Incorrect linestring clipping: missing geometries

[springmeyer]

A testcase came in that shows that, in some rare cases, the line clipping inside vtcomposite will discard geometries incorrectly that are within the clipping bbox.

[springmeyer]

Added a testcase in 466d654. To replicate do:

git checkout linestring-clipping-bug
make
make test

Until this bug is fixed you should see:

ok 9 should be falsy
not ok 10 should be equal
  ---
    operator: equal
    expected: 5
    actual:   4
    at: composite (/Users/danespringmeyer/projects/vtcomposite/test/vtcomposite-linestrings.test.js:89:12)
    stack: |-
      Error: should be equal
          at Test.assert [as _assert] (/Users/danespringmeyer/projects/vtcomposite/node_modules/tape/lib/test.js:225:54)
          at Test.bound [as _assert] (/Users/danespringmeyer/projects/vtcomposite/node_modules/tape/lib/test.js:77:32)
          at Test.equal.Test.equals.Test.isEqual.Test.is.Test.strictEqual.Test.strictEquals (/Users/danespringmeyer/projects/vtcomposite/node_modules/tape/lib/test.js:388:10)
          at Test.bound [as equal] (/Users/danespringmeyer/projects/vtcomposite/node_modules/tape/lib/test.js:77:32)
          at composite (/Users/danespringmeyer/projects/vtcomposite/test/vtcomposite-linestrings.test.js:89:12)
  ...

What I think is happening is that a multilinestring is disappearing during the call to boost::geometry::intersection(multi_line, bbox_, result); at:

vtcomposite/src/feature_builder.hpp

Line 191 in baf3b42

boost::geometry::intersection(multi_line, bbox_, result);

Very curiously the multipolygon does not disappear if the buffer_size in the test is set to 4079 or 4081.

/cc @artemp

[andrewharvey]

I have a handful other test cases I can provide if needed, but I suspect they are all the same root cause.

When I first ran into the problem, I tried to replicate using vtcomposite but I couldn't replicate, but I used buffer_size: 0, which could have been the reason, if it only happens for some buffer_size values.

[artemp]

@andrewharvey - could you verify df0f462 fixes this issue

/cc @springmeyer

[andrewharvey]

@andrewharvey - could you verify df0f462 fixes this issue

@artemp Yes, thank you! I tested two different places, and it fixed it in both cases, when buffer was set to 4080, which I assume is what's in use in the Mapbox API.

@springmeyer thank you for the investigation work.

[mapsam]

@andrewharvey appreciate your patience with this issue! The fix will go out in 0.2.1, and should go live into Mapbox's vector tile endpoints sometime tomorrow.

[mapsam]

@andrewharvey 👋 vtcomposite 0.2.1 is now deployed to Mapbox's vector tile APIs 😄