fix(area): decide segment touching by cross product, not by division

[DennisOSRM]

The problem

intersect() already means to treat a touch as not a crossing. With std::less_equal it
rejects the parameters at 0 and at 1, so a segment that ends on another does not cross it.
It says this by dividing, and the division does not land on 1.

Whether it does depends on the compiler. For a segment ending exactly on the endpoint of
another, at the scale we store coordinates, the parameter is exactly 1 when multiply-add is
contracted into fma, and 0.99999999999985489385 when it is not:

-ffp-contract=on    t = 1                      touch rejected, correct
-ffp-contract=off   t = 0.99999999999985489385  counted as a crossing

So the same predicate on the same points can answer differently in two translation units of
one build. That is how a unit test on these coordinates passed while the fuzz harness saw
the same line blocked.

The fix

A parameter is exactly 0 or 1 exactly when the matching endpoint lies on the other segment's
line. That is a cross product, not a quotient. It is exactly zero whenever the two
differences it multiplies are zero, which is this whole case, and no rounding or contraction
can move it off zero.

Degenerate cases are now decided by the cross products. Only genuinely interior crossings
are decided by the division. The change is additive: four cross products and a snap.

Why it matters

A wrong answer here deletes a sight line from a visibility graph. Walking along a wall is
allowed, so a line that grazes an obstacle's corner and then runs along its face is real,
and dropping it sends the planner the long way round.

Two places in this repository already work around the same weakness by hand. One of them
says so directly:

intersect() is meant to ignore endpoint hits by itself, but it cannot be relied on

Testing

At the scale plaza coordinates live at, near 1.0 with differences around 1e-6, 543 of 2400
touching configurations were reported as crossings before this change and none are after.
That test fails without the fix, checked by reverting it.

A single hand written case is also included but is deliberately not the guard. Under
contraction it passes either way, so on its own it would keep passing while the predicate
was wrong everywhere else.

Full cucumber suite: 1478 scenarios, 1463 passed, 15 skipped, 0 failed, the same as before
the change. Unit suites for engine, extractor and util all pass.

🤖 Claude Code, Claude Opus 5

[Copilot]

Pull request overview

This PR makes the extractor::area::intersect() predicate deterministic for “touching at endpoints” cases by avoiding reliance on floating-point division landing exactly on 0/1, preventing compiler-dependent endpoint touches from being misclassified as proper crossings (which can incorrectly remove visibility edges in area routing).

Changes:

• Snap the intersection parameters to 0/1 when an endpoint lies exactly on the other segment’s supporting line (via cross products), and only rely on the quotient for non-degenerate cases.
• Add regression tests covering both a concrete real-world coordinate case and a larger deterministic grid of lon/lat-scale “T-touch” configurations.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
include/extractor/area/util.hpp	Makes intersect() robust to FMA/contraction differences by snapping endpoint-touch parameters using cross products before endpoint-exclusive comparisons.
unit_tests/extractor/area/util.cpp	Adds targeted and grid-based unit tests ensuring endpoint touches are never reported as crossings at relevant coordinate scales.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov]

Codecov Report

❌ Patch coverage is 95.83333% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 94.74%. Comparing base (551f3f0) to head (f0d440a).

Files with missing lines	Patch %	Lines
unit_tests/extractor/area/util.cpp	94.28%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #7692      +/-   ##
==========================================
- Coverage   94.74%   94.74%   -0.01%     
==========================================
  Files         519      519              
  Lines       41534    41582      +48     
==========================================
+ Hits        39353    39398      +45     
- Misses       2181     2184       +3     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.