Integrate Harffbuzz freetype to render Khmer Burmese and hindi

[alanchenboy]

refactor for #1289

This is a demo of how to use harfbuzz to render complex text.

Example style:
https://alanchenboy.github.io/harfbuzzResource/hindi.json.

Screenshots:
.

https://github.com/maplibre/maplibre-native/assets/5135389/f8f294ec-1d20-4106-b745-51d5f4a1af1d.

https://github.com/maplibre/maplibre-native/assets/5135389/7e3857e9-74c4-405e-b12c-e5722cf76919.

[codecov]

Codecov Report

Attention: Patch coverage is 93.22382% with 33 lines in your changes are missing coverage. Please review.

Project coverage is 85.78%. Comparing base (04be6ec) to head (6b02cba).

❗ Current head 6b02cba differs from pull request most recent head d65c81d

Please upload reports for the commit d65c81d to get more accurate results.

Files	Patch %	Lines
src/mbgl/text/shaping.cpp	85.56%	14 Missing ⚠️
src/mbgl/text/harfbuzz.cpp	78.94%	8 Missing ⚠️
src/mbgl/text/glyph_manager.cpp	92.59%	4 Missing ⚠️
src/mbgl/text/glyph.cpp	91.66%	2 Missing ⚠️
src/mbgl/layout/layout.hpp	50.00%	1 Missing ⚠️
src/mbgl/layout/symbol_layout.cpp	98.90%	1 Missing ⚠️
src/mbgl/storage/resource.cpp	87.50%	1 Missing ⚠️
src/mbgl/tile/geometry_tile_worker.cpp	96.77%	1 Missing ⚠️
src/mbgl/util/i18n.cpp	91.66%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1439       +/-   ##
===========================================
+ Coverage   59.30%   85.78%   +26.48%     
===========================================
  Files         580      566       -14     
  Lines       28674    28188      -486     
  Branches    11276        0    -11276     
===========================================
+ Hits        17006    24182     +7176     
+ Misses       4139     4006      -133     
+ Partials     7529        0     -7529     

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

[alanchenboy]

Swipe out macos end is such a pain for me, I have to use my personal computer to fix unit tests

[ntadej]

I wonder if we want to make this an optional feature? This would allow us to fix Qt later for example.

Also what is the binary size increase? (@louwers, didn't we have a CI test for that in the past? I guess the rest of the actions need to pass.)

[alanchenboy]

https://github.com/maplibre/maplibre-native/actions/runs/5747452827/job/15578480347 does anybody known what is the reason for this failure?

[alanchenboy]

I didn't have linux end, will fix linux CI in next weeks.

[louwers]

@alanchenboy That means the iOS render tests are failing. Did you run them locally?

[alanchenboy]

Pipelines fixed， lucky me. let me add some complex text render tests to finish this PR.

[louwers]

@ramSeraph

This is surprising.. I do mean pulling in data, not code. I was asking this because a lot of harfbuzz binary size might be coming from unicode data it loads, which can be pulled in lazily.

That is an interesting idea.

Doesn't maplibre-native already have a plugin architecture? Can this particular feature be moved to a plugin?

Not for the C++ Core at the moment.

@alanchenboy Congrats! 💯 Can you merge main in?

[wipfli]

If this is ready for review then I would propose to have a discussion in the style spec repo about the changes needed to the style spec.

[robinhood245]

@alanchenboy Thank you so much for the effort you are putting in. I was hoping to try it out, but whenever I am trying to build from the code for Android, it's failing. Can you please help me build from the code?

[alanchenboy]

@alanchenboy Thank you so much for the effort you are putting in. I was hoping to try it out, but whenever I am trying to build from the code for Android, it's failing. Can you please help me build from the code?

Hi @robinhood245 the pipeline is passed so I think the CMake change is correct, but the new vendor requires higher CMake version, so if you installed cmake 3.18 in your Android studio, it will use that and cause the build failure. I update cmakeVersion in dependencies.gradle to from 3.18.1+ to 3.19.1+. And the app can build by android studio.
So maybe we should also update this file, or we build our own cmake file to build freetype and harfbuzz, cause most vendors are using Maplibe cmake instead of vendors' cmake file.

I roll back cmake upgrade, and I can compile the branch in my Android studio (it failed last time because of VPN setting)

[alanchenboy]

https://app.codecov.io/gh/maplibre/maplibre-native/commit/2f76cee32dff34d10f43f8abc4985f77a9578680 How can I fix this pipeline, is it a requirement for approve?

[louwers]

It means there are some code paths that are not reached during testing.

It's not a hard requirement at the moment.

[maxammann]

Just came across this? What's the status of this?

I believe the testing coverage of the layouting is not very good overall, so I think this should not be a blocker. But maybe I just have not yet found comprehensive tests for the symbol_layout.cpp file :)

[alanchenboy]

@maxammann
I think this approach needs to edit the style specification to support downloading font files via style config. Before this done we should not merge this PR.
I think @wipfli once helped me edit the specification, any update?

[1ec5]

edit the style specification to support downloading font files via style config

See also the discussion in maplibre/maplibre-gl-js#4564 (comment). Maybe we should coordinate on a strategy for making font selection more flexible across platforms.

[louwers]

Ideally we should create an issue (and then a PR) in the MapLibre Style Spec repo to achieve consensus on the style spec changes and have it officially changed before merging.

Edit: as Oliver mentioned:

If this is ready for review then I would propose to have a discussion in the style spec repo about the changes needed to the style spec.

@alanchenboy Could you create an issue there? It would be great to get this merged.

[alanchenboy]

Hi @louwers here is the issue on style spec repo: maplibre/maplibre-style-spec#868.