22 (2/2): auto-catchment, the outline reduction (stacked on #110) - #111
Merged
Merged
Conversation
The red suite for increment 22 PR 2 (docs/increments/22-auto-catchment.md, "The reduction", "The command" and "The red suites", PR 2). No production code. - tests/cpp/unit/test_area_collapse.cpp: statuses (negative, NaN, +-inf tolerance; clockwise; fewer than 4 vertices); tolerance 0 drops exactly the collinear vertices; traced rectangles reduce to at most 8 vertices at one cell; a keep-point in a tab stays strictly inside; a zigzag corridor 1.5 cells wide at 5 cells stays simple; equal bits twice; 300 random star polygons. Every case checks the design's guarantees from input and output alone: area to 1e-9, simple by brute-force classify<DefaultKernel>, keep strictly inside by exact winding, tolerance both ways, and input - collinear - collapses = output. Registered behind if(EXISTS vector_simplify/area_collapse.hpp): REMOVE AT GREEN. - test_core_reduce.py: the binding, on 12 random traced blobs at five tolerances and on the PR 1 bowl and plus catchments: the five guarantees (Hausdorff within tolerance plus a cell), determinism, statuses, ValueError on bad shapes, and a median reduction below 0.8 at two cells. - test_cli_catchment.py: --outline-tolerance 0 writes the fine ring less its collinear vertices; the default (twice the cell) reduces, keeps the area and stays within the tolerance; the properties and the stderr line; -1, nan and inf refused; Bygdin (skipped without Ola's data): reduced area equal to fine to 1e-9, at most 17,812 / 5 vertices, and mesh --domain at --tolerance 10 succeeds. Test amendment: PR 1's two area checks compare at 1e-9, the design's area guarantee, since the default output is now reduced. The refusal helper also rejects Typer's "No such option", so refusals are red until the option exists. The C++ harness was run in the scratchpad against an identity stub: every fixture premise holds and the 16 failures are the behaviour an identity lacks. Red because area_collapse.hpp, _core.reduce_ring and --outline-tolerance do not exist. Main build exits 0, ctest 814/814; the rest of pytest, ruff, format and mypy are green. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Makes the PR 2 red suite (039cf3c) pass, per docs/increments/22-auto-catchment.md, "The reduction" and "The command". No test file changed except removing the red-step if(EXISTS vector_simplify/area_collapse.hpp) guard. - include/terrain/vector_simplify/area_collapse.hpp: reduce_ring<K>. Collinear pass, then APSC-style collapses (E on line A-B or C-D, else the foot of B-C's midpoint) ordered by (deviation, id). Each edge carries a fine-vertex range; a collapse splits its range between A-E and E-D at the split that minimises the larger distance. Crossing tests use classify<K> over a grid of current edges; keep-points use an exact winding round A-B-C-D-E. Stale heap entries are dropped by a per-node version. - bindings/core.cpp, _core.pyi: reduce_ring, ReduceOutcome, ReduceStatus. - catchment.py: CatchmentRequest.outline_tolerance (None: twice the cell); the fine ring is reduced from its lower-left corner, seed kept inside. - cli.py: --outline-tolerance; writes the reduced polygon, with the properties reduced_vertices, reduced_area_m2 and outline_tolerance_m and a "reduced outline" stderr line. Bygdin at 20 m: 17,812 -> 740 vertices, area 304.909550 km2 both, difference -3e-7 m2. mesh --domain at --tolerance 10: 52,198 triangles in 0.7 s (the fine outline gives 94,374); at --tolerance 1: 1,129,026 triangles in 3.2 s. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…project structure Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
test_core_reduce.py's docstring now names four guarantees (area, simple, seed inside, tolerance at vertices) and the Hausdorff bound as measured, not guaranteed, matching the design at 698b19f. check() says it takes traced lattice rings only; every caller passes one (random blobs traced by outline.trace, the PR 1 bowl and plus catchments), so no assertion moved. The C++ random star rings (non-lattice) carry no Hausdorff assertion. test_cli_catchment.py's one Hausdorff check gets the same note. No test matches 'grown: touches' (the stderr tests look for 'window'), so the coming 'grown on north' wording breaks nothing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- catchment.py: module and Catchment docstrings describe the reduced outline and its tolerance as built; LakeError (a CatchmentError) for the lake refusals. - cli.py: catchment --help and --outline-tolerance help as built; a lake refusal is reported under --lakes, not --dem; stderr 'grown on <sides>'. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Area -0.206 % against NVE's 305.54 km2 (within the design's 2 %); node overlap 99.12 % / 99.33 %; 740 vertices at 20 m, Hausdorff 19.92 m. Meshes at 10 m and 1 m on the reduced and fine outlines, and with CORINE. First run: no performance baseline. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
As the design says: the licence and attribution terms are not recorded. README records the curl command, the date fetched and the sha256; analyse.py takes the polygon's path as an argument. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
test_cli_catchment.py: a seed in no lake, and a seed in two overlapping lakes (a GeoJSON of the lake box and the box grown by one cell), are refused with Typer's 'Invalid value for --lakes:' and without 'Invalid value for --dem:' (6757925, LakeError). The whole phrase is asserted, as plain() flattens it (Typer prints the hint unquoted): with the old --dem hint the line reads 'Invalid value for --dem: ...', which a bare '--lakes' substring would still match when the message names the flag, and this phrase does not (checked on a scratch Typer app raising the same BadParameter with param_hint '--dem'). Replaces test_a_seed_in_no_lake_is_refused, whose refusal it keeps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # ROADMAP.md # docs/increments/22-auto-catchment.md # project_structure.md
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # project_structure.md
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
1 participant
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.
Increment 22, PR 2 of 2: auto-catchment, the outline reduction. Stacked on #110: review after it; this PR's base is
increment22-pr1and moves tomasteronce #110 merges. Design:docs/increments/22-auto-catchment.md; acceptance:docs/benchmarks/2026-09-29/bygdin.md.What it does
--outline-tolerance(default twice the DEM spacing, 20 m on DTM10).Numbers (Bygdin, AC, M1 Max)
docs/benchmarks/2026-09-29/bygdin/README.md.Size
Net +406 production lines (CLAUDE.md §2).
🤖 Generated with Claude Code