Skip empty multipolygons in union_many() to avoid invalid R-tree box

[geoneutrino]

Generating planet tiles crashes during the mbtile write phase with a
Boost.Geometry assertion:

boost/geometry/index/rtree.hpp:1661:
Assertion `(detail::is_valid(m_members.translator()(value)))
&& ("Indexable is invalid")' failed.
Aborted

Environment

Debian 13.6
libboost-dev 1.83.0.2+b2
git pull master to update my version with the commits of the last 3 months
make clean
make => NDEBUG not set so an assert causes abort() which results in the crash (while with NDEBUG this error is silently ignored)
planet dump of this week (Aug. 27), problematic tile extract attached

Cause

boost::geometry::envelope() of an empty multipolygon returns Boost's inverse
initialisation box — min = +DBL_MAX, max = -DBL_MAX — which fails
the indexable validity check on insert.

Clipping a polygon against a tile boundary can legitimately produce an
empty result, so ProcessObjects() (tile_worker.cpp:402) can hand
union_many() a vector containing empty entries.

Backtrace at the point of failure:

#4 ...rtree<...>::raw_insert (...) at boost/geometry/index/rtree.hpp:1661
#6 union_many (to_unify=std::vector of length 2) at src/geom.cpp:257
#7 ProcessObjects (...) at src/tile_worker.cpp:402
#8 ProcessLayer (..., zoom=13, ...) at src/tile_worker.cpp:496

with to_unify[0].empty() == true and the box printing as
±1.7976931348623157e+308.

Fix

Skip empty multipolygons before the R-tree query and insert. An empty
multipolygon cannot intersect anything and contributes nothing to the
union, so skipping it is semantically correct rather than just
silencing the assertion.

Why this matters for release builds

Builds that define NDEBUG — including CMake Release and most
distribution packages — compile the assertion out. They do not abort,
but they do store the invalid box in the R-tree, which makes
subsequent intersects() queries unreliable for that tile. Polygons
may then be merged that should not be, or vice versa. The result is
silently incorrect output rather than a crash.

The Makefile build does not define NDEBUG, which is why the problem
surfaces there.

Reproduction

Planet dump, z13 within z6/0/14 (north-west Alaska). A cut-down
extract is enough (attached)

osmium extract -b -180,67.5,-172,71.5 planet-latest.osm.pbf -o problem.osm.pbf
tilemaker --input problem.osm.pbf --output problem.mbtiles \
  --config config.json --process process.lua

[systemed]

Thanks! I think this might be similar to 67f1e45 which is in a branch hopefully to be merged later today. I'll compare the two in practice.

[systemed]

Yep, it's the same. I'll merge this, back out the fix in the declutter branch and then re-merge. Thank you!