Use only bounds for geojson diff tile reload #6800

HarelM · 2025-12-02T11:08:53Z

Launch Checklist

Fixes geoJsonSource.updateData not working properly after updating to 5.13.0 #6795

This changes the logic of tile refreshing for geojson
The idea behind this fix:

When updating the data, we know exactly which geometries were changed, so we use this information to return the relevant bounds of these geometries
It simplifies to only using bounds and not using ids of previous features

Confirm your changes do not include backports from Mapbox projects (unless with compliant license) - if you are not sure about this, please ask!
Briefly describe the changes in this PR.
Link to related issues.
Write tests for all new functionality.
Add an entry to CHANGELOG.md under the ## main section.

HarelM · 2025-12-02T11:15:34Z

@lucaswoj @wayofthefuture let me know what you think about the direction of this PR before I invest more time in fixing the tests and adding new tests.

codecov · 2025-12-02T11:24:25Z

Codecov Report

❌ Patch coverage is 92.59259% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.41%. Comparing base (50d412d) to head (e430c63).

Files with missing lines	Patch %	Lines
src/source/geojson_source.ts	94.11%	1 Missing ⚠️
src/source/geojson_worker_source.ts	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6800      +/-   ##
==========================================
- Coverage   92.42%   92.41%   -0.01%     
==========================================
  Files         288      288              
  Lines       23807    23803       -4     
  Branches     5056     5056              
==========================================
- Hits        22003    21998       -5     
- Misses       1804     1805       +1

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

🚀 New features to boost your workflow:

📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

…plibre-gl-js into 6795-use-only-bounds

Add a newline at the end of the HTML file.

wayofthefuture · 2025-12-02T12:52:39Z

I like the way this is headed.

However, let's say a feature was moved from 1 tile to another tile. How do you know to invalidate the previous tile without the feature id? The new bounds would not affect the previous tile.

HarelM · 2025-12-02T12:56:30Z

If a feature was moved from one tile to the other - I have the previous geometry just before I apply the update, and the new geometry.
So I "invalidate" tiles that have one of those geometries.

HarelM · 2025-12-02T13:51:38Z

I've gone and updated the relevant tests. I've removed some tests that are "duplicated" in the sense that they are being tested in lat lng bounds and there's no need to retest them in geojson source.
I've added tests that are using only public API to mimic how the class is behaving.
There's still a weird issue with the "covering world tile", which causes a test to fail.
I added a similar test to lnglatbounds tests, which also fails to highlight the issue in a simpler test.
@lucaswoj, do let me know what you think about this test and if there's something wrong about it.

lucaswoj · 2025-12-02T14:54:53Z

Please hold off on merging these changes until I have time to review later today. I think this introduces some new bugs.

HarelM · 2025-12-02T14:59:50Z

Sure, take your time. I wouldn't merge this before you review it.

src/source/geojson_source.ts

lucaswoj · 2025-12-02T16:29:33Z

src/source/geojson_source.ts

+        if (tile.state === 'unloaded') {
+            return false;
        }
-


Removing this codepath will cause performance regressions in some cases. Although having two codepaths is ugly, invalidating tiles by id is much faster and more accurate than doing so by bounds.

src/source/geojson_source.ts

lucaswoj · 2025-12-02T16:36:06Z

Thanks for diving into this code @HarelM! This does fix the bug but it may regress performance in some cases. I've proposed another fix at #6801, ready for your review, that is performance neutral.

HarelM · 2025-12-02T18:13:57Z

Can we measure how much performance degradation this is causing?
This code is a lot simpler to maintain and doesn't relay on feature index (including the timing related to it, which I find hard to make sure won't break in the future).

lucaswoj · 2025-12-02T18:29:59Z

Can we measure how much performance degradation this is causing?

There's not a single number that captures the whole picture.

We can say that #6562 would no longer apply to sources loaded from a URL, so updating 1 point out of 100k points would become ~100% slower after this PR merges.

#6562 would also no longer apply in some cases involving more complex geometries, like an L-shaped line that crosses three out of 4 tiles in the viewport. We could create another benchmark that would become ~100% slower after this PR merges, for a non-URL-backed source.

HarelM · 2025-12-02T18:34:36Z

Fair enough, can you take a look at the lat lang bounds failing test? I'm not sure I understand why it's failing.

HarelM · 2025-12-02T18:38:09Z

It's worth mentioning that for tiles with a lot of features and small updates to the geojson, this route has better performance.
I'm not sure what I think about loading the geojson into memory when using url to improve performance, but that's also a possible route here.

lucaswoj · 2025-12-02T18:47:54Z

Fair enough, can you take a look at the lat lang bounds failing test? I'm not sure I understand why it's failing.

Just created #6802 to investigate

HarelM · 2025-12-02T20:03:31Z

@mwilsnd does the fix related to geojson undefined properties has to do with the feature index etc?
I'm wondering if the fix here is helps with your fix or if they are not related.

mwilsnd · 2025-12-02T20:25:56Z

Take a look: #6803 - this just changes how the layers are constructed in the feature index for geojson sources, I'm not sure if this is fully related.

HarelM · 2025-12-02T21:11:03Z

The features from the geojson are sent back for every tile, so it might be somewhat related, I'm not sure, I need to wrap my head around it...

wayofthefuture · 2025-12-03T19:08:44Z

src/source/geojson_source.ts

-     * Refresh all tiles that PREVIOUSLY DID contain these feature ids.
-     */
-    prevIds: Set<GeoJSON.Feature['id']>;
+    affectedBounds: LngLatBounds[];


Since affectedBounds is the only property, would it be better to remove shouldReloadTileOptions and simply pass affectedBounds in the data event?

I think keeping it this way allows adding properties in the future if needed, so I'm fine with keeping it that way.
I had the same thought myself, but decided to keep it as is.

Let me know if the change I made here makes sense now: 755ec5f

src/source/geojson_source.ts

src/source/geojson_source_diff.ts

wayofthefuture · 2025-12-03T19:17:05Z

src/source/geojson_source_diff.ts

        for (const feature of diff.add) {
            const id = getFeatureId(feature, promoteId);
            if (id != null) {
+                affectedGeometries.push(feature.geometry);


Currently updateable allows the addition of duplicate (replacing) feature ids, so you might want to consider checking/adding existing affected feature geometry here...

Can you clarify a bit what you mean here?
I can move the push before the "if", if that's what you mean.

src/source/geojson_source_diff.ts

wayofthefuture

After giving this a look I can say that I definitely prefer this solution. Considering it's a geojson source, and geojson supports string ids, the fact that this also supports string ids just makes sense. It also saves us (and many) a breaking refactor to eliminate the warning. (correct me if I'm wrong).

Also, @HarelM's implementation is similar to my implementation used in the differential GeoJSON-VT PR that is pending. In this PR, updates are applied to the source and affected features are returned as well. Not sure if that would benefit in the future, but I like the idea of both sides using similar methods.

HarelM · 2025-12-04T06:29:36Z

Thanks for the review!
Note that I updated my initial message as feature updates were supported in the previous version, I just read the code wrong.

There are some performance changes (in some flows they are improved while in others they degrade).
I can't say which is better.
I can say that I think this approach is a bit less fragile as it doesn't relay on feature index.

I'll work on the review comments nevertheless.

wayofthefuture · 2025-12-04T15:45:37Z

#6809

HarelM added 2 commits December 2, 2025 13:08

Use only bounds for tile reload

cce1503

Fix lint

e5caba2

HarelM mentioned this pull request Dec 2, 2025

geoJsonSource.updateData not working properly after updating to 5.13.0 #6795

Open

HarelM added 7 commits December 2, 2025 13:33

Update min test

bb535d1

Remove console log

c9799d5

Revert accidental file commit.

fcf2d11

revert display-a-map file

811a12c

Reduce indentation

82cfc63

Merge branch '6795-use-only-bounds' of https://github.com/maplibre/ma…

b74b747

…plibre-gl-js into 6795-use-only-bounds

Fix missing newline at end of display-a-map.html

bc023ca

Add a newline at the end of the HTML file.

Updated tests

e6398a6

HarelM added 4 commits December 2, 2025 15:55

Remove loaded and move it to beforeEach

d13c69a

Remove white space

574e03d

Update changelog

2d67669

Fix lint

6f9fd25

lucaswoj reviewed Dec 2, 2025

View reviewed changes

src/source/geojson_source.ts Show resolved Hide resolved

lucaswoj reviewed Dec 2, 2025

View reviewed changes

src/source/geojson_source.ts Show resolved Hide resolved

lucaswoj reviewed Dec 2, 2025

View reviewed changes

src/source/geojson_source.ts Show resolved Hide resolved

lucaswoj mentioned this pull request Dec 2, 2025

Make LngLatBounds#intersects return true when bounds touch along an edge or at a corner #6802

Merged

8 tasks

Merge branch 'main' into 6795-use-only-bounds

be30613

HarelM marked this pull request as ready for review December 3, 2025 07:34

wayofthefuture reviewed Dec 3, 2025

View reviewed changes

src/source/geojson_source.ts Outdated Show resolved Hide resolved

wayofthefuture reviewed Dec 3, 2025

View reviewed changes

src/source/geojson_source_diff.ts Outdated Show resolved Hide resolved

wayofthefuture reviewed Dec 3, 2025

View reviewed changes

src/source/geojson_source_diff.ts Show resolved Hide resolved

wayofthefuture reviewed Dec 3, 2025

View reviewed changes

HarelM added 3 commits December 4, 2025 08:30

Merge branch 'main' into 6795-use-only-bounds

e1593bf

Code review changes

755ec5f

Fix tests, fix lint

e430c63

lucaswoj mentioned this pull request Dec 4, 2025

Send GeoJSON data from worker to main if loaded from a URL #6811

Open

8 tasks

Uh oh!

Use only bounds for geojson diff tile reload #6800

Are you sure you want to change the base?

Use only bounds for geojson diff tile reload #6800

Uh oh!

Conversation

HarelM commented Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Launch Checklist

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

codecov bot commented Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

wayofthefuture commented Dec 2, 2025

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

lucaswoj commented Dec 2, 2025

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

Uh oh!

lucaswoj Dec 2, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

lucaswoj commented Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

lucaswoj commented Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

HarelM commented Dec 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

lucaswoj commented Dec 2, 2025

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

mwilsnd commented Dec 2, 2025

Uh oh!

HarelM commented Dec 2, 2025

Uh oh!

wayofthefuture Dec 3, 2025

Choose a reason for hiding this comment

Uh oh!

HarelM Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

HarelM Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

wayofthefuture Dec 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HarelM Dec 4, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

wayofthefuture left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HarelM commented Dec 4, 2025

Uh oh!

wayofthefuture commented Dec 4, 2025

Uh oh!

HarelM commented Dec 2, 2025 •

edited

Loading

codecov bot commented Dec 2, 2025 •

edited

Loading

lucaswoj commented Dec 2, 2025 •

edited

Loading

lucaswoj commented Dec 2, 2025 •

edited

Loading

HarelM commented Dec 2, 2025 •

edited

Loading

wayofthefuture Dec 3, 2025 •

edited

Loading

wayofthefuture left a comment •

edited

Loading