Skip to main content

Upgrade Guide

Upgrading to v4.2

  • It is recommended (but not required) to add dataType and batchType fields to any custom loaders.

Upgrading to v4.1

  • WKBLoader/TWKBLoader/HexWKBLoader - The default shape is now geojson-geometry rather than binary-geometry. If you were relying on binary-geometry, just add add a shape: 'binary-geometry' option, as in load(..., WKBLoader, {wkb: {shape: 'binary-geometry}}).
  • The geometry shape is deprecated, and now called geojson-geometry.

Upgrading to v4.0

Node.js v18+

When using on Node.js your application should import the module before calling any functions.

Typed Loaders

Loaders now return typed data. This addition of types into previously untyped code can generated type errors in existing applications. These type errors can just be a result of the typescript compiler making sure your application knows what it is doing, but they can also mean that you have been making wrong or unsafe assumptions about what is being returned from the loader in question. We recommend you review the errors rather than just defeat them with comments as some may be valid and should be fixed in the application.

Shape of Returned Data

Some loaders can return multiple formats, often controlled with the loader options shape parameter. Note that many returned data types now also include a shape field which contain a string value the specifies the shape of the data. This goal is that this should result in a "discriminated union". Switching on the returnedData.shape field will then allow typescript to correctly determine which type of data was returned.

Apache Arrow JS now imports apache-arrow v13 which is a major upgrade but Apache Arrow JS v9 introduces breaking change (compared with Apache Arrow v4 which is used by v3.x_).

If your application is using the Apache Arrow API directly to work with Apache Arrow tables returned from, note that the Apache Arrow v9 API contains a number of breaking changes.

On the upside, the new Apache Arrow API is more modular and "tree shakeable" (meaning that only the Apache Arrow functionality your application is actually using is included in your application bundle).

Since Apache Arrow JS does yet not come with upgrade notes, you can refer to the Arrow documentation.

Table Schemas

If you are referencing table schemas returned by loaders, they will no longer be Apache Arrow schemas, but instead equivalent "serialized" lower-overhead schemas. You can recover Arrow schemas as follows

import {deserializeArrowSchema} from '';
const table = load(url, ParquetLoader);
const arrowSchema = deserializeArrowSchema(table.schema);


If you were relying on module to install a global fetch() function under Node.js that supported fetching from local files. v4 uses the built-in fetch in Node.js v18+ (which doesn't support fetching from local files), so fetch from local files, you now need to use fetchFile() instead.

import {fetchFile} from '';
const response = await fetchFile('/path/to/local/file');

Note that fetchFile is called by all core load() functions unless the fetch function is overridden through loader options.

Details: The expectation is that v4.0+ will be used with Node.js version 18 and higher, which now provide a built-in browser-compatible fetch() function by default. This new built-in Node.js fetch function does not support reading from the file system, and v4.0 aligns with this practice.

Loader module changes, in order of estimated impact to applications:

  • loadImage() has moved to

  • GLTFLoader - no longer post processes data. Applications need to import and call the postProcessGLTF function after calling the loader to get the same result.

  • All hashes now require an encoding parameter. To get previous behavior, just specify .hash...(..., 'base64').

  • Batches now contain a Table with a single RecordBatch (instead of just a RecordBatch).

  • options.geopackage.shape replaces all other format specification optiojs, such as options.gis.format
  • options.geopackage.shape: 'tables', The default data format returned is now tables which returns the type Tables<ObjectRowTable>, where the data of each table is an array of GeoJSON features. (The Tables and ObjectRowTable types are exported from You can use options.geopackage.shape: 'geojson-table'.

  • options.kml.shape replaces all other mechanisms for specifying format of returned data (.format etc), and aligns with the geojson-table table shape as this is compatible with a GeoJSON FeatureCollection.
  • options.gpx.shape same changes as options.shape.kml
  • options.tcx.shape same changes as options.shape.kml

Upgrading to v3.4

This module is still marked as experimental and had some breaking changes.

  • WMSService class
    • The srs parameters has been renamed to crs in alignment with the most recent WMS 1.3.0 conventions.
  • ``WMSCapabilitiestype (returned byWMSServiceandWMSCapabilitiesLoader`)
    • WMSCapabilities.layer is now WMSCapabilities.layers
    • WMSCapabilities.boundingBox is now WMSCapabilities.geographicBoundingBox (in generic lng/lats) and WMSCapabilities.boundingBoxes (array of bounding boxes in supported projections)

Upgrading to v3.2

  • The default data format returned is now options.gis.format: 'tables', which returns the type Tables<ObjectRowTable>, where the data of each table is an array of GeoJSON features. (The Tables and ObjectRowTable types are exported from You can use options.gis.format: 'geojson' to return an object mapping table names to an array of GeoJSON features.

Upgrading to v3.0

Platform Support Updates

  • Node.js 10 is no longer supported (LTS maintenance of Node.js 10 ended in May 2021).
  • The module entry point in the published module is no longer transpiled to ES5 as modern JavaScript is supported now support recent evergreen browsers resulting bundle size reduction and better performance. The main entry point is still fully transpiled to es5, you can select this entry point using e.g. the mainField option in webpack.

Worker Concurrency

Default number of worker threads for each loader has been reduced from 5 to 3 on non-mobile devices and to 1 on mobile devices to reduce memory use. Generally, increasing the number of workers has diminishing returns.

  • Passing fetch options (such as headers) to load() and parse() etc. should now be done via the options.fetch options object (e.g. options.fetch.headers). Specifying fetch options on the root options object was deprecated in v2.3 and is no longer supported in v3.0.

  • The Transform API in v2.3 has been replaced with Compression classes that handle both compression and decompression. Please refer to the documentation.

  • The Transform API in v2.3 has been replaced with Hash classes that handle both compression and decompression. Please refer to the documentation.

  • The format of parsed data (i.e. whether table rows are objects or arrays) is now controlled by options.csv.shape instead of being dynamically selected based on whether the CSV file has a header row. The default shape is now object-row-table. Default column names are generated if no header row is present. This is a breaking change when loading CSV files without header rows.
  • Duplicate column names will have a counter suffix added to ensure that they are unique. In rare cases this could be a breaking change.

  • GLTFScenegraph is updated to provide modifying capabilities. Signatures of some methods have been changed to use named parameters (rather than positional parameters).
  • The deprecated GLBBuilder class and encodeGLTFSync functions have now been removed.

  • Module has been moved to

  • The texture API loadImage, loadImageArray, loadImageCube has been moved to the new module, and have been renamed to loadImageTexture*.
  • The binary image API has been consolidated in a single function getBinaryImageMetadata():
  • A number of previously deprecated exports have been removed:
getBinaryImageSize(arrayBuffer)getBinaryImageMetadata(arrayBuffer)?.{width, height}
HTMLImageLoaderUse ImageLoader with options {image: type: 'image'}
getSupportedImageType(imageType?) NA

  • The KMLLoader, GPXLoader, and TCXLoader now require a value for options.gis.format. Previously, the lack of a value would return data in "raw" format, i.e. not normalized to GeoJSON. To return GeoJSON-formatted data, use options.gis.format: 'geojson'. Other options are binary and raw.
  • The kml.normalize option has been deprecated. When options.gis.format is geojson, coordinates will always be in longitude-latitude ordering.

  • createWorker() now creates a generic worker. For loader workers use the new createLoaderWorker() function.

  • Added featureIds attribute to i3s tile content. It is an array of feature ids which specify which feature each vertex belongs to. Can be used for picking functionality.

  • The (undocumented) module has been renamed to

Upgrading to v2.3

  • selectLoader() is now async and returns a Promise that resolves to a loader.
  • selectLoaderSync() is available for situations when calling an async function is inconvenient.
  • Passing fetch options to load() and parse() etc. should now be done via the options.fetch sub-options object. fetch options on the root object are now deprecated.

  • The KMLAsGeoJsonLoader has been removed, use KMLLoader, with options.gis.format: 'geojson'.

Upgrading to v2.2

  • selectLoader is no longer experimental. If you were using the experimental export, replace _selectLoader with selectLoader. Also note that argument order has changed and now aligns with load and parse
  • parseInBatchesSync has been removed, all batched parsing is now performed asynchronously.

Some iterator utilities that are mostly used internally have been changed.

FunctionReplacement / Status
makeChunkIteratorcombined into makeIterator
makeStreamIteratorcombined into makeIterator
getStreamIteratorDeprecated in 2.1, now removed in 2.2
contatenateAsyncIteratorDeprecated in 2.1, now removed in 2.2

  • Header auto-detection now requires options.csv.header to be set to 'auto' instead of undefined. 'auto' is the new default value for this option, so this change is unlikely to affect applications.

  • The experimental json._rootObjectBatches option is now deprecated. Use the top-level metadata: true option instead. Note that the batchType names have also changed, see the JSONLoader docs for details.

The experimental streaming _PLYStreamingLoader has been removed. Use the non-streaming PLYLoader instead.

The new function getBinaryImageMetadata() replaces isBinaryImage(), getBinaryImageSize() and getBinaryImageMIMEType(). The old functions are now deprecated, but still available.

Upgrading to v2.1

Some iterator helper functions have been renamed, the old naming is now deprecated.

Old NameNew Name

  • Experimental exports have been removed JSONParser, StreamingJSONParser, ClarinetParser.

The experimental ImageLoaders for individual formats introduced in 2.0 have been removed, use ImageLoader for all formats.

  • getImageData(image) now returns an object with {data, width, height} instead of just the data array. This small breaking change ensures that the concept of image data is consistent across the API.
  • ImageLoader: options.image.type: The html and ndarray image types are now deprecated and replaced with image and data respectively.

Tileset3DLoader and Tile3DLoader are replaced by Tiles3DLoader, which supports loading both a 3D tileset file and a tile. Check for loaded data format.

Upgrading to v2.0

Version 2.0 is a major release that consolidates functionality and APIs, and a number of deprecated functions have been removed.

Some general changes:

  • All exported loader and writer objects now expose a mimeType field. This field is not yet used by but is available for applications (e.g. see selectLoader).
  • All (non-worker) loaders are now required to expose a parse function (in addition to any more specialized parseSync/parseText/parseInBatches functions). This simplifies using loaders without, which can reduce footprint in small applications.

TextEncoderUse global TextEncoder instead and if needed
TextDecoderUse global TextDecoder instead and if needed
createReadStreamfetch().then(resp => resp.body)

ImageHTMLLoaderImageLoader with options.images.format: 'image'
ImageBitmapLoaderImageLoader with options.images.format: 'imagebitmap'
decodeImageparse(arrayBuffer, ImageLoader)
getImageMetadatagetBinaryImageMIMEType + getBinaryImageSize

Loader Objects

  • Loaders can no longer have a loadAndParse method. Remove it, and just make sure you define parse on your loaders instead.

The GLTFLoader now always uses the new v2 parser, and the original GLTFParser has been removed.


Note that automatic packing of binary data (aka "packed JSON" support) was only implemented in the v1 GLTFLoader and has thus also been removed. Experience showed that packing of binary data for .glb files is best handled by applications.

GLTFLoader option changes

The foillowing top-level options are deprecated and will be removed in v2.0

Removed OptionReplacementDescriptions
gltf.parserVersionN/ANo longer needs to be specied, only the new gltf parser is available.
fetchLinkedResourcesgltf.fetchBuffers, gltf.fetchImages
createImagesN/AImages are now always created when fetched
decompressgltf.decompressMeshesDecompress Draco compressed meshes (if DracoLoader available).
DracoLoaderN/ASupply DracoLoader to parse, or call registerLoaders(pDracoLoader])
postProcessgltf.postProcessPerform additional post processing before returning data.
uribaseUriAuto-populated when loading from a url-equipped source
fetchN/Afetch is automatically available to sub-loaders.


Loader Objects

  • Loaders no longer have a loadAndParse removed. Just define parse on your loaders.

Upgrading from v1.2 to v1.3

  • As with v1.1, GLTFLoader will no longer return a GLTFParser object in v2.0. A new option options.gltf.parserVersion: 2 is provided to opt in to the new behavior now.

Upgrading from v1.0 to v1.1

A couple of functions have been deprecated and will be removed in v2.0. They now emit console warnings. Start replacing your use of these functions now to remove the console warnings and ensure a smooth future upgrade to v2.0.

Also, Node support now requires installing before use.

  • Removal: Node support for fetchFile now requires importing before use.
  • Removal: Node support for TextEncoder, and TextDecoder now requires importing before use.
  • Deprecation: TextEncoder and TextDecoder will not be exported from in v2.0.

  • Removal: Node support for images now requires importing before use.

  • Deprecation: GLBParser/GLBBuilder - These will be merged into GLTF classes..
  • Deprecation: GLTFParser/GLTFBuilder - The new GLTF class can hold GLTF data and lets application access/modify it.
  • Deprecation: GLTFLoader will no longer return a GLTFParser object in v2.0. Instead it will return a pure javascript object containing the parse json and any binary chunks. This object can be accessed through the GLTF class. Set options.GLTFParser to false to opt in to the new behavior now.


First official release of