main moved each tool's option table to file scope so that usage.cpp can
generate the usage message from it (#409), which conflicted with the
tables the MLT options had been added to. The three MLT options move to
the file-scope tables, under the same headings they were listed with
before: "Setting or disabling tile size limits" in tippecanoe and
tile-join, and "Output tile" in tippecanoe-overzoom.
Also combine the new radix-sort-test with the MLT tests in the test
target, add usage.o to the tools that gained MLT objects, keep both the
MLT=0 and the new docs CI jobs, and regenerate the man page for the
README changes, which the docs job now checks.
The MLT build fails on aarch64: FastPFOR's cmake passes -msse4.2 to the
compiler in every SIMD mode, including the portable one that MLT selects,
so the build dies with "unrecognized command-line option '-msse4.2'" on
ubuntu-24.04-arm, and on Apple Silicon for the same reason. Nothing about
the MLT integration is architecture-specific, so this is as true of the
output format as of the decoder; it went unnoticed because CI never ran
on the branch that added it.
Upstream fixed it in maplibre/maplibre-tile-spec#1492, by vendoring a
FastPFOR snapshot with the SIMD codecs removed -- which is no loss, since
the SIMD encoder's output isn't compatible with the non-SIMD decoder and
was never used. Move the submodule forward four commits to pick that up
along with the platform fixes in #1497.
FastPFOR is no longer a cmake option there, so drop the two -D flags for
it from the configure step and follow the library's rename from FastPFOR
to fastpfor-lib. FastPFOR decoding is still compiled in, so no tile
becomes unreadable, and the checked-in standards are unchanged. The
submodule also no longer needs its own fastpfor and simde submodules, so
the recursive checkout gets a little smaller.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LxhELiLtpUFwrPSWwTYdHG
MapLibre Tile support pulls in the maplibre-tile-spec submodule and needs
cmake to build it, which is a lot to ask of anyone who only wants to work
with Mapbox Vector Tiles.
`make MLT=0` compiles with -DNO_MLT, skips the submodule and its cmake
build entirely, and drops the MLT tests from `make test`. The result needs
no dependencies beyond the ones MVT already needed, and can be built from
a checkout with no submodules at all.
Everything that touches the MLT library is behind the #ifdef, which is
just the two files that were written for it. The option parsing and the
tile format helpers stay compiled either way, so nothing else needs to
know: --output-format=mlt reports that the build has no MLT support rather
than being an unrecognized value, and a tile that is recognized as MLT
reports the same instead of being misparsed as a protobuf, since the
format sniffing itself doesn't need the library.
CI builds and tests this configuration from a checkout without submodules,
so it can't quietly stop working.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LxhELiLtpUFwrPSWwTYdHG
Both tools could read MapLibre Tiles but only ever wrote Mapbox Vector
Tiles, so there was no way to convert a tileset into MLT, or to keep a
tileset in MLT once it had been through either of them.
Give them the same --output-format, --pretessellate, and
--no-mlt-feature-sort options that tippecanoe has. tile-join writes the
chosen format to mbtiles files, PMTiles archives, and tile directories,
naming directory tiles and the tileset metadata format accordingly, and
tippecanoe-overzoom writes it to its output tile. Either tool will read
whichever format its sources are in regardless of what it is writing.
The output format selection and the MLT encoder options now live in
mlt.cpp, shared by all three tools rather than defined in main.cpp for
tippecanoe alone, along with encode_tile() for encoding a tile in the
selected format. overzoom() takes the format as a parameter, since
tile-join uses it internally to rescale tiles that will be re-encoded
afterward, and those intermediate tiles should stay MVT.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LxhELiLtpUFwrPSWwTYdHG
Tilesets written with --output-format=mlt could not be read back by any of
tippecanoe's own tools, which made MLT a dead end rather than a tile format.
Build the mlt-cpp decoder from the vendored maplibre-tile-spec submodule
alongside the encoder, and convert a decoded MapLibre Tile back into the
equivalent mvt_tile: geometry, feature ids, and property columns. MLT rings
come back explicitly closed, so the repeated final point becomes an MVT
closepath, and the property columns, which are held in an unordered map,
are sorted by name so that the attributes of a decoded tile come out in a
stable order.
Rather than adding an option to each tool, mvt_tile::decode() detects the
encoding and dispatches, so everywhere tippecanoe already reads a vector
tile can read MLT. An MLT tile begins with a varint layer length followed
by a varint layer tag whose only defined value is 1, while an MVT tile is
a protobuf whose only field is the repeated layer field 3, so it begins
with 0x1a followed by a layer length that can never be as short as the one
byte that would be needed to look like an MLT layer tag.
Tile directories also needed a fix: enumerate_dirtiles() recognized .mlt
file names but still recorded .pbf as the extension to read them back
with, so decoding an MLT directory failed to find any of its tiles.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LxhELiLtpUFwrPSWwTYdHG
An MLT property column has one type for the whole layer, while MVT values
are individually typed. Converting each value on its own left the encoder
to reconcile the mismatch, and its fallback for a column holding both
integers and doubles is to encode the whole column as strings, so an
attribute like gdp_md_est in tests/ne_110m_admin_0_countries turned into
values like "904.200000".
Summarize each attribute across the layer first and pick one type that can
hold all of its values, so mixed integer and floating point columns become
doubles. Only columns that mix numbers with strings, or booleans with
numbers, still fall back to strings, which is as close as MLT's typed
columns can get.
Also stop encoding JSON-object-valued attributes as MLT struct columns.
Struct children can only be strings, and a struct column is flattened into
"column name + child name" when it is read back, so an attribute `meta`
holding {"en": "one", "de": "eins"} decoded as separate `metaen` and
`metade` attributes. Nested JSON now stays JSON text, the way MVT
carries it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LxhELiLtpUFwrPSWwTYdHG
Claude