Alois Kraus


  Home  |   Contact  |   Syndication    |   Login
  133 Posts | 8 Stories | 368 Comments | 162 Trackbacks



Post Categories


I was at the Jazoon 2011 in Zurich (Switzerland). It was a really cool event and it had many top notch speaker not only from the Microsoft universe. One of the most interesting talks was from Don Syme with the title: F# Today/F# Tomorrow. He did show how to use F# scripting to browse through open databases/, OData Web Services, Sharepoint, …interactively. It looked really easy with the help of F# Type Providers which is the next big language feature in a future F# version.

The object returned by a Type Provider is used to access the data like in usual strongly typed object model. No guessing how the property of an object is called. Intellisense will show it just as you expect. There exists a range of Type Providers for various data sources where the schema of the stored data can somehow be dynamically extracted. Lets use e.g. a free database it would be then

let data = DbProvider(http://.....);

data the object which contains all data from e.g. a chemical database. It has an elements collection which contains an element which has the properties: Name, AtomicMass, Picture, …. You can browse the object returned by the Type Provider with full Intellisense because the returned object is strongly typed which makes this happen.

The same can be achieved of course with code generators that use an input the schema of the input data (OData Web Service, database, Sharepoint, JSON serialized data, …) and spit out the necessary strongly typed objects as an assembly.

This does work but has the downside that if the schema of your data source is huge you will quickly run against a wall with traditional code generators since the generated “deserialization” assembly could easily become several hundred MB.

*** The following part contains guessing how this exactly work by asking Don two questions ****

Q: Can I use Type Providers within C#?

D: No.

Q: F# is after all a library. I can reference the F# assemblies and use the contained Type Providers?

D: F# does annotate the generated types in a special way at runtime which is not a static type that C# could use.

The F# type providers seem to use a hybrid approach. At compilation time the Type Provider is instantiated with the url of your input data. The obtained schema information is used by the compiler to generate static types as usual but only for a small subset (the top level classes up to certain nesting level would make sense to me). To make this work you need to access the actual data source at compile time which could be a problem if you want to keep the actual url in a config file.

Ok so this explains why it does work at all. But in the demo we did see full intellisense support down to the deepest object level. It looks like if you navigate deeper into the object hierarchy the type provider is instantiated in the background and attach to a true static type the properties determined at run time while you were typing. So this type is not really static at all. It is static if you define as a static type that its properties shows up in intellisense. But since this type information is determined while you are typing and it is not used to generate a true static type and you cannot use these “intellistatic” types from C#.

Nonetheless this is a very cool language feature. With the plotting libraries you can generate expressive charts from any datasource within seconds to get quickly an overview of any structured data storage.

My favorite programming language C# will not get such features in the near future there is hope. If you restrict yourself to OData sources you can use LINQPad to query any OData enabled data source with LINQ with ease.

There you can query Stackoverflow with


The output is also nicely rendered which makes it a very good tool to explore OData sources today.

posted on Thursday, June 23, 2011 9:29 AM