Upgrade Guide

Upgrading from v8.4 to v8.5

Transpilation

The module entry point is now only lightly transpiled for the most commonly used evergreen browsers. This change offers significant savings on bundle size. If your application needs to support older browsers such as IE 11, make sure to include node_modules in your babel config.

Upgrading from v7.3 to v8.0

The key goals of luma.gl v8.0 were to simplify the core architecture and significantly improve performance. The number of modules has been reduced from 14 to 9, and they are now structured around layers of abstraction (high-level to low-level) to clarify the relationship between them. Unfinished and rarely-used components have been removed, reducing the complexity of the API and leading to significant performance gains (see What's New) for details).

Module Restructure

The module structure has been significantly changed for v8.0 with the intention of clarifying the purpose of each module and the relationships between them.

Breaking changes

• Texture2D's unpackFlipY option is removed. This change ensures that all data sources (Image, ImageBitmap, typed array) are treated consistently. As a result, textures created from Image objects and URL strings are now y-flipped from the v7.3 default. To get the old behavior, specify the pixelStore option:

new Texture2D({
  data,
  pixelStore: {
    [GL.UNPACK_FLIP_Y_WEBGL]: true
  }
});

• createGLContext will no longer attempt to create a headlessgl context under node.js. One can either create headless context externally and use instrumentGLContext to prepare it for use with luma.gl or use createHeadlessContext from @luma.gl/test-utils to create a headlessgl context.
• registerShaderModules has been removed. Modules can be imported and used directly where necessary.
• createShaderHook and createModuleInjection have been removed. Use ProgramManager.getDefaultProgramManger(gl).addShaderHook && the shader module inject field instead.
• ProgramManager.getDefaultProgramManger(gl).addModuleInjection been removed. Use the shader module inject field instead.
• getParameter and setParameter have been removed. Use getParameters and setParameters instead.
• The following are no longer exported by @luma.gl/core, but can still be imported from the modules indicated:

Smaller changes

• Functions are no longer accepted as uniform values to the Model class. The same effect can be achieved by updating the uniform values each frame prior to drawing.
• BaseModel and Model have been consolidated in Model. Model can be used as a substitute for BaseModel where necessary.
• AmbientLight, DirectionalLight, PointLight, PhongMaterial, PBRMaterial, CameraNode have been removed from @luma.gl/core. These were either empty classes or simple data objects and so can be replaced by plain JavaScript objects in most cases.
• ShaderCache has been removed and superseded by ProgramManager.
• VertexArray.getDrawParams no longer takes overrides as an argument. The calling function can manually override values as needed.
• @luma.gl/main has been removed. Use individual modules instead.
• Multipass classes have been removed.
• Seer support has been removed.
• Timeline and Keyframes have been moved from @luma.gl/addons to @luma.gl/engine.

Upgrading from v7.2 to v7.3

ProgramManager has replaced ShaderCache in the Model class as a more robust resource manager. Use of the ShaderCache with Model will not affect functionality in any way, but it is now a no-op.

Upgrading from v6.x to v7.0

luma.gl v7.0 represents a major overhaul of the API. The majority of changes are in areas that are only infrequently used by applications, and the intention is that most applications should only require very light porting.

Core API Removals

Loading Functions Removed

Extensive loading functionality is now provided by a new companion framework loaders.gl and because of this, most of the limited legacy luma.gl loading functions have been removed.

For the most common case (loading of images for use with textures), loading functions are no longer needed as the data prop in the Texture2D constructor now accepts url strings and Promise objects (this is the new Async Textures function).

Attribute Class Removed

This experimental class has been moved to deck.gl and is now an internal class. Attribute accessor API improvements in luma.gl v7 should cover any issue.

WebGL API Removals

Sampler Class Removed

The Sampler class has been removed as its utility was limited and it added complexity to the library. It may be added back in the future if a clear use case arises.

Texture2DArray Class Removed

The Texture2DArray class has been removed as its utility was limited and the status of support was unclear due to limited testing. It may be added back in the future if a clear use case arises.

FenceSync Class Removed

The FenceSync class has been removed as its utility was limited. It may be added back in the future if a clear use case arises. If required, syncing can be done directly through the WebGLFenceSync object.

Framebuffer and Texture: Copy and Blit methods

Following member function of Framebuffer and Texture classes are no longer supported, instead use the corresponding new global methods:

Parameters have also changed in some cases, see separate section.

Module Structure Changes

Debug functionality moved to separate npm module

To reduce bundle size and increase separation of concerns, debug functionality is now more cleanly separated from the core library and needs to be imported from a separate npm module:

To upgrade, install the new module

npm install @luma.gl/debug

And replace

import 'luma.gl/debug';

with

import '@luma.gl/debug';

Model

Changes:

• Model no longer extends ScenegraphNode. This ensures that applications that do not need scenegraph support do not need to include scenegraph related code. Use the new ModelNode class to inject Models into scenegraphs.

Deletions:

• Redraw flag handling has been removed: Model.setNeedsRedraw() and Model.getNeedsRedraw().

Additions:

• A new Model.isAnimated() method is provided, indicating that redraws are required every frame.

Geometry

The Geometry class has been simplified and is now a conceptually "immutable" class that holds typed arrays and accessor metatadata describing attributes for a geometry.

Buffer

Framebuffer

To maximize rendering performance, the default framebuffer is no longer preserved between frames.

The most common use case for preserving the draw buffer is capturing canvas contents into an image via toDataURL. This can now be done via AnimationLoop.toDataURL which returns a Promise that resolves to the canvas data URL:

dataURL = await animationLoop.toDataURL();
snapshotImage.src = dataURL;

More generally, moving code that depends on canvas contents to the end of onRender, after all draw operations, will ensure that canvas contents are available.

Prior behaviour can re-enabled using the glOptions argument to the createGLContext or AnimationLoop constructors:

new AnimationLoop({
  glOptions: {
    preserveDrawingBuffer: true
  }
});

Note that setting preserveDrawingBuffers may result in a performance drop on some platforms.

Query

Use Query.getTimerMilliseconds to retrieve timer results in milliseconds. Query.getResult now returns raw query results.

To improve performance and simplify the library, support for tracking Query instances with promises has changed: The Query constructor no longer takes onComplete and onError callbacks, and pollGLContext has been removed. Instead Query.createPoll now provides a simple, optional promise-based API.

Copy And Blit Parameter Unification

Names of certain parameters to these methods have been unified in an effort to reduce confusion and use the same conventions across all functions implementing image read-back, copy or blit.

This table lists parameter mapping between old and new function.

Geometry Scenegraph Models

Geometry scenegraph models have been deprecated. Simply create a Model or ModelNode and explicitly pass a Geometry instance as an argument, e.g.:

const sphere = new Model(gl, {
  geometry: new SphereGeometry({
    nlat: 30,
    nlong: 30,
    radius: 2
  })
});

Upgrading from v5.3 to v6.0

luma.gl v6.0 underwent a major API cleanup, resulting in a smaller, easier-to-learn API and smaller application bundles. While there are many smaller changes, the impact on most applications should be limited:

• Most removed functions were in practice rarely used by applications, and the impact on typical luma.gl applications should be limited.
• A number of API changes are related to moving attribute management from Program to VertexArray, however for higher level applications that work with the Model class rather than Program directly, there should not be much impact.

GL Constants Import Path

The biggest change for many apps will probably be that the static GL symbol (that contains all WebGL 2 constants) must now be separately imported GL from 'luma.gl/constants'.

Experimental Exports: New Naming Convention

Experimental exports are now prefixed with underscore (_). The experimental "name space" export has been removed.

// NOW: luma.gl v6
import {_Attribute as Attribute} from 'luma.gl';

// BEFORE: luma.gl v5.x
import {experimental} from 'luma.gl';
const {Attribute} = experimental;

This change will enable tree-shaking bundlers to remove unused experimental exports, resulting in smaller final application bundles.

Removed symbols

Math functions were moved from luma.gl to the separate math.gl module in v4.1. As of v6.0, they are no longer forwarded by luma.gl and now need to be imported directly from math.gl:

import {radians, degrees, Vector2, Vector3, Vector4, Matrix4} from '@math.gl/core';

luma.gl v6.0 removes a number of previously deprecated symbols. luma.gl will now issue an error rather than a warning if the old usage is detecated.

Constants

Context

Global Functions

AnimationLoop

Program

TransformFeedback

VertexArray

Upgrading from v5.2 to v5.3

v5.3 deprecates a number of symbols. It is recommended that you replace their usage in your source code.

Upgrading from v5.1 to v5.2

Running under Node.js

Using with Node: "import luma.gl/headless" is no longer required for luma.gl to load headless gl and the usage has been deprecated. You can now simply remove any such import statements from your code.

Using Debug Contexts

Debugging: The Khronos group's WebGLDeveloperTools are automatically installed when luma.gl is installed, but are not actually bundled into the application unless explicitly imported. This avoids impacting the size of production bundles built on luma.gl that typically do not need debug support.

To use debug support, first import the debug tools, then call getDebugContext to create a debug contexts from a normal WebGL context:

import 'luma.gl/debug';
const gl = getDebugContext(gl);

Upgrading from v4 to v5

Please read this documentation before upgrading your luma.gl dependency from v4 to v5. In v5 a number of previously deprecated features have been removed and a number of additional deprecations have been made at the same time.

Before upgrading to v5, it is highly recommended to run your application using latest v4 release, and check the console for any deprecated warnings, if there are any replace deprecated API with newer API as listed below.

Model Class

The Model constructor expects a gl context as the first argument.

// v5
Model(gl);
Model(gl, {...opts});
Model(gl, {program});

Following style construction was deprecated in v4 and is now removed in v5.

// NOT SUPPORTED
Model({gl});
Model({gl, ...opts});
Model({program});

useDevicePixelRatio

useDevicePixelRatio is used as a an argument in AnimationLoop class constructor and pickModels method. It is now deprecated in v5, but still supported with a warning message and will be removed in next major version update. It is recommended to use useDevicePixels instead.

Geometry

Geometry class construction with inline attributes was deprecated in v4 and now removed in v5.

// NOT SUPPORTED
new Geometry({
  positions: new Float32Array([ ... ]),
  colors: {
    size: 4,
    value: new Float32Array([ ... ])
  }
});

All attributes should be grouped inside attribute object.

// SUPPORTED
new Geometry({
 attributes: {
   positions: new Float32Array([ ... ]),
   colors: {
     size: 4,
     value: new Float32Array([ ... ])
   }
 }
});

Removed Features

Following features were deprecated in v3 and v4 are now removed in v5.

• Global symbols:

• Texture construction options:

Upgrading from v3 to v4

luma.gl v4 is a major release with API changes. Please read this documentation before upgrading your luma.gl's dependency from v3 to v4. In addition, a number of previously deprecated features have been removed and a number of additional deprecations have been made at the same time in this version.

Removed Features

Some previously deprecated classes and functions have been removed in luma.gl v4 and applications must be updated with the new classes and functions if they are still using these.

Deprecated Features

Some classes and functions have been deprecated in luma.gl v4. They will continue to function in v4, but a warning in the console will be generated. These functions are expected to be removed in a future major versions of luma.gl.

API Change

Model Class

The Model constructor now expects a gl context as the first argument.

// v3
Model({gl});
Model({gl, ...opts});
Model({program});

// v4
Model(gl);
Model(gl, {...opts});
Model(gl, {program});

the gl context used to be extracted from the supplied program or provided along side with other options, but in luma.gl v4, it is expected as a separate argument to the constructor. This change is because luma.gl v4 emphasizes sharing shaders rather than programs (often indirectly via shader caching / shader assembly), it is less common that a gl context is available.

Upgrading from V2 to V3

V3 was a fairly minor release, a number of deprecations were made.

Deprecations