Rough Notes on Porting “glTF Viewer” from Mixed Reality Toolkit (MRTK) V1 to MRTK V2 (RC2.1)

NB: The usual blog disclaimer for this site applies to posts around HoloLens. I am not on the HoloLens team. I have no details on HoloLens or Azure Mixed Reality other than what is on the public web and so what I post here is just from my own experience experimenting with pieces that are publicly available and you should always check out the official developer site for the product documentation.

Around 6 months ago, I wrote a simple application for HoloLens 1 and published it to the Windows Store.

It’s called “glTF Viewer” and it provides a way to view models stored in glTF format on the HoloLens with basic move, rotate, scale manipulations. It also provides a way via which one user can put such a model onto their HoloLens, open it up and then share it automatically to other users on the same local network such that they will also be able to see the same model and the manipulations performed on it. This includes downloading the files for the model from the originating device and caching them onto the requesting device.

You can find the application in the store here;

glTF Viewer in the Microsoft Store

and you can find the original blogpost that I wrote about the process of writing this application here;

A Simple glTF Viewer for HoloLens

and you can find the source code for the application over here;

glTF Viewer on GitHub

I’d like to keep this application up to date and so with the arrival of MRTK V2 (release candidates) I thought that it would be a good idea to port the application over to MRTK V2 such that the application was “more modern” and better suited to work on HoloLens 2 when the device becomes available.

In doing that work, I thought it might be helpful to document the steps that I have taken to port this application and that’s what this blog post is all about – it’s a set of ‘rough notes’ made as I go through the process of moving the code from V1 to V2.

Before beginning, though, I want to be honest about the way in which I have gone about this port. What I actually did was;

  1. Begin the port thinking that I would write it up as I went along.
  2. Get bogged down in some technical details.
  3. Complete the port.
  4. Realise that I had not written anything much down.

So it was a bit of a failure in terms of writing anything down.

Consequently, what I thought that I would do is to revisit the process and repeat the port from scratch but, this time, write it down Smile as I went along.

That’s what the rest of this post is for – the step-by-step process of going from MRTK V1 to MRTK V2 on this one application having done the process once already.

Before I get started though, I’d like to point out some links.

Some Links…

There are a number of links that relate to activities and reading that you can do if you’re thinking of getting started with a mixed reality application for HoloLens 2 and/or thinking of porting an existing application across from HoloLens 1. The main sites that I find myself using are;

Armed with those docs, it’s time to get started porting my glTF Viewer to MRTK V2.

Making a New Branch, Getting Versions Right

I cloned my existing repo from using a recursive clone and made sure that it would still build.

There are quite a few steps necessary to build this project right now described in the readme at

Specifically, the repo contains a sub-module which uses UnityGLTF from the Kronos Group. There’s nothing too unusual about that except that the original MRTK also included some pieces around GLTF which clashed with UnityGLTF and so I had to write some scripts to as to set a few things up and remove one or two toolkit files in order to get things to build.

I described this process in the original blog post under the section entitled ‘A Small Challenge with UnityGLTF’.

One of the expected benefits of porting to MRTK V2 with its built-in support for GLTF is to be able to get rid of the sub-module and the scripts needed to hack the build process and end up with a much cleaner project all round Smile 

I made a new branch for my work named V2WorkBlogPost as I already had the V2Work branch where I first tried to make a port and from which I intend to merge back into master at some later point.

With that branch in play, I made sure that I had the right prerequisites for what I was about to do, taking them from the ‘Getting Started’ page here;

  • Visual Studio 2017.
    • I have this although I’m actually working in 2019 at this point.
  • Unity 2018.4.x.
    • I have 2018.4.3f1 – I have a particular interest in this version because it is supposed to fix a (UWP platform) issue which I raised here where the UWP implementations of System.IO.File APIs got reworked in Windows SDK 16299 which broke existing code which used those file APIs. You can see more on that in the original blog post under the title “Challenge 3 – File APIs Change with .NET Standard 2.0 on UWP”. It’s nice that Unity has taken the effort to try and fix this so I’ll be super keen to try it out.
  • Latest MRTK release.
    • I took the V2.0.0 RC2.1 release and I only took the Foundation package rather than the examples as I do not want the examples in my project here. Naturally, I have the examples in another place so that I can try things out.
  • Windows SDK 18362+.
    • I have 18362 as the latest installed SDK on this machine.

It is worth noting at this point a couple of additional things about my glTF Viewer application as it is prior to this port;

  • It has already been built in a Unity 2018.* version. It was last built with 2018.3.2f1.
  • It is already building on the IL2CPP back-end

Why is my application already building for Il2CPP?

Generally, I would much prefer to work on the .NET back-end but it has to be acknowledged that IL2CPP is inevitable given that Unity 2019 versions no longer have the .NET back-end support but there is a bigger reason for my use of IL2CPP. My application using classes from .NET Standard 2.0 (specifically HttpListener) and due to the deprecation of the .NET back-end Unity did not add support for .NET Standard 2.0 into the .NET back-end. So, if I want to use HttpListener then I have to use IL2CPP. I wrote about this in gory detail at the time that I wrote the application so please refer back to the original blog post  (in the section entitled ‘Challenge Number 1 – Picking up .NET Standard 2.0 ’) if you want the blow-by-blow.

So, armed with the right software and an application that already builds in Unity 2018 on the IL2CPP back-end, I’m ready to make some changes.

Opening the Project in Unity

I opened up my project in the 2018.4.3f1 version of Unity and allowed it to upgrade it from 2018.3.2f1.

I didn’t expect to see problems in this upgrade but it did seem to get stuck on this particular error;


which says;

“Project has invalid dependencies:
    com.unity.xr.windowsmr.metro: Package [com.unity.xr.windowsmr.metro@1.0.10] cannot be found”

so my best thought was to use the Package Manager which offered to upgrade this to Version 1.0.12


and that seemed to do the trick. I had a look at my build settings as well and switched platform over to the UWP;


A quick note on the debugging settings here. For Il2CPP, you can either choose to debug the C# code or the generated C++ code and Unity has all the details over here.

UWP: Debugging on IL2CPP Scripting Backend

Take extra care to ensure that you have the right capabilities set in your project for this to work as mentioned in the first paragraph of that page.

Because of this, I generally build Release code from Visual Studio and attempt to use the Unity C# debugging first. If that doesn’t help me out, I tend to debug the generated C++ code using the native debugger in Visual Studio and, sometimes, I rebuild from Visual Studio in Debug configuration to help with that debugging on native code.

I’d note also that I do toggle “Scripts Only Build” when I think it is appropriate in order to try and speed up build times but it’s “risky” as it’s easy to leave it on when you should have turned it off so beware on that one Smile

With that done, Unity was opening my project in version 2018.4.3f1 and it would build a Visual Studio project for me and so I committed those changes and moved on.

The commit is here.

A Word on Scenes

An important thing to note about the glTF Viewer application is that it’s really quite simple. There’s a bit of code in there for messaging and so on but there’s not much to it and, as such, it’s built as a single scene in Unity as you can see below;


If you have a multi-scene application then you’re going to need to take some steps to work with the MRTK V2 across those multiple scenes to ensure that;

  1. The MRTK doesn’t get unloaded when scenes change
  2. More than one MRTK doesn’t get loaded when scenes change

I’ve seen a few apps where this can be a struggle and there’s an issue raised on the MRTK V2 around this over here with a long discussion attached which I think leads to the approach of having a “base” scene with the MRTK embedded into it and then loading/unloading scenes with the “additive” flag set but you might want to check out that whole discussion if this is an area of interest for you as it doesn’t impact my app here.

Adding the New Toolkit

This is much easier than the previous 2 steps in that I just imported the Unity package that represents MRTK V2 RC 2.1.

I hit one error;

“Assembly has reference to non-existent assembly ‘Unity.TextMeshPro’ (Assets/MixedRealityToolkit.SDK/MixedRealityToolkit.SDK.asmdef)”

but that was easily fixed by going back into the Package Manager and installing the Text Mesh Pro package into my project and I, once again, ensured that the project would build in Unity. It did build but it spat out this list of “errors” that I have seen many times working on these pieces so I thought I would include a screenshot here;


These errors all relate to the “Reference Rewriter” and all seem to relate to System.Numerics and I have seen these flagged as errors by Unity in many projects recently and yet the build is still flagged as Succeeded and seems to deploy and work fine on a device.

Consequently, I ignore them although the last error listed there about a failure to copy from the Temp folder to the Library folder is an actual problem that I have with Unity at the moment and I have to fix that one by restarting the editor and the hub until it goes away Confused smile

When it did go away, I then hit this error;

“Scripted importers UnityGLTF.GLTFImporter and Microsoft.MixedReality.Toolkit.Utilities.Gltf.Serialization.Editor.GlbAssetImporter are targeting the glb extension, rejecting both.

but I can fully understand why Unity is complaining here because I do have two versions of UnityGLTF in the project right now so I’m not surprised that Unity is a bit puzzled but I’m hoping to address this shortly and Unity seems to be tolerating the situation for now and so, with those caveats, I do now have a project that contains both the old MRTK V1 and the new MRTK V2 as below;


The big question for me at this point is whether to take a dependency on the MRTK V2 as a Git sub-module or whether to just include the code from the MRTK V2 in my Unity project.

I much prefer to take a dependency on it as a sub-module but I figure that while it is not yet finished I will have the code in my project and then I can do the sub-module step at a later point. Consequently, I had quite a lot of folders to add to my Git repo and it leaves my repo in a slightly odd state because the MRTK V1 is in there as a sub-module and the MRTK V2 is in there as code but I’m about to remove MRTK V1 anyway so it won’t be in this hybrid state for too much longer.

The commit is here.

Removing the MRTK V1 – Surgical Removal or the Doomsday Option?

I now have a project with both the MRTK V1 and the MRTK V2 within it but how do I go about removing the V1 and replacing it with the V2?

So far when I’ve worked on applications that are doing this it feels to me like there are 2 possibilities;

  1. The “Doomsday” option – i.e. delete the MRTK V1 and see what breaks.
  2. The “Surgical” option – i.e. make an inventory of what’s being used from the MRTK V1 and consider what replacement is needed.

For the blog post, I’m going to go with option 2 but I’ve seen developers try both approaches and I’m not convinced that one is any better than the other.

In my particular application, I did a survey of my scene to try and figure out what is being used from the toolkit.

Firstly, I had some objects in my scene which I think I used in their default configuration;

  • Cursor object
  • InputManager object
  • MixedRealityCameraParent object

I’m expecting all of these to be replaced by the MRTK V2 camera system and input system without too much effort on my part.

I also noticed that I had a ProgressIndicator. At the time of writing, I’m asking for this to be replaced into the MRTK V2 but it’s not there as far as I know and so my expectation here is to simply keep these pieces from the MRTK V1 in my application for now and continue to use the progress indicator as it is.

Having taken a look at my scene, I wanted to see where I was using the MRTK V1 from my own code. My first thought was to attempt to use the “Code Map” feature of Visual Studio but I don’t think there’s enough “differentiation” between my code and the code in the toolkit to be able to make sense of what’s going on.

Abandoning that idea, I looked at the entire set of my scripts that existed in the scripting project;


There are around 30 or so scripts there, it’s not huge and so I opened them all up in the editor and searched for HoloToolkit in all of them and I came up with a list of 8 files;


I then opened those files and did a strategic search to try and find types from the HoloToolkit and I found;

  • A use of the interface IFocusable in FocusWatcher.cs a class which was trying to keep track of which (if any) object has focus.
  • A use of the ObjectCursor in a class CursorManager.cs which tried to make the cursor active/inactive at suitable times, usually while something was asynchronously loading.
  • The ModelUpdatesManager class which adds the type TwoHandManipulatable to a GameObject such that it can be moved, rotated, scaled and this class needs a BoundingBox prefab in order to operate.
  • A use of the ProgressIndicator type which I use in order to show/hide progress when a long running operation is going on.

Additionally, I know that I am also using UnityGLTF from the Kronos repo in order to load GLTF models from files whether they be JSON/binary and whether they be an object packaged into a single file or into multiple files which all need loading.

The application also makes use of voice commands but I know that in the MRTK V1 I had to avoid the speech support as it caused me some issues. See back to the original blog post under the section entitled “Challenge 7” for the blow-by-blow on problems I had using speech as pre

While it’s probably not a perfect list, this then gives me some things to think about – note that I am mostly building this list by looking back at the porting guide and finding equivalents for the functionality that I have used;

  1. Input – Replace the Cursor, InputManager, MixedRealityCameraParent in the scene with the new MRTK systems.
  2. Speech – Look into whether speech support in MRTK V2 works better in my scenario than it did in MRTK V1.
  3. GLTF – Replace the Unity GLTF use from the Kronos repo with the new pieces built into MRTK V2.
  4. Focus – Replace the use of IFocusable with the use of IMixedRealityFocusHandler.
  5. Cursor – Come up with a new means for showing/hiding the cursor across the various pointers that are used by the MRTK V2.
  6. Manipluations – Replace the TwoHandManipulatable script with use of the new ManipulationHandler, NearInteractionGrabbable and BoundingBox scripts with suitable options set on them.
  7. Rework – Look into which pieces of the application could benefit from being reworked, re-architected based on the new service-based approach in MRTK V2.

That’s a little backlog to work on and I’ll work through them in the following sub-sections.


Firstly, I removed the InputManager, Cursor and MixedRealityCameraParent from my scene and then used the Mixed Reality Toolkit –> Add to Scene and Configure menu to add the MRTK V2 into the scene. At this point, the “Mixed Reality Toolkit” menu is a little confusing as both the MRTK V1 and V2 are contributing to it but, for now, I can live with that.

I chose the DefaultHoloLens2ConfigurationProfile for my toolkit profile as below;


A word about “profiles”. I think it’s great that a lot of behaviour is moving into “profiles” or what an old-fashioned person like me might call “configuration by means of a serialized object” Smile

The implication of this though is that if you were to lose these profiles then your application would break. I’ve seen these profiles be lost more than once by someone who allowed them to be stored in the MRTK folders (by default the MixedRealityToolkit.Generated folder) themselves & then deleted one version of the MRTK in order to add another losing the MixedRealityToolkit.Generated folder in the process.

Additionally, imagine that in one of today’s Default profiles a setting is “off”. What’s to say that a future profile won’t replace it with a value of “on” and change your application behaviour?

Maybe I’m just paranoid Winking smile but my way of managing these profiles is to create a “Profiles” folder of my own and then duplicate every single profile that is in use into that folder and give it a name that lines up with my app. That way, I know exactly where my profiles are coming from and I don’t run the risk of deleting them by mistake or having them overwritten by a newer toolkit.

While doing this, I noticed that the DefaultMixedRealityToolkitConfigurationProfile allows for “copy and customize”;


whereas the DefaultHoloLens2ConfigurationProfile doesn’t seem to;


but I might be missing how this is supposed to work. Regardless, I started with the DefaultMixedRealityToolkitConfigurationProfile and I cloned it to make a copy in Profiles\GLTFViewerToolkitConfigurationProfile.

I then went through that profile and;

  • Changed the Target Scale to be World.
  • Changed the Camera profile to be the DefaultHoloLens2CameraProfile before cloning that to make Profiles\GLTFViewerCameraProfile
  • Changed the Input profile to be the DefaultHoloLens2InputSystemProfile before cloning that to make Profiles\GLTFViewerInputSystemProfile
    • In doing this, I cloned all of the 8 sub-sections for Input Actions, Input Action Rules, Pointer, Gestures, Speech Commands, Controller Mapping, Controller Visualization, Hand Tracking
  • I switched off the Boundary system, leaving it configured with its default profile
  • I switched off the Teleport system, leaving it configured with its default profile
  • I switched off the Spatial Awareness system, leaving it with its default profile and removing the spatial observer (just in case!)
  • I cloned the DefaultMixedRealityDiagnosticsProfile to make my own and left it as it was.
  • I cloned the Extensions profile to make my own and left it as it was.
  • I left the editor section as it was.

With that in place, I then have all these profiles in my own folder and they feel like they are under my control.


At this point, I thought I’d risk pressing “Play” in the editor and I was surprised that I didn’t hear the welcome message that I had built into the app but, instead, spotted a “not implemented exception”.

Speech and Audio, Editor and UWP

I dug into this exception and realised that I had written a class AudioManager which decides whether to play voice clips or not and that class had been built to work only on UWP devices, not in the editor – i.e. it was making use of ApplicationData.Current.LocalSettings so I quickly tried to rewire that in order to use PlayerPrefs instead so that it could work both in editor and on device.

With that done, I got my audible welcome message on pressing play, I could see the framerate counter from the MRTK V2 and I seemed to be able to move around in the editor.

I couldn’t open any files though because I’d also written some more code which was editor specific.

My application uses voice commands but I had a major challenge with voice commands on the MRTK V1 in that they stopped working whenever the application lost/regained focus.

Worst of all this included when the application lost focus to make use of the file dialog so a user of the application was able to use the voice command “Open” to raise the file dialog, thereby breaking the voice commands before their model file had even been chosen.

I wrote about this in the original blog post under the section “Challenge 7”. The upshot is that I removed anything related to MRTK V1 speech or Unity speech from my application and I fell back to purely using SpeechRecognizer from the UWP for my application and that worked out fine but, of course, not in the Unity editor.

I only have 3 speech commands – open, reset, remove and so what I would ideally like to do is to work in the way of MRTK V2 by defining new input actions for these commands along with a profiler command to toggle the profile display as below in my input actions profile;


and then I could define some speech commands in my speech settings profile;


and then in my class which handles speech commands, I could add a property to map the MixedRealityInputAction (open etc.) to a handler using my own internal class ActionHandler because I don’t think Unity can serialize dictionaries for me;


and then configure them to their respective values in the editor…


and then I should be able to implement IMixedRealityInputActionHandler to invoke the actions here (rather than directly tie myself to those actions coming from only voice commands);


In doing so, I think I also need to register my GameObject as a “global” handler for these commands and so I need to add a call to do;


and that seemed to work really, really nicely.

That said, I am still pretty concerned that this isn’t going to work on the device itself reliably across invocations of the file dialog as I see the new WindowsSpeechInputProvider implementation using the KeywordRecognizer and I’m not sure that this type behaves well on the device when the application loses/gains focus.

Consequently, I figured that I would use all of this MRTK V2 infrastructure to deliver speech commands to me in the editor but, on the device, I would like to switch it off and rely on the mechanism that I’d previously built which I know works.

I edited my Input system profile in order to try and remove the WindowsSpeechInputProvider outside of the editor and I disabled the WindowsDicationInputProvider altogether;


and I then changed my startup code such that it did different things depending on whether it was in the editor or not;


and my own speech handling code is super, super simple and inefficient but I know that it works on a V1 device so I am trying to largely keep intact and here it is below – it essentially keeps creating a SpeechRecognizer (UWP not Unity) and using it for a single recognition before throwing it away and starting again;

    /// <summary>
    /// Why am I using my own speech handling rather than relying on SpeechInputSource and
    /// SpeechInputHandler? I started using those and they worked fine.
    /// However, I found that my speech commands would stop working across invocations of
    /// the file open dialog. They would work *before* and *stop* after.
    /// I spent a lot of time on this and I found that things would *work* under the debugger
    /// but not without it.
    /// That led me to think that this related to suspend/resume and perhaps HoloLens suspends
    /// the app when you move to the file dialog because I notice that dialog running as its
    /// own app on HoloLens.
    /// I tried hard to do work with suspend/resume but I kept hitting problems and so I wrote
    /// my own code where I try quite hard to avoid a single instance of SpeechRecognizer being
    /// used more than once - i.e. I create it, recognise with it & throw it away each time
    /// as this seems to *actually work* better than any other approach I tried.
    /// I also find that SpeechRecognizer.RecognizeAsync can get into a situation where it
    /// returns "Success" and "Rejected" at the same time & once that happens you don't get
    /// any more recognition unless you throw it away and so that's behind my approach.
    /// </summary>
    async void StartSpeechCommandHandlingAsync()
        while (true)
            var command = await this.SelectSpeechCommandAsync();

            if (command.Action != MixedRealityInputAction.None)
                // Just being paranoid in case we start spinning around here
                // My expectatation is that this code should never/rarely
                // execute.
                await Task.Delay(250);
    async Task<SpeechCommands> SelectSpeechCommandAsync()
        var registeredCommands = MixedRealityToolkit.InputSystem.InputSystemProfile.SpeechCommandsProfile.SpeechCommands;

        SpeechCommands command = default(SpeechCommands);

        using (var recognizer = new SpeechRecognizer())
                new SpeechRecognitionListConstraint(registeredCommands.Select(c => c.Keyword)));

            await recognizer.CompileConstraintsAsync();

            var result = await recognizer.RecognizeAsync();

            if ((result.Status == SpeechRecognitionResultStatus.Success) &&
                ((result.Confidence == SpeechRecognitionConfidence.Medium) ||
                 (result.Confidence == SpeechRecognitionConfidence.High)))
                command = registeredCommands.FirstOrDefault(c => string.Compare(c.Keyword, result.Text, true) == 0);
        return (command);

I suspect that I’ll be revisiting this code once I try and deploy to a device but, for now, it works in the editor and moves me onto my next little challenge.

I also switched off the frame rate profiler by default in the profile;


and implemented my handler to toggle it on/off;


Opening File Dialogs

My application has, initially, a single voice command, “Open”, which raises a file dialog in order to open a glTF model.

However, I’d only written the file open code in order to support opening the file dialog on a UWP device. I hadn’t done the work to make it open in the editor and I realised that this needed addressing so I quickly amended the method that I have to add an additional piece of code for the non-UWP platform case;

    async Task<string> PickFileFrom3DObjectsFolderAsync()
        var filePath = string.Empty;

        var known3DObjectsFolder = KnownFolders.Objects3D.Path.ToLower().TrimEnd('\\');

            filePath = await FileDialogHelper.PickGLTFFileAsync();

            if (!string.IsNullOrEmpty(filePath) &&
                filePath = string.Empty;
        } while (filePath == string.Empty);
        filePath = EditorUtility.OpenFilePanelWithFilters(
            "Select GLTF File",
            new string[] { "GLTF Files", "gltf,glb", "All Files", "*" });

        return (filePath);

but I found that even if I could raise the file dialog, I was still getting exceptions opening files…

Loading GLTF Models

The problem that I was hitting was that the GLTFParser was struggling to read the files that I was feeding it and so I decided to take the leap to stop using that code and start using the GLTF code bundled into the MRTK V2.

In the existing code, I make use of a class GLTFSceneImporter to load the one or more files that might make up a GLTF model. In my original blog post I had a few struggles using this in a deterministic way as it’s very coroutine based and I found it hard to be in control of a couple of things;

  • Knowing when it had finished
  • Knowing when it had thrown exceptions

I mentioned these challenges in the original post under the title of “A Small Challenge with Async/Await and CoRoutines” and also “Another Small Challenge with CoRoutines and Unity’s Threading Model”.

At the time, I largely worked around them by writing a base class named ExtendedMonoBehaviour which did some work for me in this regard. It’s in the repo so I won’t call it out in any detail here.

The GLTFSceneImporter delegated the responsibility for actually opening files to an implementation of an interface named ILoader which looks as below;

namespace UnityGLTF.Loader
	public interface ILoader
		IEnumerator LoadStream(string relativeFilePath);

		void LoadStreamSync(string jsonFilePath);

		Stream LoadedStream { get; }

		bool HasSyncLoadMethod { get; }

This was very useful for me as the user might choose to open a multi-file GLTF file with various separate material files etc. and this is the way in which my code gets to “know” which files have actually been opened. I need this list of files to be able to offer the model over HTTP to other devices that might request it in a shared experience.

In order to use this, I had a class RecordingFileLoader which implemented this ILoader interface and kept track of every file that it successfully opened on behalf of the loader and I passed this around into a couple of places that needed to know about the file list.

Looking at the new MRTK V2 support for GLTF, things seem much improved in that there is a new class GltfUtility which seems to offer an ImportGltfObjectFromPathAsync method. The built-in support for async makes my base class ExtendedMonoBehaviour redundant Smile but it does leave me with the challenge of trying to figure out how to know which files the code has actually loaded a model from.

That method returns a GltfObject and I wrote some code which attempts to work out which files loaded by interrogating the buffers property after it has been populated. I already had this class ImportedModelInfo which wrapped around my RecordingFileLoader and so I modified it to take on this extra functionality;

using Microsoft.MixedReality.Toolkit.Utilities.Gltf.Schema;
using System.Collections.Generic;
using System.IO;
using System.Linq;
using UnityEngine;

public class ImportedModelInfo
    public ImportedModelInfo(
        string fullFilePath,
        GltfObject gltfObject)
        // Where were these files loaded from?
        this.BaseDirectoryPath = Path.GetDirectoryName(fullFilePath);

        // What's the name of the file itself?
        this.relativeLoadedFilePaths = new List<string>();

        // Note: At the time of writing, I'm unsure about what the URI property
        // might contain here for buffers and images given that the GLTF spec
        // says that it can be file URIs or data URIs and so what does the GLTF
        // reading code return to me in these cases?

        // I'm expecting Uris like 
        //  foo.bin
        //  subfolder/foo.bin
        //  subfolder/bar/foo.bin

        // and will probably fail if I encounter something other than that.
        var definedUris =
                .Where(b => !string.IsNullOrEmpty(b.uri))
                .Select(b => b.uri)
                    .Where(i => !string.IsNullOrEmpty(i.uri))
                    .Select(i => i.uri));


        this.GameObject = gltfObject.GameObjectReference;
    public string BaseDirectoryPath { get; private set; }
    public IReadOnlyList<string> RelativeLoadedFilePaths => this.relativeLoadedFilePaths.AsReadOnly();
    public GameObject GameObject { get; set; }

    List<string> relativeLoadedFilePaths;

with the reworking of one or two other pieces of code that then allowed me to delete my classes RecordingFileLoader and ExtendedMonoBehaviour which felt good Smile

I had to do another slight modification to code which had never been run in the editor before because it was expecting to export world anchors but, other than that, it was ok and I could now load at least one GLTF model in the editor as below;


What I couldn’t do was any kind of manipulations on the object so that was perhaps where I needed to look next although I suspect that depends on focus and I also suspect it relies on having a collider which might not be present…

The commit for these pieces is here.


The earlier code would attach this behaviour;

using HoloToolkit.Unity.InputModule;
using UnityEngine;

public class FocusWatcher : MonoBehaviour, IFocusable
    public void OnFocusEnter()
        focusedObject = this.gameObject;
    public void OnFocusExit()
        focusedObject = null;
    public static bool HasFocusedObject => (focusedObject != null);
    public static GameObject FocusedObject => focusedObject;
    static GameObject focusedObject;

to the models that had been loaded such that when voice commands like “reset” or “remove” were used, the code could check the HasFocusedObject property, get the FocusedObject value itself and then would typically look for some other component on that GameObject and make a method call on it to reset its position or remove it from the scene.

It’s questionable as to whether this behaviour should be attached to the objects themselves or whether it should just be a global handler for the whole scene but the effect is the same either way.

I need an equivalent in the new MRTK V2 and the natural thing to do would seem to be to reach into the MixedRealityToolkit.InputSystem.FocusProvider and make a call to GetFocusedObject() but that method expects that the caller knows which pointer is in use and I’m not sure that I do.

Instead, I chose to just update the existing class so as to implement IMixedRealityFocusHandler and keep doing what it had been doing before;

using HoloToolkit.Unity.InputModule;
using Microsoft.MixedReality.Toolkit.Input;
using UnityEngine;

public class FocusWatcher : MonoBehaviour, IMixedRealityFocusHandler
    public void OnFocusEnter(FocusEventData eventData)
        focusedObject = this.gameObject;
    public void OnFocusExit(FocusEventData eventData)
        focusedObject = null;
    public static bool HasFocusedObject => (focusedObject != null);
    public static GameObject FocusedObject => focusedObject;
    static GameObject focusedObject;

but I noticed that I still wasn’t able to interact with the duck – there’s still work to be done Smile

The commit for this stage is here.


My class which manipulates the cursor for me was still stubbed out and so I attempted to update that from what it had been;

using HoloToolkit.Unity.InputModule;
using UnityEngine;

public class CursorManager : MonoBehaviour
    private ObjectCursor cursor;

    public void Show()
    public void Hide()

to this version;

using Microsoft.MixedReality.Toolkit;
using Microsoft.MixedReality.Toolkit.Input;
using System.Collections.Generic;
using System.Linq;
using UnityEngine;

public class CursorManager : MonoBehaviour
    public CursorManager()
        this.hiddenPointers = new List<IMixedRealityPointer>();
    public void Hide()
        // TODO: I need to understand how you are supposed to do this on V2, I just want
        // to switch all cursors off when the user cannot do anything useful with them.
        foreach (var inputSource in MixedRealityToolkit.InputSystem.DetectedInputSources)
            foreach (var pointer in inputSource.Pointers)
                if ((pointer.IsActive) && (pointer.BaseCursor != null))
        MixedRealityToolkit.InputSystem.GazeProvider.Enabled = false;
    public void Show()
        foreach (var pointer in this.hiddenPointers)

        MixedRealityToolkit.InputSystem.GazeProvider.Enabled = true;
    List<IMixedRealityPointer> hiddenPointers;

I’m not sure whether this is “right” or not – once again I find myself puzzling a little over all these pointers and cursors and trying to figure which ones I’m meant to interact with but the code feels reasonably “safe” in that it attempts to put back what it did in the first place so, hopefully, I’m not breaking the toolkit with this.

That commit is here.


Up until now, I’ve left the code which attempts to handle manipulations as it was. That is, there is code in the application;


which attempts to add TwoHandManipulatable to a model which has been loaded from the disk (rather than one which has been received over the network where I don’t allow local manipulations). That TwoHandManipulatable wants a BoundingBoxPrefab and so you can see that my code here has passed such a thing through to it.

It’s probably not too surprising that this isn’t working as it’s mixing MRTK V1 classes with MRTK V2 in the scene so I wouldn’t really expect it to do anything.

Additionally, I’m not sure from looking at the objects in the editor that there is any type of collider being added by the glTF loading code so I probably need to deal with that too.

I suspect then that I’m going to need to add a few pieces here;

  • A BoxCollider to allow for interactions on the model.
  • ManipulationHandler to allow the model to be moved, rotated, etc.
  • NearInteractionGrabbable so that the manipulations cater for both near and far interactions on a HoloLens 2.
  • BoundingBox to provide some visualisation of the interactions with the model.

Additionally, I think that I’m going to want to be able to have quite a bit of control over the settings of some of the materials etc. on the BoundingBox and some of the axes of control on the other pieces and so it feels like it might be a lot easier to set this all up as a prefab that I can build in the editor and then just pass through to this code.

Previously, when loading a model my code took an approach of something like this;

  • load the GLTF model, giving a new GameObject with a collider already on it
  • create a new object to act as the model’s parent, parenting this object itself off some root parent within the scene
  • position the parent object 3m down the user’s gaze vector, facing the user
  • attach a world anchor to the parent object both for stability but also so it can be exported to other devices
  • add manipulation behaviours to the GLTF model itself so that it can be moved, rotated, scaled underneath its parent which is anchored

I decided to change this slightly for the new toolkit to;

  • load the GLTF model, giving a new GameObject ( M )
  • create a new object ( A ) to act as the anchored parent
  • create a new object to act as the model’s parent ( P ) from a prefab where BoxCollider, ManipulationHandler, NearInteractionGrabbable, BoundingBox are already present and configured on that prefab
  • parent M under P, P under A, A under R
  • add a world anchor to A

and that lets me slip this prefab into the hierarchy like adding an item into a linked-list so as to let the prefab bring a bunch of behaviour with it.

That prefab is as below;


and I tweaked a few materials and settings both on the BoundingBox largely based on examples that I looked at in the example scenes from the toolkit;




Changing the hierarchy of the components that are set up when a model is loaded into the scene had some impact on my scripts which create/access world anchors and on my scripts which tried to watch for object transformations to send/receive over the network and so I had to make a few related changes here to patch that up and pass a few objects to the right place but I’ll keep that detail out of the post.

It also broke my simplistic FocusWatcher class because that class expected that the GameObject which had focus would be the model itself with direct excess to various behaviours that I have added to it whereas, now, that object is buried in a bit of hierarchy and so I got rid of the FocusWatcher altogether at this point and tried to write this method which would hopefully return to me all focused objects which had a particular component within their hierarchy;

    IEnumerable<T> GetFocusedObjectWithChildComponent<T>() where T : MonoBehaviour
        // TODO: I need to figure whether this is the right way to do things. Is it right
        // to get all the active pointers, ask them what is focused & then use that as
        // the list of focused objects?
        var pointers = MixedRealityToolkit.InputSystem.FocusProvider.GetPointers<IMixedRealityPointer>()
            .Where(p => p.IsActive);

        foreach (var pointer in pointers)
            FocusDetails focusDetails;

            if (MixedRealityToolkit.InputSystem.FocusProvider.TryGetFocusDetails(
                pointer, out focusDetails))
                var component = focusDetails.Object?.GetComponentInChildren<T>();

                if (component != null)
                    yield return component;

whether this is a good thing to do or not, I’m not yet sure but for my app it’s only called on a couple of voice commands so it shouldn’t be executing very frequently.

I tried this out in the editor and I seemed to be at a place where I could open GTLF models and use near and far interactions to transform them as below;


the commit for this stage is here.

Removing the MRTK V1

At this point, I felt like I was done with the MRTK V1 apart from the ProgressRingIndicator which I am still using so I need to preserve it in my project for now.

I made a new folder named TookitV1 and I moved across the Progress related pieces which appeared to be;

  • Animations – the contents of the Progress folder
  • Fonts – I copied all of these
  • Materials – I copied only ButtonIconMaterial here
  • Prefabs – the contents of the Progress folder
  • Scripts – the contents of the Progress folder

I did a quick commit and then deleted the HoloToolkit folder and I also deleted the UnityGLTF folder as I should, at this point, not be using anything from those 2 places.

At this point, the ProgressIndicator blew up compiling and told me that it was missing the HoloToolkit.Unity namespace (easily fixed) and that it wanted to derive from Singleton<T> but I found that easy enough to fix by just changing the base class to MonoBehaviour and adding a static Instance property which was set to the first instance which spun up in the application.

I still had problems though in that I had a couple of missing scripts in the prefab for the ProgressIndicator and I tried to replicate what had been there previously with the SolverHandler and Orbital as below


and I had to patch a couple of materials but, other than that, the MRTK V1 was gone and the app seemed to continue to function in the editor.

The commit is here.

Removing MRTK V1 and UnityGLTF as Submodules

I had previously included the MRTK V1 and UnityGLTF github repos as submodules of my repo and I no longer need them so removing them would make the repo a lot cleaner.

Additionally, I had a setup.bat script which attempted to move a lot of files around, do some preliminary building of Unity GLTF etc. and I no longer need that either.

I should be in a state on this branch where the project can “simply” be pulled from github and built.

With that in mind, I attempted to remove both of those submodules following the procedure described here as I’ve done this once or twice but I can never remember how you’re meant to do it.

I also removed the setup.bat and altered the

Now, usually, when I do so many things at once some thing goes wrong so the next step was to…

Make a Clean Folder, Clone the Repo, Fix Problems

I cloned the repo again recursively into a new, clean folder with git clone –recursive and then switched to the V2WorkBlogPost and I noticed that git struggled to remove the MixedRealityToolkit-Unity and the UnityGLTF folders which had been created/populated as part of bringing down the recursive repo so I got rid of them manually (I’ll admit that the finer details of submodules are a bit of a mystery to me).

I reopened that project in Unity and, remarkably, all seemed to be fine – the project ran fine in the editor once I’d switched platforms & I didn’t seem to have missed files from my commits.

The commit is here.

Deploying to a Device

At this point, it felt like it was time to build for a device and see how the application was running as I find that there are often pieces of functionality that work ok in the editor but fail on a device.

I only have a HoloLens 1 device with me at the time of writing and so I used HoloLens 1, I can’t try on HoloLens 2 right now.

In trying to build for the device I hit an immediate failure;

“IOException: Sharing violation on path C:\Data\temp\blogpost\GLTF-Model-Viewer\GLTFModelViewer\Temp\StagingArea\Data\Managed\tempStrip\UnityEngine.AudioModule.dll”

but I see this quite frequently with Unity at the moment and so did a quick re-start (and shutdown Visual Studio) but then I got hit with the;

“Copying assembly from ‘Temp/Unity.TextMeshPro.dll’ to ‘Library/ScriptAssemblies/Unity.TextMeshPro.dll’ failed”

which is another transient error I see quite a lot so I did some more restarts (of both Unity and the Unity Hub) and managed to produce a successful VS build which seemed to deploy ok and run fine;


In deploying to the device, I also did some basic tests of the multi-user network sharing functionality which also seemed to be working fine.

Other Rework – Mixed Reality Extension Services

There are a few places in this code base where I make use of “services” which are really “global” across the project. As examples;

  • I have a class StorageFolderWebServer which, in a limited way, takes a UWP StorageFolder and makes some of its content available over HTTP via HttpListener
  • I have a NetworkMessageProvider which facilitates the shared experience by multicasting and receiving New Model, Transformed Model, Deleted Model messages around the local network.
    • This sits on top of a MessageService which simply knows how to Send/Receive messages having initially joined a multicast group.
  • I have a MessageDialogHelper which shows message boxes without blowing up the Unity/UWP threads.
  • I have a FileDialogHelper which shows a file dialog without blowing up the Unity/UWP threads.

Most of these are probably static classes but I feel that they are really providing services which may/not have some configurable element to them and which other pieces of code just need to look up somewhere in a registry and make use of thereby allowing them to be replaced at some point in the future.

As the MRTK V2 provides a form of service registry via the means of “extensions” to the toolkit, I thought it would make sense to try that out and see if I could refactor some code to work that way.

By way of example, I started with my MessageService class and extracted an interface from it deriving it from IMixedRealityExtensionService;

using Microsoft.MixedReality.Toolkit;
using System;

namespace MulticastMessaging
    public interface IMessageService : IMixedRealityExtensionService
        MessageRegistrar MessageRegistrar { get; set; }
        void Close();
        void Open();
        void Send<T>(T message, Action<bool> callback = null) where T : Message;

and then I defined a profile class for my service with the sorts of properties that I might want to set on it;

using Microsoft.MixedReality.Toolkit;
using UnityEngine;

namespace MulticastMessaging

        menuName = "Mixed Reality Toolkit/Message Service Profile",
        fileName = "MessageServiceProfile")]
    public class MessageServiceProfile : BaseMixedRealityProfile
        [Tooltip("The address to use for multicast messaging")]
        public string multicastAddress = "";

        [Tooltip("The port to use for multicast messaging")]
        public int multicastPort = 49152;

and then implemented that on my MessageService class deriving that from BaseExtensionService and marking it with a MixedRealityExtensionService attribute as you see below;

namespace MulticastMessaging
    using Microsoft.MixedReality.Toolkit;
    using Microsoft.MixedReality.Toolkit.Utilities;
    using System;
    using System.Diagnostics;
    using System.IO;
    using System.Net;
    using System.Net.Sockets;

    [MixedRealityExtensionService(SupportedPlatforms.WindowsUniversal | SupportedPlatforms.WindowsEditor)]
    public class MessageService : BaseExtensionService, IMessageService
        // Note: is the start of the UDP multicast addresses reserved for
        // private use.
        // Note: 49152 is the result I get out of executing;
        //      netsh int ipv4 show dynamicport udp
        // on Windows 10.
        public MessageService(
            IMixedRealityServiceRegistrar registrar,
            string name,
            uint priority,
            BaseMixedRealityProfile profile) : base(registrar, name, priority, profile)

        MessageServiceProfile Profile => base.ConfigurationProfile as MessageServiceProfile;

Clearly, that’s not the whole code but note the use of the MixedRealityExtensionService attribute and also the reach into the base class to get the ConfigurationProfile and cast it up as the concrete type of my actual profile.

With that in place, I can now use the editor to create one of those profiles;


and then I can add my new service to extensions of the toolkit;


and then change my code to grab hold of the instance via


whenever I want to get hold of the instance of that service.

In this branch, I only added two services this way – my networking provider and my messaging service but in my V2Work branch I’ve made more of these services and plan to rework a few more pieces in this way;


The commit at this point is here.

Wrapping Up

I wanted to go around the loop again on this exercise partly to make my own notes around things that I have perhaps forgotten and partly in case there were some pieces that others might pick up on and share.

I’m not planning to take this V2WorkBlogPost branch any further or release anything from it because I’ve already done the port in my V2Work branch and I want to move that forward and, ultimately, merge back into master from there but I did learn a few things by repeating the exercise, namely;

  1. I can do a better job of making speech work in the editor and at runtime.
  2. I should make more extension services for some of the other pieces of my app.
  3. I did a better job of leaving the MRTK V1 in the code base until I really no longer needed it whereas first time around I removed it too early and got in a bit of a mess Smile
  4. I realised that more of the app functionality needs to work in the editor and I can improve that but there’s still a way to go as I haven’t made attempts to have all of it work in the editor.

I hope that there was something useful in here for readers (if any get this far to the end of the post) and good luck in porting your own apps across to MRTK V2 Smile