XML Type Provider

This article demonstrates how to use the XML Type Provider to access XML documents in a statically typed way. We first look at how the structure is inferred and then demonstrate the provider by parsing a RSS feed.

The XML Type Provider provides statically typed access to XML documents. It takes a sample document as an input (or document containing a root XML node with multiple child nodes that are used as samples). The generated type can then be used to read files with the same structure. If the loaded file does not match the structure of the sample, a runtime error may occur (but only when accessing e.g. non-existing element). Starting from version 3.0.0 there is also the option of using a schema (XSD) instead of relying on samples.

Introducing the provider

The type provider is located in the FSharp.Data.dll assembly. Assuming the assembly is located in the ../../bin directory, we can load it in F# Interactive as follows: (note we also need a reference to System.Xml.Linq, because the provider uses the XDocument type internally):

#r "System.Xml.Linq.dll"
open FSharp.Data

Inferring type from sample

The XmlProvider<...> takes one static parameter of type string. The parameter can be either a sample XML string or a sample file (relative to the current folder or online accessible via http or https). It is not likely that this could lead to ambiguities.

The following sample generates a type that can read simple XML documents with a root node containing two attributes:

type Author = XmlProvider<"""<author name="Paul Feyerabend" born="1924" />""">
let sample = Author.Parse("""<author name="Karl Popper" born="1902" />""")

printfn "%s (%d)" sample.Name sample.Born

The type provider generates a type Author that has properties corresponding to the attributes of the root element of the XML document. The types of the properties are inferred based on the values in the sample document. In this case, the Name property has a type string and Born is int.

XML is a quite flexible format, so we could represent the same document differently. Instead of using attributes, we could use nested nodes (<name> and <born> nested under <author>) that directly contain the values:

type AuthorAlt = XmlProvider<"<author><name>Karl Popper</name><born>1902</born></author>">
let doc = "<author><name>Paul Feyerabend</name><born>1924</born></author>"
let sampleAlt = AuthorAlt.Parse(doc)

printfn "%s (%d)" sampleAlt.Name sampleAlt.Born

The generated type provides exactly the same API for reading documents following this convention (Note that you cannot use AuthorAlt to parse samples that use the first style - the implementation of the types differs, they just provide the same public API.)

The provider turns a node into a simply typed property only when the node contains just a primitive value and has no children or attributes.

Types for more complex structure

Now let's look at a number of examples that have more interesting structure. First of all, what if a node contains some value, but also has some attributes?

type Detailed = XmlProvider<"""<author><name full="true">Karl Popper</name></author>""">
let info = Detailed.Parse("""<author><name full="false">Thomas Kuhn</name></author>""")

printfn "%s (full=%b)" info.Name.Value info.Name.Full

If the node cannot be represented as a simple type (like string) then the provider builds a new type with multiple properties. Here, it generates a property Full (based on the name of the attribute) and infers its type to be boolean. Then it adds a property with a (special) name Value that returns the content of the element.

Types for multiple simple elements

Another interesting case is when there are multiple nodes that contain just a primitive value. The following example shows what happens when the root node contains multiple <value> nodes (note that if we leave out the parameter to the Parse method, the same text used for the schema will be used as the runtime value).

type Test = XmlProvider<"<root><value>1</value><value>3</value></root>">

Test.GetSample().Values
|> Seq.iter (printfn "%d")

The type provider generates a property Values that returns an array with the values - as the <value> nodes do not contain any attributes or children, they are turned into int values and so the Values property returns just int[]!

Processing philosophers

In this section we look at an example that demonstrates how the type provider works on a simple document that lists authors that write about a specific topic. The sample document data/Writers.xml looks as follows:

<authors topic="Philosophy of Science">
  <author name="Paul Feyerabend" born="1924" />
  <author name="Thomas Kuhn" />
</authors>

At runtime, we use the generated type provider to parse the following string (which has the same structure as the sample document with the exception that one of the author nodes also contains a died attribute):

let authors = """
  <authors topic="Philosophy of Mathematics">
    <author name="Bertrand Russell" />
    <author name="Ludwig Wittgenstein" born="1889" />
    <author name="Alfred North Whitehead" died="1947" />
  </authors> """

When initializing the XmlProvider, we can pass it a file name or a web URL. The Load and AsyncLoad methods allows reading the data from a file or from a web resource. The Parse method takes the data as a string, so we can now print the information as follows:

type Authors = XmlProvider<"../data/Writers.xml", ResolutionFolder=__SOURCE_DIRECTORY__>
let topic = Authors.Parse(authors)

printfn "%s" topic.Topic
for author in topic.Authors do
  printf " - %s" author.Name
  author.Born |> Option.iter (printf " (%d)")
  printfn ""

The value topic has a property Topic (of type string) which returns the value of the attribute with the same name. It also has a property Authors that returns an array with all the authors. The Born property is missing for some authors, so it becomes option<int> and we need to print it using Option.iter.

The died attribute was not present in the sample used for the inference, so we cannot obtain it in a statically typed way (although it can still be obtained dynamically using author.XElement.Attribute(XName.Get("died"))).

Global inference mode

In the examples shown earlier, an element was never (recursively) contained in an element of the same name (for example <author> never contained another <author>). However, when we work with documents such as XHTML files, this can often be the case. Consider for example, the following sample (a simplified version of data/HtmlBody.xml):

<div id="root">
  <span>Main text</span>
  <div id="first">
    <div>Second text</div>
  </div>
</div>

Here, a <div> element can contain other <div> elements and it is quite clear that they should all have the same type - we want to be able to write a recursive function that processes <div> elements. To make this possible, you need to set an optional parameter Global to true:

type Html = XmlProvider<"../data/HtmlBody.xml", Global=true, ResolutionFolder=__SOURCE_DIRECTORY__>
let html = Html.GetSample()

When the Global parameter is true, the type provider unifies all elements of the same name. This means that all <div> elements have the same type (with a union of all attributes and all possible children nodes that appear in the sample document).

The type is located under a type Html, so we can write a printDiv function that takes Html.Div and acts as follows:

/// Prints the content of a <div> element
let rec printDiv (div:Html.Div) =
  div.Spans |> Seq.iter (printfn "%s")
  div.Divs |> Seq.iter printDiv
  if div.Spans.Length = 0 && div.Divs.Length = 0 then
      div.Value |> Option.iter (printfn "%s")

// Print the root <div> element with all children
printDiv html

The function first prints all text included as <span> (the element never has any attributes in our sample, so it is inferred as string), then it recursively prints the content of all <div> elements. If the element does not contain nested elements, then we print the Value (inner text).

Loading Directly from a File or URL

In many cases we might want to define schema using a local sample file, but then directly load the data from disk or from a URL either synchronously (with Load) or asynchronously (with AsyncLoad).

For this example I am using the US Census data set from https://api.census.gov/data.xml, a sample of which I have used here for ../data/Census.xml. This sample is greatly reduced from the live data, so that it contains only the elements and attributes relevant to us:

<census-api
    xmlns="http://thedataweb.rm.census.gov/api/discovery/"
    xmlns:dcat="http://www.w3.org/ns/dcat#"
    xmlns:dct="http://purl.org/dc/terms/">
    <dct:dataset>
        <dct:title>2006-2010 American Community Survey 5-Year Estimates</dct:title>
        <dcat:distribution
            dcat:accessURL="https://api.census.gov/data/2010/acs5">
        </dcat:distribution>
    </dct:dataset>
    <dct:dataset>
        <dct:title>2006-2010 American Community Survey 5-Year Estimates</dct:title>
        <dcat:distribution
            dcat:accessURL="https://api.census.gov/data/2010/acs5">
        </dcat:distribution>
    </dct:dataset>
</census-api>

When doing this for your scenario, be careful to ensure that enough data is given for the provider to infer the schema correctly. For example, the first level <dct:dataset> element must be included at least twice for the provider to infer the Datasets array rather than a single Dataset object.

type Census = XmlProvider<"../data/Census.xml", ResolutionFolder=__SOURCE_DIRECTORY__>

let data = Census.Load("https://api.census.gov/data.xml")

let apiLinks = data.Datasets
               |> Array.map (fun ds -> ds.Title,ds.Distribution.AccessUrl)

This US Census data is an interesting dataset with this top level API returning hundreds of other datasets each with their own API. Here we use the Census data to get a list of titles and URLs for the lower level APIs.

Bringing in Some Async Action

Let's go one step further and assume here a sligthly contrived but certainly plausible example where we cache the Census URLs and refresh once in a while. Perhaps we want to load this in the background and then post each link over (for example) a message queue.

This is where AsyncLoad comes into play:

let enqueue (title,apiUrl) =
  // do the real message enqueueing here instead of
  printfn "%s -> %s" title apiUrl

// helper task which gets scheduled on some background thread somewhere...
let cacheJanitor() = async {
  let! reloadData = Census.AsyncLoad("https://api.census.gov/data.xml")
  reloadData.Datasets |> Array.map (fun ds -> ds.Title,ds.Distribution.AccessUrl)
                      |> Array.iter enqueue
}

Reading RSS feeds

To conclude this introduction with a more interesting example, let's look how to parse a RSS feed. As discussed earlier, we can use relative paths or web addresses when calling the type provider:

type Rss = XmlProvider<"http://tomasp.net/rss.xml">

This code builds a type Rss that represents RSS feeds (with the features that are used on http://tomasp.net). The type Rss provides static methods Parse, Load and AsyncLoad to construct it - here, we just want to reuse the same URI of the schema, so we use the GetSample static method:

let blog = Rss.GetSample()

Printing the title of the RSS feed together with a list of recent posts is now quite easy - you can simply type blog followed by . and see what the autocompletion offers. The code looks like this:

// Title is a property returning string
printfn "%s" blog.Channel.Title

// Get all item nodes and print title with link
for item in blog.Channel.Items do
  printfn " - %s (%s)" item.Title item.Link

Transforming XML

In this example we will now also create XML in addition to consuming it. Consider the problem of flattening a data set. Let's say you have xml data that looks like this:

[<Literal>]
let customersXmlSample = """
  <Customers>
    <Customer name="ACME">
      <Order Number="A012345">
        <OrderLine Item="widget" Quantity="1"/>
      </Order>
      <Order Number="A012346">
        <OrderLine Item="trinket" Quantity="2"/>
      </Order>
    </Customer>
    <Customer name="Southwind">
      <Order Number="A012347">
        <OrderLine Item="skyhook" Quantity="3"/>
        <OrderLine Item="gizmo" Quantity="4"/>
      </Order>
    </Customer>
  </Customers>"""

and you want to transform it into something like this:

[<Literal>]
let orderLinesXmlSample = """
  <OrderLines>
    <OrderLine Customer="ACME" Order="A012345" Item="widget" Quantity="1"/>
    <OrderLine Customer="ACME" Order="A012346" Item="trinket" Quantity="2"/>
    <OrderLine Customer="Southwind" Order="A012347" Item="skyhook" Quantity="3"/>
    <OrderLine Customer="Southwind" Order="A012347" Item="gizmo" Quantity="4"/>
  </OrderLines>"""

We'll create types from both the input and output samples and use the constructors on the types generated by the XmlProvider:

type InputXml = XmlProvider<customersXmlSample>
type OutputXml = XmlProvider<orderLinesXmlSample>

let orderLines =
  OutputXml.OrderLines [|
    for customer in InputXml.GetSample().Customers do
      for order in customer.Orders do
        for line in order.OrderLines do
          yield OutputXml.OrderLine
                  ( customer.Name,
                    order.Number,
                    line.Item,
                    line.Quantity ) |]

Using a schema (XSD)

The Schema parameter can be used (instead of Sample) to specify an XML schema. The value of the parameter can be either the name of a schema file or plain text like in the following example:

type Person = XmlProvider<Schema = """
  <xs:schema xmlns:xs="http://www.w3.org/2001/XMLSchema"
    elementFormDefault="qualified" attributeFormDefault="unqualified">
    <xs:element name="person">
      <xs:complexType>
        <xs:sequence>
          <xs:element name="surname" type="xs:string"/>
          <xs:element name="birthDate" type="xs:date"/>
        </xs:sequence>
      </xs:complexType>
    </xs:element>
  </xs:schema>""">

let turing = Person.Parse """
  <person>
    <surname>Turing</surname>
    <birthDate>1912-06-23</birthDate>
  </person>
  """

printfn "%s was born in %d" turing.Surname turing.BirthDate.Year

The properties of the provided type are derived from the schema instead of being inferred from samples.

Usually a schema is not specified as plain text but stored in a file like data/po.xsd and the uri is set in the Schema parameter:

type PurchaseOrder = XmlProvider<Schema="../data/po.xsd">

When the file includes other schema files, the ResolutionFolder parameter can help locating them. The uri may also refer to online resources:

type RssXsd = XmlProvider<Schema = "https://www.w3schools.com/xml/note.xsd">

The schema is expected to define a root element (a global element with complex type). In case of multiple root elements:

type TwoRoots = XmlProvider<Schema = """
  <xs:schema xmlns:xs="http://www.w3.org/2001/XMLSchema"
    elementFormDefault="qualified" attributeFormDefault="unqualified">
    <xs:element name="root1">
      <xs:complexType>
        <xs:attribute name="foo" type="xs:string" use="required" />
        <xs:attribute name="fow" type="xs:int" />
      </xs:complexType>
    </xs:element>
    <xs:element name="root2">
      <xs:complexType>
        <xs:attribute name="bar" type="xs:string" use="required" />
        <xs:attribute name="baz" type="xs:date" use="required" />
      </xs:complexType>
    </xs:element>
  </xs:schema>
""">

the provided type has an optional property for each alternative:

let e1 = TwoRoots.Parse "<root1 foo='aa' fow='2' />"
match e1.Root1, e1.Root2 with
| Some x, None ->
    printfn "Foo = %s and Fow = %A" x.Foo x.Fow
| _ -> failwith "Unexpected"

let e2 = TwoRoots.Parse "<root2 bar='aa' baz='2017-12-22' />"
match e2.Root1, e2.Root2 with
| None, Some x ->
    printfn "Bar = %s and Baz = %O" x.Bar x.Baz
| _ -> failwith "Unexpected"

Common XSD constructs: sequence and choice

A sequence is the most common way of structuring elements in a schema. The following xsd defines foo as a sequence made of an arbitrary number of bar elements followed by a single baz element.

type FooSequence = XmlProvider<Schema = """
    <xs:schema xmlns:xs="http://www.w3.org/2001/XMLSchema"
      elementFormDefault="qualified" attributeFormDefault="unqualified">
        <xs:element name="foo">
          <xs:complexType>
            <xs:sequence>
              <xs:element name="bar" type="xs:int" maxOccurs="unbounded" />
              <xs:element name="baz" type="xs:date" minOccurs="1" />
            </xs:sequence>
          </xs:complexType>
        </xs:element>
    </xs:schema>""">

here a valid xml element is parsed as an instance of the provided type, with two properties corresponding to barand baz elements, where the former is an array in order to hold multiple elements:

let fooSequence = FooSequence.Parse """
<foo>
    <bar>42</bar>
    <bar>43</bar>
    <baz>1957-08-13</baz>
</foo>"""

printfn "%d" fooSequence.Bars.[0] // 42
printfn "%d" fooSequence.Bars.[1] // 43
printfn "%d" fooSequence.Baz.Year // 1957

Instead of a sequence we may have a choice:

type FooChoice = XmlProvider<Schema = """
    <xs:schema xmlns:xs="http://www.w3.org/2001/XMLSchema"
      elementFormDefault="qualified" attributeFormDefault="unqualified">
        <xs:element name="foo">
          <xs:complexType>
            <xs:choice>
              <xs:element name="bar" type="xs:int" maxOccurs="unbounded" />
              <xs:element name="baz" type="xs:date" minOccurs="1" />
            </xs:choice>
          </xs:complexType>
        </xs:element>
    </xs:schema>""">

although a choice is akin to a union type in F#, the provided type still has properties for bar and baz directly available on the foo object; in fact the properties representing alternatives in a choice are simply made optional (notice that for arrays this is not even necessary because an array can be empty). This decision is due to technical limitations (discriminated unions are not supported in type providers) but also preferred because it improves discoverability: intellisense can show both alternatives. There is a lack of precision but this is not the main goal.

let fooChoice = FooChoice.Parse """
<foo>
  <baz>1957-08-13</baz>
</foo>"""

printfn "%d items" fooChoice.Bars.Length // 0 items
match fooChoice.Baz with
| Some date -> printfn "%d" date.Year // 1957
| None -> ()

Another xsd construct to model the content of an element is all, which is used less often and it's like a sequence where the order of elements does not matter. The corresponding provided type in fact is essentially the same as for a sequence.

Advanced schema constructs

XML Schema provides various extensibility mechanisms. The following example is a terse summary mixing substitution groups with abstract recursive definitions.

type Prop = XmlProvider<Schema = """
    <xs:schema xmlns:xs="http://www.w3.org/2001/XMLSchema"
      elementFormDefault="qualified" attributeFormDefault="unqualified">
        <xs:element name="Formula" abstract="true"/>
        <xs:element name="Prop" type="xs:string" substitutionGroup="Formula"/>
        <xs:element name="And" substitutionGroup="Formula">
          <xs:complexType>
            <xs:sequence>
              <xs:element ref="Formula" minOccurs="2" maxOccurs="2"/>
              </xs:sequence>
          </xs:complexType>
        </xs:element>
    </xs:schema>""">

let formula = Prop.Parse """
    <And>
        <Prop>p1</Prop>
        <And>
            <Prop>p2</Prop>
            <Prop>p3</Prop>
        </And>
    </And>
    """

printfn "%s" formula.Props.[0] // p1
printfn "%s" formula.Ands.[0].Props.[0] // p2
printfn "%s" formula.Ands.[0].Props.[1] // p3

Substitution groups are like choices, and the type provider produces an optional property for each alternative.

Validation

The GetSchema method on the generated type returns an instance of System.Xml.Schema.XmlSchemaSet that can be used to validate documents:

open System.Xml.Schema
let schema = Person.GetSchema()
turing.XElement.Document.Validate(schema, validationEventHandler = null)

The Validate method accepts a callback to handle validation issues; passing null will turn validation errors into exceptions. There are overloads to allow other effects (for example setting default values by enabling the population of the XML tree with the post-schema-validation infoset; for details see the documentation).

Remarks on using a schema

The XML Type Provider supports most XSD features. Anyway the XML Schema specification is rich and complex and also provides a fair degree of openness which may be difficult to handle in data binding tools; but in FSharp.Data, when providing typed views on elements becomes too challenging (take for example wildcards) the underlying XElement is still available.

An important design decision is to focus on elements and not on complex types; while the latter may be valuable in schema design, our goal is simply to obtain an easy and safe way to access xml data. In other words the provided types are not intended for domain modeling (it's one of the very few cases where optional properties are preferred to sum types). Hence, we do not provide types corresponding to complex types in a schema but only corresponding to elements (of course the underlying complex types still affect the shape of the provided types but this happens only implicitly). Focusing on element shapes let us generate a type that should be essentially the same as one inferred from a significant set of valid samples. This allows a smooth transition (replacing Sample with Schema) when a schema becomes available.

Related articles

• Using JSON provider in a library also applies to XML type provider
• API Reference: XmlProvider type provider
• API Reference: XElementExtensions