Speckle 2.0 stream & object structure

Ritesh_Solanke · 25 January 2021 13:16

I want to know that how the stream and object structure are different compared to speckle 1.0. We are developing data structure based on it. It will helpful if you provide the details.

@Ritesh_Solanke

dimitrie · 25 January 2021 16:20

Hi @Ritesh_Solanke! I’m simple question, long answer

In 2.0, Speckle supports pretty much any kind of structure for data. Some examples:

From Rhino, we group objects by layer.
From Revit, we currently group objects by their type. Nested elements, such as windows, are nested inside their parents.
From Grasshopper, we allow end users to create the structure dynamically.

On the receiving end, what we do, is we iterate through this structure and do whatever action we need to. For example, let’s assume the following data structure coming from Grasshopper, with a bunch of elements:


var commit = new Base(); 
commit["levelOne"] = new List<Base>() { some walls, some floor slabs, some furniture, etc. }
commit["levelTwo"] = new Dictionary<string, object>() { [ "levelTwo-LeftWing" = a list of objects], [ "levelTwo-CentralAtrium" = a different list of objects] }

When received in Revit, we initially flatten the structure and extract the objects we can convert to Revit; then we convert said objects via the kit conversion routines. In Rhino, we have a similar approach: we traverse the commit, and as we find convertable objects, we create them into the current Rhino document on nested layers that replicate the commit structure:

github.com

specklesystems/speckle-sharp/blob/0efa64f756262b19e03e27df5dc53dba6dd0f2fd/ConnectorRhino/ConnectorRhino/UI/ConnectorBindingsRhino.cs#L252-L372


private void HandleAndConvert(object obj, ISpeckleConverter converter, Layer layer, StreamState state, Action updateProgressAction = null)
{
  Layer myLayer = null;

  // The rhino layer api is a bit sucky, hence the result below. It probably can be cleaned up and optimised.
  if (!layer.HasIndex || layer.Index == -1)
  {
    // Try and recreate layer structure if coming from Rhino.
    if (layer.Name.Contains("::") || layer.FullPath.Contains("::"))
    {
      var layers = layer.Name.Split(new string[] { "::" }, StringSplitOptions.RemoveEmptyEntries);
      var ancestors = new List<Layer>();
      var currentPath = currentRootLayerName;
      foreach (var linkName in layers)
      {
        currentPath += $"::{linkName}";
        var existingIndex = Doc.Layers.FindByFullPath(currentPath, -1);
        if (existingIndex != -1)
        {
          ancestors.Add(Doc.Layers[existingIndex]);

This file has been truncated. show original

For the 3d viewer, we do exactly the same thing; we always traverse the data structure and we convert it to buffer geometries that we can later display in threejs:

github.com

specklesystems/speckle-server/blob/a793dc433a35de7c9834bb4d454f9f4c853cbd40/packages/viewer/src/modules/Converter.js#L25-L74


async traverseAndConvert( obj, callback ) {
  // Exit on primitives (string, ints, bools, bigints, etc.)
  if ( typeof obj !== 'object' ) return

  if ( obj.referencedId ) obj = await this.resolveReference( obj )

  // Traverse arrays, and exit early (we don't want to iterate through many numbers)
  if ( Array.isArray( obj ) ) {
    for ( let element of obj ) {
      if ( typeof element !== 'object' ) return // exit early for non-object based arrays
      ( async() => await this.traverseAndConvert( element, callback ) )() //iife so we don't block
    }
  }

  // If we can convert it, we should invoke the respective conversion routine.
  const type = this.getSpeckleType( obj )
  if ( this[`${type}ToBufferGeometry`] ) {
    try {
      callback( await this[`${type}ToBufferGeometry`]( obj.data || obj ) )
      return

This file has been truncated. show original

TL;DR

There is no given or fixed structure of a commit - see it like everybody organises their code project files in different ways. In order to operate on, or with, objects in Speckle you have to traverse the commit data structure and operate on it accordingly:

either act on each object individually as it comes along - i.e., as part of a map and reduce` operation - this can be very powerful and memory efficient
either recreate the entire data structure in memory, and operate on it as such.

Ritesh_Solanke · 27 January 2021 05:18

thank you @dimitrie for explanation. We are right now focusing on the Revit, where you are grouping the similar type of elements in current version, which is not their in previous version. I would like to know are making any structural changes in tessellated geometry structure in 2.0.

dimitrie · 27 January 2021 18:24

Not really, though it would be good to check the classes out. Meshes stay the same!

Ritesh_Solanke · 28 January 2021 11:20

If I am correct the mesh data structure won’t have any changes. But if you remember we have discussed that you can add single instance for repeating instance and reduce the mesh size. Have you added that in 2.0.

dimitrie · 28 January 2021 16:44

Unfortunately not yet - but this does make it easier for development purposes It’s a good optimisation, but quite complex to enable reliably. We’re saving off much more by using compression and caching, so we’ve put it on the backburner for now.

Ritesh_Solanke · 1 February 2021 07:32

you mean the JSON data output for mesh will be same ? will it affect due to change in database to Postgres from mangodb ?

dimitrie · 1 February 2021 07:43

Morning @Ritesh_Solanke! Yep, meshes stay the same. Here’s an example. You’ll notice the good old vertices and faces array. As a reminder, the face array is structured as

[ faceType, vertexIndex_0, vertexIndex_1, vertexIndex_2, (vertexIndex_3) ]

where faceType can be 0 for triangles and 1 for quads.

The main thing to keep an eye out is the chunking of large arrays that we’ve introduced recently. Here’s our internal research doc re why we needed this. The .NET and Python SDKs handle this for you on deserialisation.

Ritesh_Solanke · 23 February 2021 05:15

You have changed storage of streams and objects from mongodb to postgres for 2.0 right ? So I want to understand how you are keeping the data in tabulated format. If possible can you share the doc or code base where I can get better clarity.

dimitrie · 23 February 2021 08:38

Morning @Ritesh_Solanke! Sure, here’s the actual migrations that create the objects tables:

github.com

specklesystems/speckle-server/blob/main/packages/server/modules/core/migrations/000-core.js#L115-L135


await knex.schema.createTable( 'objects', table => {
  table.string( 'id' ).primary( )
  table.string( 'speckleType', 1024 ).defaultTo( 'Base' ).notNullable( )
  table.integer( 'totalChildrenCount' )
  table.jsonb( 'totalChildrenCountByDepth' )
  table.timestamp( 'createdAt' ).defaultTo( knex.fn.now( ) )
  table.jsonb( 'data' )
} )

// Closure table for tracking the nesting relationships of objects.
// Note: the usecase optimised for is that when we request an object, we either:
// a) interactively request/query for its subchildren (sequentially)
// or
// b) we want all of it!
await knex.schema.createTable( 'object_children_closure', table => {
  table.string( 'parent' ).notNullable( ).index( )
  table.string( 'child' ).notNullable( ).index( )
  table.integer( 'minDepth' ).defaultTo( 1 ).notNullable( ).index( )
  table.unique( [ 'parent', 'child' ], 'obj_parent_child_index' )
  table.index( [ 'parent', 'minDepth' ], 'full_pcd_index' )
} )

The decomposition logic explains more in depth how things are structured, so it can be considered “required” reading to understand the approach in detail.

Ritesh_Solanke · 23 February 2021 09:14

hey @dimitrie If I understand correctly there is no stream in 2.0, then how you are going to handle the versioning is it at object level ?

dimitrie · 23 February 2021 09:18

Wait up, streams exist! They’re much more like git repos now though. @izzylys did a nice write up here re the basic concepts (streams, commits, branches, etc)

Ritesh_Solanke · 2 March 2021 07:24

hey @dimitrie ,

I am using the speckle 1.0 where I have noticed error like “if file size is exceed 2MB the elements are skipped” we have noticed that some of elements are missing. Is this issue is fixed in 2.0 !
I have attached error message image for your reference.
MicrosoftTeams-image (2)|410x500

dimitrie · 2 March 2021 09:36

Yes it is! Though i am quite skeptical of those 2mb - can you let us know what those elements are? Usually it’s an out-of-control mesh for a simple extrusion…