F# Compiler Services


Interactive Service: Embedding F# Interactive

This tutorial demonstrates how to embed F# interactive in your application. F# interactive is an interactive scripting environment that compiles F# code into highly efficient IL code and executes it on the fly. The F# interactive service allows you to embed F# evaluation in your application.

NOTE: There is a number of options for embedding F# Interactive. The easiest one is to use the fsi.exe process and communicate with it using standard input and standard output. In this tutorial, we look at calling F# Interactive directly through .NET API. However, if you have no control over the input, it is a good idea to run F# interactive in a separate process. One reason is that there is no way to handle StackOverflowException and so a poorly written script can terminate the host process. Remember that while calling F# Interactive through .NET API, --shadowcopyreferences option will be ignored. For detailed discussion, please take a look at this thread. NOTE: If FsiEvaluationSession.Create fails with an error saying that FSharp.Core.dll cannot be found, add the FSharp.Core.sigdata and FSharp.Core.optdata files. More info here.

However, the F# interactive service is still useful, because you might want to wrap it in your own executable that is then executed (and communicates with the rest of your application), or if you only need to execute limited subset of F# code (e.g. generated by your own DSL).

Starting the F# interactive

First, we need to reference the libraries that contain F# interactive service:

1: 
2: 
3: 
#r "FSharp.Compiler.Service.dll"
open Microsoft.FSharp.Compiler.SourceCodeServices
open Microsoft.FSharp.Compiler.Interactive.Shell

To communicate with F# interactive, we need to create streams that represent input and output. We will use those later to read the output printed as a result of evaluating some F# code that prints:

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
11: 
12: 
13: 
14: 
15: 
16: 
17: 
open System
open System.IO
open System.Text

// Intialize output and input streams
let sbOut = new StringBuilder()
let sbErr = new StringBuilder()
let inStream = new StringReader("")
let outStream = new StringWriter(sbOut)
let errStream = new StringWriter(sbErr)

// Build command line arguments & start FSI session
let argv = [| "C:\\fsi.exe" |]
let allArgs = Array.append argv [|"--noninteractive"|]

let fsiConfig = FsiEvaluationSession.GetDefaultConfiguration()
let fsiSession = FsiEvaluationSession.Create(fsiConfig, allArgs, inStream, outStream, errStream)  

Evaluating and executing code

The F# interactive service exposes several methods that can be used for evaluation. The first is EvalExpression which evaluates an expression and returns its result. The result contains the returned value (as obj) and the statically inferred type of the value:

1: 
2: 
3: 
4: 
5: 
/// Evaluate expression & return the result
let evalExpression text =
  match fsiSession.EvalExpression(text) with
  | Some value -> printfn "%A" value.ReflectionValue
  | None -> printfn "Got no result!"

This takes a string as an argument and evaluates (i.e. executes) it as F# code.

1: 
evalExpression "42+1" // prints '43'

This can be used in a strongly typed way as follows:

1: 
2: 
3: 
4: 
5: 
6: 
7: 
/// Evaluate expression & return the result, strongly typed
let evalExpressionTyped<'T> (text) = 
    match fsiSession.EvalExpression(text) with
    | Some value -> value.ReflectionValue |> unbox<'T>
    | None -> failwith "Got no result!"

evalExpressionTyped<int> "42+1"  // gives '43'

The EvalInteraction method can be used to evaluate side-effectful operations such as printing, declarations, or other interactions that are not valid F# expressions, but can be entered in the F# Interactive console. Such commands include #time "on" (and other directives), open System all declarations and other top-level statements. The code does not require ;; at the end. Just enter the code that you want to execute:

1: 
fsiSession.EvalInteraction "printfn \"bye\""

The EvalScript method allows to evaluate a complete .fsx script.

1: 
2: 
File.WriteAllText("sample.fsx", "let twenty = 10 + 10")
fsiSession.EvalScript "sample.fsx"

Catching errors

EvalExpression, EvalInteraction and EvalScript are awkward if the code has type checking warnings or errors, or if evaluation fails with an exception. In these cases you can use EvalExpressionNonThrowing, EvalInteractionNonThrowing and EvalScriptNonThrowing. These return a tuple of a result and an array of FSharpErrorInfo values. These represent the errors and warnings. The result part is a Choice<_,_> between an actual result and an exception.

The result part of EvalExpression and EvalExpressionNonThrowing is an optional FSharpValue. If that value is not present then it just indicates that the expression didn't have a tangible result that could be represented as a .NET object. This siutation shouldn't actually occur for any normal input expressions, and only for primitives used in libraries.

1: 
2: 
3: 
4: 
5: 
6: 
7: 
File.WriteAllText("sample.fsx", "let twenty = 'a' + 10.0")
let result, warnings = fsiSession.EvalScriptNonThrowing "sample.fsx"

// show the result
match result with 
| Choice1Of2 () -> printfn "checked and executed ok"
| Choice2Of2 exn -> printfn "execution exception: %s" exn.Message

Gives:

1: 
execution exception: Operation could not be completed due to earlier error
1: 
2: 
3: 
// show the errors and warnings
for w in warnings do 
   printfn "Warning %s at %d,%d" w.Message w.StartLineAlternate w.StartColumn 

Gives:

1: 
2: 
Warning The type 'float' does not match the type 'char' at 1,19
Warning The type 'float' does not match the type 'char' at 1,17

For expressions:

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
let evalExpressionTyped2<'T> text =
   let res, warnings = fsiSession.EvalExpressionNonThrowing(text)
   for w in warnings do 
       printfn "Warning %s at %d,%d" w.Message w.StartLineAlternate w.StartColumn 
   match res with 
   | Choice1Of2 (Some value) -> value.ReflectionValue |> unbox<'T>
   | Choice1Of2 None -> failwith "null or no result"
   | Choice2Of2 (exn:exn) -> failwith (sprintf "exception %s" exn.Message)

evalExpressionTyped2<int> "42+1"  // gives '43'

Executing in parallel

By default the code passed to EvalExpression is executed immediately. To execute in parallel, submit a computation that starts a task:

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
11: 
12: 
13: 
open System.Threading.Tasks

let sampleLongRunningExpr = 
    """
async { 
    // The code of what you want to run
    do System.Threading.Thread.Sleep 5000
    return 10 
}
  |> Async.StartAsTask"""

let task1 = evalExpressionTyped<Task<int>>(sampleLongRunningExpr)  
let task2 = evalExpressionTyped<Task<int>>(sampleLongRunningExpr)  

Both computations have now started. You can now fetch the results:

1: 
2: 
task1.Result // gives the result after completion (up to 5 seconds)
task2.Result // gives the result after completion (up to 5 seconds)

Type checking in the evaluation context

Let's assume you have a situation where you would like to typecheck code in the context of the F# Interactive scripting session. For example, you first evaluation a declaration:

1: 
fsiSession.EvalInteraction "let xxx = 1 + 1"

Now you want to typecheck the partially complete code xxx + xx

1: 
let parseResults, checkResults, checkProjectResults = fsiSession.ParseAndCheckInteraction("xxx + xx")

The parseResults and checkResults have types ParseFileResults and CheckFileResults explained in Editor. You can, for example, look at the type errors in the code:

1: 
checkResults.Errors.Length // 1

The code is checked with respect to the logical type context available in the F# interactive session based on the declarations executed so far.

You can also request declaration list information, tooltip text and symbol resolution:

1: 
2: 
3: 
4: 
5: 
6: 
7: 
open Microsoft.FSharp.Compiler

// get a tooltip
checkResults.GetToolTipTextAlternate(1, 2, "xxx + xx", ["xxx"], FSharpTokenTag.IDENT) 

checkResults.GetSymbolUseAtLocation(1, 2, "xxx + xx", ["xxx"]) // symbol xxx
  

The 'fsi' object

If you want your scripting code to be able to access the 'fsi' object, you should pass in an implementation of this object explicitly. Normally the one fromm FSharp.Compiler.Interactive.Settings.dll is used.

1: 
let fsiConfig2 = FsiEvaluationSession.GetDefaultConfiguration(fsi)

Collectible code generation

Evaluating code in using FsiEvaluationSession generates a .NET dynamic assembly and uses other resources. You can make generated code collectible by passing collectible=true. However code will only be collected if there are no outstanding object references involving types, for example FsiValue objects returned by EvalExpression, and you must have disposed the FsiEvaluationSession. See also Restrictions on Collectible Assemblies.

The example below shows the creation of 200 evaluation sessions. Note that collectible=true and use session = ... are both used.

If collectible code is working correctly, overall resource usage will not increase linearly as the evaluation progresses.

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
11: 
12: 
13: 
14: 
15: 
16: 
let collectionTest() = 

    for i in 1 .. 200 do
        let defaultArgs = [|"fsi.exe";"--noninteractive";"--nologo";"--gui-"|]
        use inStream = new StringReader("")
        use outStream = new StringWriter()
        use errStream = new StringWriter()

        let fsiConfig = FsiEvaluationSession.GetDefaultConfiguration()
        use session = FsiEvaluationSession.Create(fsiConfig, defaultArgs, inStream, outStream, errStream, collectible=true)
        
        session.EvalInteraction (sprintf "type D = { v : int }")
        let v = session.EvalExpression (sprintf "{ v = 42 * %d }" i)
        printfn "iteration %d, result = %A" i v.Value.ReflectionValue

// collectionTest()  <-- run the test like this
Fork me on GitHub