Hiperspace.SQL 2.5.21

• net10.0

  • FSharp.Core (>= 10.0.100)
  • FsLexYacc (>= 11.3.0)
  • FsLexYacc.Runtime (>= 11.3.0)
  • HiLang (>= 2.5.18)
  • Hiperspace (>= 2.5.21)
  • Microsoft.Bcl.HashCode (>= 6.0.0)
  • Microsoft.CodeAnalysis (>= 5.0.0)
  • Microsoft.CodeAnalysis.Analyzers (>= 4.14.0)
  • Microsoft.CodeAnalysis.CSharp (>= 5.0.0)
  • Parquet.Net (>= 5.4.0)
  • protobuf-net.Core (>= 3.2.56)
  • System.CodeDom (>= 10.0.0)

• net8.0

  • FSharp.Core (>= 10.0.100)
  • FsLexYacc (>= 11.3.0)
  • FsLexYacc.Runtime (>= 11.3.0)
  • HiLang (>= 2.5.18)
  • Hiperspace (>= 2.5.21)
  • Microsoft.Bcl.HashCode (>= 6.0.0)
  • Microsoft.CodeAnalysis (>= 5.0.0)
  • Microsoft.CodeAnalysis.Analyzers (>= 4.14.0)
  • Microsoft.CodeAnalysis.CSharp (>= 5.0.0)
  • Parquet.Net (>= 5.4.0)
  • protobuf-net.Core (>= 3.2.56)
  • System.CodeDom (>= 10.0.0)
  • System.Text.Json (>= 10.0.0)

• net9.0

  • FSharp.Core (>= 10.0.100)
  • FsLexYacc (>= 11.3.0)
  • FsLexYacc.Runtime (>= 11.3.0)
  • HiLang (>= 2.5.18)
  • Hiperspace (>= 2.5.21)
  • Microsoft.Bcl.HashCode (>= 6.0.0)
  • Microsoft.CodeAnalysis (>= 5.0.0)
  • Microsoft.CodeAnalysis.Analyzers (>= 4.14.0)
  • Microsoft.CodeAnalysis.CSharp (>= 5.0.0)
  • Parquet.Net (>= 5.4.0)
  • protobuf-net.Core (>= 3.2.56)
  • System.CodeDom (>= 10.0.0)
  • System.Text.Json (>= 10.0.0)

https://www.cepheis.com/hiperspace/20251210
# Overview
This release introduces GPU optimization for graph searches involving `HiperEdge` and other vector operations used in nearest-neighbor searches (*commonly required when providing information for AI prompts*).

It also enhances *Hiperspace* functions for LINQ queries by adding automatic null-coalescing for values selected from Hiperspace, supporting non-sargable join conditions, and improving performance for multi-table joins and **HiperspaceDB** cube aggregation.

-----

## CalculationGPU
[Graph Theory](https://en.wikipedia.org/wiki/Graph_theory) is a branch of Mathematics that can be used to describe all knowledge as `Node` and `Edge`'s  between them; because there is a simple consistent model across domains it has wide application for visualizing information, but has the downside that many graph views appear as a tangled web of nodes and lines: *often* conveying complexity, *but not* clarity.

[HyperGraph](https://en.wikipedia.org/wiki/Hypergraph)  is a generalization of [Graph](https://en.wikipedia.org/wiki/Graph_theory) that replaces the tangle of edges with a *hyperedge* that encapsulates all the intermediate nodes that are not interesting: I have a *cousin* (Robert), which can be described as "`Me->(cousin)->Robert`", but *cousin* is actually a *hyperedge* that is derived from "`me->mother2->mother1->daughter->son where mother2 is not daughter`"

[HiperEdge](https://www.cepheis.com/hiperspace/hiperedge) is a *HiperSpace* implementation of *hyperedge* that encapsulates the source of the *hyperedge* for a simple view of connections ***or*** the intermediate nodes in the path. When viewed graphically a `HiperEdge` view looks like a tree of connections extending out from the subject `Node`.

Hiperspace provides functions to query  [HiperEdge](https://www.cepheis.com/hiperspace/hiperedge)s using rules that project {*source node type, edge type, end node type*} to create [HiperEdge](https://www.cepheis.com/hiperspace/hiperedge)s whenever needed (*e.g. as a member function of a `Node`*).

### *Graph support*
This version introduces `ICalculationGPU` for optional acceleration of Graph Queries using the hardware created for [Ray Tracing](https://en.wikipedia.org/wiki/Ray_tracing_(graphics)),

### *Calculation support*

Search for *Nearest* node proximity in a [Vector Space](https://en.wikipedia.org/wiki/Vector_space) is supported via GPU, together with aggregation of Vectors used for complex calculation.

-----

### Null Coalesce
Consider a cube analytics model with {*Sector, Customer, Account, Trade*} with dimensions *Customer* and *Sector*; with *Trade* Facts aggregated. Null-coalescing is necessary when querying dimensions associated with a *Trade* fact.
```
@CubeDimension, CubeHierarchy(Parent), Versioned
entity Sector
 (Id : Int32)
 {Name : String, Parent : Sector}
 [Children : Sector (Parent = this), Customers : Customer (Sector = this)];

@CubeDimension, DeltaIndex
entity Customer
 (Id : String)
 {Sector : Sector}
 [Name : String, Accounts : Account (Customer = this)];
entity Instrument;

entity Account
 (Id : String)
 {Name : String, Customer : Customer};

@CubeFact, DeltaIndex
entity Trade : Versioned
 (id : Int64)
 {@CubeMeasure(Aggregation.Sum) Quantity : Decimal, Instrument : Instrument, Account : Account};
```
[image]Sites/hiperspace/sample/trade-null.png[/image]

The query  `from trade in Space.Trades select trade.Account.Customer.Sector` provides the *Sector* for a trade, but if any {*Trade.Account, Account.Customer, Customer.Sector*} is missing the expression will break.  The work around is to manually add null-coalesce to the query:
```
from trade in Space.Trades
select trade.Account ==  null ? null :  
      trade.Account.Customer == null ? null :
      trade.Account.Customer.Sector;
```
**This release adds coalesce automatically**. This is especially useful when Hiperspace.SQL is used to query a *hiperspace* from a scripting language where null-coalesce is not available, but we want to access the *graph* of properties.
```
select c.Id as SectorId, a.Id as AccountId
from Account as a join
    Customer as c on a.Customer.Id = c.Id
group by c.Sector.Id, a.Id;
```
This query would have ***previously*** failed if `a.Customer` is missing.

-----

### Non-Saragable predicates
In *Hiperspace* the following query cannot be [Sargable](https://en.wikipedia.org/wiki/Sargable) since *Customer* does not contain a *Sector* but a `KeyRef<Sector.KeyType, Sector>` with *lazy* loading of the element if needed.
```
select Sector.Name as "Sector Name", c.Name as "Customer Name"
from   Sector join
      Customer on Sector.Name = Customer.Sector.Name
order by Sector.Name, Customer.Name;
```
This query would ***previously*** have failed since `Customer.Sector` is a derived property of *Customer* and cannot be applied to a template for a `Find()` request of matching *Customers*

This release splits join processing into [Sargable](https://en.wikipedia.org/wiki/Sargable) predicates and residual conditions and is constructed dynamically as compiled expressions.