change; CI measured 806,965 B gzipped (+118 B) and 2,087,266 B raw (+4 B) on the head with the rules. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The CI build of the bridge with the building-passage rule: 808,271 B gzipped (WASM and glue), 2,090,936 B raw, on stable 1.100.0. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… tag end Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
trailview: a building passage is never a tunnel, also on older tiles
The bridge's recorded size takes ground-plan's for now; this branch records its own CI size on the merged base next. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ls rules on the passage rule CI on 681070b measured 808,099 B gzipped (-172 B) and 2,090,936 B raw (+0 B); the bridge's behaviour does not change. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ground plan GP12.4 step 1: tunnel rules A and B in the bake's levels (no pixel change)
Contributor
|
Important Review skippedToo many files! This PR contains 293 files, which is 193 over the limit of 100. To get a review, reduce the PR to 100 files or fewer by splitting it into smaller PRs or changing its base branch. Upgrade to a paid plan to raise the limit. This review couldn't start because sufficient usage credits or metered capacity aren't available. Add credits or update usage-based reviews in the billing tab, then retry.
|
The test builds three real fixture cells and meshes each face alone. It takes several seconds, and it timed out at the default limit of 5 s in a CI run. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
test: give the fixture test of the debug mesh a longer time limit
20 tasks
Open
10 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
This pull request merges the integration branch
ground-plan(head 8a399ce) intodevelop. It brings the TrailView road rebuild of epic #3096 as it stands today: the ground plan crate, its WASM bridge, the new road solver, line water with culverts, bridge decks, the tunnel levels rules and the viewer's plan path.The owner gave the go for this merge on 2026-11-15, after he read the merge document.
After this merge, new work on the epic targets
develop. The branchground-planstays, but gets no more pull requests.What changes for a viewer
ground=plan. The owner decided on 2026-11-15 that the plan path becomes the only path and that no off switch stays. That is the last step of the epic (TrailView ground plan GP19: the ground plan becomes the only path, the old code goes, docs, budgets, merge into develop #3110) and is not in this pull request, because the plan path does not yet draw sidewalks, kerbs, zebras, rails near the eye, step treads and guardrails.trailview-featuresrevision 21. The viewer reads older tiles; a house with a passage draws correctly on them (trailview: a building passage is never a tunnel, also on older tiles #3489).What is in it
Each part had its own pull request into
ground-plan, with a fresh review, a delta review and green CI. The last ones:developmerged intoground-plan(the preparation of this merge; the proofs for both paths are there).developmoved by 6 commits since #3481 (#3326 and #3485: the seam of the reference terrain inobc-dem, one test name). They merge without a conflict. This pull request's CI runs on the merged tree.Known faults that come along
All are on the plan path, which is off by default:
Checks
develop's stills on the GPU.Public docs
No guide page changes beyond what
develophas. The TrailView README rows and the generateddocs/design/map.jsonchange.Mode: change #3096
Requirements: none
🤖 Generated with Claude Code