Enhancing the F# developer experience with MongoDB



This is a guest post by Max Hirschhorn, who is currently an intern at MongoDB.

About the F# programming language

F# is a multi-paradigm language built on the .NET framework. It is functional-first and prefers immutability, but also supports object-oriented and imperative programming styles.

Also, F# is a statically-typed language with a type inference system. It has a syntax similar to Ocaml, and draws upon ideas from other functional programming languages such as Erlang and Haskell.

Using the existing .NET driver

The existing .NET driver is compatible with F#, but is not necessarily written in a way that is idiomatic to use from F#.

Part of the reason behind this is that everything in F# is explicit. For example, consider the following example interface and implementing class.

type I =
    abstract Foo : unit -> string

type C() =
    interface I with
        member __.Foo () = "bar"

// example usage
let c = C()
(c :> I).Foo()

So in order to use any of the interface members, the class must be upcasted using the :> operator. Note that this cast is still checked at compile-time.

In a similar vein, C# supports implicit operators, which the BSON library uses for converting between a primitive value and its BsonValue equivalent, e.g.

new BsonDocument {
    { "price", 1.99 },
    { "$or", new BsonDocument {
        { "qty", new BsonDocument { { "$lt", 20 } } },
        { "sale", true }
    } }

whereas F# does not. This requires the developer to explicitly construct the appropriate type of BsonValue, e.g.

BsonDocument([ BsonElement("price", BsonDouble(1.99))
               BsonElement("$or", BsonArray([ BsonDocument("qty", BsonDocument("$lt", BsonInt32(20)))
                                              BsonDocument("sale", BsonBoolean(true)) ])) ])

with the query builder, we can hide the construction of BsonDocument instances, e.g.

Query.And([ Query.EQ("price", BsonDouble(1.99))
            Query.OR([ Query.LT("qty", BsonInt32(20))
                       Query.EQ("sale", BsonBoolean(true)) ]) ])

It is worth noting that the need to construct the BsonValue instances is completely avoided when using a typed QueryBuilder.

type Item = {
    Price : float
    Quantity : int
    Sale : bool

let query = QueryBuilder<Item>()

query.And([ query.EQ((fun item -> item.Price), 1.99)
            query.Or([ query.LT((fun item -> item.Quantity), 20)
                       query.EQ((fun item -> item.Sale), true) ]) ])

What we are looking for is a solution that matches the brevity of F# code, offers type-safety if desired, and is easy to use from the language.

New features

The main focus of this project is to make writing queries against MongoDB as natural from the F# language as possible.

bson quotations

We strive to make writing predicates as natural as possible by reusing as many of the existing operators as possible.

A taste

Consider the following query

{ price: 1.99, $or: [ { qty: { $lt: 20 } }, { sale: true } ] }

we could express this with a code quotation

bson <@ fun (x : BsonDocument) -> x?price = 1.99 && (x?qty < 20 || x?sale = true) @>

or with type safety

bson <@ fun (x : Item) -> x.Price = 1.99 && (x.Quantity < 20 || x.Sale = true) @>
Breaking it down

The quotations are not actually executed, but instead are presented as an abstract syntax tree (AST), from which an equivalent BsonDocument instance is constructed.

The ? operator

The ? operator is defined to allow for an unchecked comparison. The F# language supports the ability to do a dynamic lookup (get) and assignment (set) via the ? and ?<- operators respectively, but does not actually provide a implementation.

So, the F# driver defines the ? operator as the value associated with a field in a document casted to a fresh generic type.

// type signature: BsonDocument -> string -> 'a
let (?) (doc : BsonDocument) (field : string) =
    unbox doc.[field]

and similarly defines the ?<- operator as the coerced assignment of a generically typed value to the associated field in the document.

// type signature: BsonDocument -> string -> 'a -> unit
let (?<-) (doc : BsonDocument) (field : string) value =
    doc.[field] = unbox value |> ignore

Unchecked expressions have the type signature Expr<BsonDocument -> bool>.

// $mod
bson <@ fun (x : BsonDocument) -> x?qty % 4 = 0 @>

Checked expressions have the type signature Expr<'DocType -> bool>.

// $mod
bson <@ fun (x : Item) -> x.Quantity % 4 = 0 @>

Unchecked expressions have the type signature Expr<BsonDocument -> unit list>. The reason for the list in the return type is to perform multiple update operations.

// $set
bson <@ fun (x : BsonDocument) -> [ x?qty <- 20 ] @>

// $inc
bson <@ fun (x : BsonDocument) -> [ x?qty <- (+) 1 ] @>
Mmm… sugar

A keen observer would notice that (+) 1 is not an int, but actually a function int -> int. We are abusing the fact that type safety is not enforced here by assigning the quantity field of the document to a lambda expression, that takes a single parameter of the current value.

Note that

// $inc
bson <@ fun (x : BsonDocument) -> [ x?qty <- x?qty + 1 ] @>

is also valid.

Checked expressions either have the type signature Expr<'DocType -> unit list> or Expr<'DocType -> 'DocType>, depending on whether the document type has mutable fields (only matters for record types).

// $set
bson <@ fun (x : Item) -> [ x.Quantity <- 20 ] @>

// $inc
bson <@ fun (x : Item) -> [ x.Quantity <- x.Quantity + 1 ] @>

mongo expressions

Uses the monadic structure (computation expression) to define a pipeline of operations that are executed on each document in the collection.

let collection : IMongoCollection<BsonDocument> = ...

mongo {
    for x in collection do
    where (x?price = 1.99 && (x?qty < 20 || x?sale = true))

or with a typed collection

let collection : IMongoCollection<Item> = ...

mongo {
    for x in collection do
    where (x.price = 1.99 && (x.qty < 20 || x.sale = true))
let collection : IMongoCollection<BsonDocument> = ...

mongo {
    for x in collection do
    set x?price 0.99
    inc x?qty 1

or with a typed collection

let collection : IMongoCollection<Item> = ...

mongo {
    for x in collection do
    set x.Price 0.99
    inc x.Quantity 1

Serialization of F# data types

Now supports



The source code is available at GitHub. We absolutely encourage you to experiment with it and provide us feedback on the API, design, and implementation. Bug reports and suggestions for improvements are welcomed, as are pull requests.

Disclaimer. The API and implementation are currently subject to change at any time. You must not use this driver in production, as it is still under development and is in no way supported by MongoDB, Inc.


Many thanks to the guidance from the F# community on Twitter, and my mentors: Sridhar Nanjundeswaran, Craig Wilson, and Robert Stam. Also, a special thanks to Stacy Ferranti and Ian Whalen for overseeing the internship program.