Plasma GitLab Archive
Projects Blog Knowledge


The PXP Preprocessor

Since PXP-1.1.95, there is a preprocessor as part of the PXP distribution. It allows you to compose XML trees and event lists dynamically, which is very handy to write XML transformations.

To enable the preprocessor, compile your source files as in:

 ocamlfind ocamlc -syntax camlp4o -package pxp-pp,... ... 

The package pxp-pp contains the preprocessor. The -syntax option enables camlp4, on which the preprocessor is based. It is also possible to use it together with the revised syntax, use -syntax camlp4r in this case.

In the toploop, type

# #use "topfind";;
# #camlp4o;;
# #require "pxp-pp";;
# #require "pxp";;

The preprocessor defines the following new syntax notations, explained below in detail:

  • <:pxp_charset< CHARSET_DECL >>
  • <:pxp_tree< EXPR >>
  • <:pxp_vtree< EXPR >>
  • <:pxp_evlist< EXPR >>
  • <:pxp_evpull< EXPR >>
  • <:pxp_text< TEXT >>
The basic notation is pxp_tree which creates a tree of PXP document nodes as described in EXPR. pxp_vtree is the variant where the tree is immediately validated - with pxp_tree the tree is not validated, but one can validate it later (e.g. when the whole output tree of the program is built up). pxp_evlist creates a list of PXP events instead of nodes, useful together with the event parser. pxp_evpull is a variation of the latter: Instead of an event list an event generator is created that works like a pull parser.

The pxp_charset notation only configures the character sets to assume. Finally, pxp_text is a notation for string literals.

Creating constant XML with pxp_tree - basic syntax

The following examples are all written for pxp_tree. You can also use one of the other XML composers instead, but see the notes below that explain a few differences.

In order to use pxp_tree, you must define two variables in the environment: spec and dtd:

let spec = Pxp_tree_parser.default_spec
let dtd = Pxp_dtd.create_dtd `Enc_iso88591

These variables are assumed to exist by the generated code. The dtd variable is the DTD object. Note that you need it even in well-formedness mode (validation turned off) - see the explanations in Parsing in well-formedness mode to understand why. The spec variable controls which classes are instantiated as node representation. See Specifying which classes implement nodes - the mysterious spec parameter for the meaning of spec.

Elements, attributes, and data nodes

Now you can create XML trees like in

let book = 
      [ <title>[ "The Lord of The Rings" ]
        <author>[ "J.R.R. Tolkien" ]

As you can see, the syntax is somehow XML-related but not really XML. (Many ideas are borrowed from CDUCE, by the way.) In particular, there are start tags like <title> but no end tags. Instead, we are using square brackets to denote where the list of the children of the XML element starts and where it ends. Furthermore, character data must be put into double quotes.

You may ask why the well-known XML syntax has been modified for this preprocessor. There are many reasons, and they will become clearer in the following explanations. For now, you can see the advantage that the syntax is less verbose, as you need not to repeat the element names in end tags (I know programmers like brevity). Furthermore, you can exactly control which characters are part of the data nodes without having to make compromises with indentation.

Attributes are written as in XML:

let book = 
    <book id="BOOK_001">
      [ <title lang="en">[ "The Lord of The Rings" ]
        <author>[ "J.R.R. Tolkien" ]

An element without children can be written


or slightly shorter:


Processing instructions and comments

You can also create processing instructions and comment nodes:

let list =
      [ <!>"Now the list of books follows!"
        <?>"formatter_directive" "one book per page"

The notation <!> creates a comment node with the following string as contents. The notation <?> for constructing processing instructions needs two strings, first the target, then the value (here, this results in <?formatter_directive one book per page?>).

Look again at the last example: The O'Caml variable book occurs, and it inserts its tree into the list of books. Identifiers without "decoration" just refer to O'Caml variables. We will see more examples below.

Elements with one child

The preprocessor syntax knows a number of shortcuts and variations. First, you can omit the square brackets when an element has exactly one child:

<element><child>"Data inside child"

This is the same as

<element>[ <child>[ "Data inside child" ] ]

Details of data nodes

Second, we already have used a common abbreviation: Strings are automatically converted to data nodes. The "expanded" syntax is

<*>"Data string"

where <*> denotes to construct a data node, and the following string is used as contents. Usually, you can omit <*>, so this is the same as

"Data string"

However, there are a few occasions where the <*> notation is still useful, see below (essentially, it also works like a type annotation: the following subexpression must be a string).

Inside strings, the usual entity references can be used: &lt;, &gt;, &amp;, &quot;, &apos;, and also numeric references work: &#n; where n is a number. Note that &lt;, &gt;, and &apos; are not obligatory, as <, >, and ' can be included directly.

Example: "Double quotes: &quot;". For a newline character, write &#10;.


The preprocessor knows two operators: ^ concatenates strings, and @ concatenates lists. Examples:

<element>[ "Word1" ^ "Word2" ]
<element>([ <a/> ] @ [ <b/> ])

Parentheses can be used to clarify precedence. For example:

<element>(l1 @ l2)

Without parentheses, the concatenation operator @ would be parsed as

(<element> l1) @ l2

Parentheses may be used in any expression.

Super root

Rarely used, there is also a notation for the "super root" nodes. For uses of this node types, see Choosing the node types to represent.

<^>[ <element> ... ]

Dynamic XML

This section describes how to insert dynamically created content into XML trees.

Let us begin with an example. The task is to convert O'Caml values of type

type book = 
  { title : string;
    author : string;
    isbn : string;

to XML trees like

<book id="BOOK_{isbn}">

(conventional syntax, with placeholders in {braces}). When b is the book variable, the solution is

let book = 
  let title = b.title
  and author =
  and isbn = b.isbn in
    <book id=("BOOK_" ^ isbn)>
      [ <title><*>title

First, we bind the simple O'Caml variables title, author, and isbn. The reason is that the preprocessor syntax does not allow expressions like b.title directly in the XML tree (but see below for another, often better workaround).

The XML tree contains the O'Caml variables. The id attribute is a concatenation of the fixed prefix BOOK_ and the contents of isbn. The title and author elements contain a data node whose contents are the O'Caml strings title, and author, respectively.

Why <*>? If we just wrote <title>title, the generated code would assume that the title variable is an XML node (of type Pxp_document.node), and not a string. From this point of view, <*> works like a type annotation, as it specialises the type of the following expression.

The (: ... :) notation

Here is an alternate solution:

let book = 
    <book id=("BOOK_" ^ (: b.isbn :))>
      [ <title><*>(: b.title :)
        <author><*>(: :)

The notation (: ... :) allows you to include arbitrary O'Caml expressions into the tree. In this solution it is no longer necessary to artificially create O'Caml variables for the only purpose of injecting values into trees.

Dynamic names

It is possible to create XML elements with dynamic names: Just put parentheses around the expression. Example:

let name = "book" in
<:pxp_tree< <(name)> ... >>

With the same notation, one can also set attribute names dynamically:

let att_name = "id" in
<:pxp_tree< <book (att_name)=...> ... >>

Dynamic attribute lists

Finally, it is also possible to include complete attribute lists dynamically:

let att_list = [ "id", ("BOOK_" ^ b.isbn) ] in
<:pxp_tree< <book (: att_list :) > ... >>

Here, att_list must be a (string*string) list with the attributes to include.


Depending on where a variable or O'Caml expression occurs, different types are assumed. Compare the following examples:

<:pxp_tree< <element>x1 >>
<:pxp_tree< <element>[x2] >>
<:pxp_tree< <element><*>x3 >>

As a rule of thumb, the most general type is assumed that would make sense at a certain location. As x1 could be replaced by a list of children, its type is assumed to be a node list. As x2 could be replaced by a single node, its type is assumed to be a node. And x3 is a string, we had this case already.

Character encodings: pxp_charset

As the preprocessor generates code that builds XML trees, it must know two character encodings:

  • Which encoding is used in the source code (in the .ml file)
  • Which encoding is used in the XML representation, i.e. in the O'Caml values representing the XML trees
Both encodings can be set independently. The syntax is:

<:pxp_charset< source="ENC" representation="ENC" >>

where ENC is the name of the selected encoding. The default is ISO-8859-1 for both encodings. For example, to set the representation encoding to UTF-8, use:

<:pxp_charset< representation="UTF-8" >>

The pxp_charset notation is a constant expression that always evaluates to (). (A requirement by camlp4 that looks artificial.)

When you set the representation encoding, it is required that the encoding stored in the DTD object is the same. Remember that we need a DTD object like

let dtd = Pxp_dtd.create_dtd `Enc_iso88591

Of course, we must change this to the representation encoding, too. In our example:

let dtd = Pxp_dtd.create_dtd `Enc_utf8

The preprocessor cannot check this at compile time, and for performance reasons, a runtime check is not generated. So it is up to the programmer that the character encodings are used in a consistent way.

Validated trees: pxp_text, calling validate, and pxp_vtree

In order to validate trees, you need a filled DTD object. In principle, you can create this object by a number of methods. For example, you can parse an external file:

let dtd = Pxp_dtd_parser.parse_dtd_entity config (from_file "sample.dtd")

It is, however, often more convenient to include the DTD literally into the program. This works by

let dtd = Pxp_dtd_parser.parse_dtd_entity config (from_string "...")

As the double quotes are often used inside DTDs, O'Caml string literals are a bit impractical, as they are also delimited by double quotes, and one needs to add backslashes as escape characters. The pxp_text notation is often more readable here:


is just another way of writing "STRING". In our DTD, we have

let dtd_text =
    <!ELEMENT book (title,author)>
    <!ELEMENT title (#PCDATA)>
    <!ATTLIST title lang CDATA "en">
    <!ELEMENT author (#PCDATA)>
let config = default_config
let dtd = Pxp_dtd_parser.parse_dtd_entity config (from_string dtd_text)

Note that pxp_text is not restricted to DTDs, as it can be used for any kind of string.

After we have the DTD, we can validate the trees. One option is to call the Pxp_document.validate function:

let book = 
      [ <title>[ "The Lord of The Rings" ]
        <author>[ "J.R.R. Tolkien" ]
let () =
  Pxp_document.validate book

(This example is invalid, and validate will throw an exception, as the id attribute is missing.)

Note that it is a misunderstanding that pxp_tree builds XML trees in well-formedness mode. You can create any tree with it, and the fact is that pxp_tree just does not invoke the validator. So if the DTD enforces validation, the tree is validated when the validate function is called. If the DTD is in well-formedness mode, the tree is effectively not validated, even when the validate function is invoked. Btw, the following statements would create a DTD in well-formedness mode:

let dtd = Pxp_dtd.create_dtd `Enc_iso88591
let () = dtd # allow_arbitrary

Validating with pxp_vtree

As an alternative of calling the validate function, one can also use pxp_vtree instead. It immediately validates every XML element it creates. However, "injected" subtrees are not validated, i.e. validation does not proceed recursively to subnodes as the validate function does it.

pxp_vtree has the same syntax as pxp_tree.

Generating events: pxp_evlist and pxp_evpull

As PXP has also an event model to represent XML, the preprocessor can also produce such events. In particular, there are two modes: The pxp_evlist notation outputs lists of events (of type Pxp_types.event list) representing the XML expression. The pxp_evpull notation creates an automaton from which one can "pull" events (like from a pull parser). The automaton has type unit -> Pxp_types.event.


Syntactically, these two notations work very much like pxp_tree. For example,

let book = 
      [ <title>[ "The Lord of The Rings" ]
        <author>[ "J.R.R. Tolkien" ]

returns this list of events:

E_start_tag ("book", [], None, <obj>);
  E_start_tag ("title", [], None, <obj>);
  E_char_data "The Lord of The Rings"
  E_end_tag ("title", <obj>);
  E_start_tag ("author", [], None, <obj>); 
  E_char_data "J.R.R. Tolkien";
  E_end_tag ("author", <obj>); 
  E_end_tag ("book", <obj>)

(Here, <obj> denotes the entity_id object for identifying the containing entity.)

Note that you neither need a dtd variable nor a spec variable in event mode.

There is one important culprit: Both single nodes and lists of nodes are represented by the same type, Pxp_types.event list. That has the consequence that in the following example x1 and x2 have the same type Pxp_types.event list:

<:pxp_evlist< <element>x1 >>
<:pxp_evlist< <element>[x2] >>

In principle, it could be checked at runtime whether x1 and x2 have the right structure. However, this is not done because of performance reasons, and because the generated XML is still well-formed. The typing is just different from pxp_tree which distinguishes between a single node and a node list.


As mentioned, pxp_evpull works like a pull parser. After defining

let book = 
      [ <title>[ "The Lord of The Rings" ]
        <author>[ "J.R.R. Tolkien" ]

book is a function unit -> Pxp_types.event option. One can call it to pull the events out of it one after the other:

let e1 = book();;       (* = Some(E_start_tag ("book", [], None, <obj>)) *)
let e2 = book();;       (* = Some(E_start_tag ("title", [], None, <obj>)) *)

After the last event, book returns None to indicate the end of the event stream.

As for pxp_evlist, it is not possible to distinguish between single nodes and node lists by type. In this example, both x1 and x2 are assumed to have type unit -> Pxp_types.event:

<:pxp_evlist< <element>x1 >>
<:pxp_evlist< <element>[x2] >>

Note that <element>x1 actually means to build a new pull automaton around the existing pull automaton x1: The children of element are retrieved by pulling events from x1 until None is returned.

A consequence of the pull semantics is that once an event is obtained from an automaton, the state of the automaton is modified such that it is not possible to get the same event again. If you need an automaton that can be reset to the beginning, just wrap the pxp_evlist notation into a functional abstraction:

let book_maker() =
  <:pxp_evpull< <book ...> ... >>
let book1 = book_maker()
let book2 = book_maker()

This way, book1 and book2 generate independent event streams.

There is another implication of the nature of the automatons: Subexpressions are lazily evaluated. For example, in

<:pxp_evpull< <element>[ <*> (: get_data_contents() :) ] >>

the call of get_data_contents is performed just before the event for the data node is constructed instead of being done at automaton construction time.


Note that none of the notations pxp_tree, pxp_vtree, pxp_evlist, or pxp_evpull is able to create documents. They just create what is equivalent to the node tree inside a document, but not the document wrapping.

In the tree case, just put the node tree into a Pxp_document.document:

let book = <:pxp_tree< ... >>
let doc = new Pxp_document.document warner dtd#encoding
doc # init_root book "book"

In the event case, the generated events do not include E_start_doc, E_end_doc, or E_end_of_stream. If required, one has to add these events manually which is quite simple. For pxp_evlist, do something like

let doc =
  E_start_doc("1.0", dtd) ::
  ( <:pxp_evlist< <book>... >> @
    [ E_end_doc("book");

For pxp_evpull, do something like

let doc =
    [ Pxp_event.of_list [ E_start_doc("1.0", dtd) ];
      <:pxp_evpull< <book>... >>;
      Pxp_event.of_list [E_end_doc("book"); E_end_of_stream ]

(See Pxp_event.concat and Pxp_event.of_list.)


By default, the preprocessor does not generate nodes or events that support namespaces. It can, however, be configured to create namespace-aware XML aggregations.

In any case, you need a namespace manager. This is an object that tracks the usage of namespace prefixes in XML nodes. For example, we can create a namespace manager that knows the html prefix:

let mng = new Pxp_dtd.namespace_manager in
mng # add_namespace "html" ""

(Also see Pxp_dtd.namespace_manager.) Here, we declare that we want to use the html prefix for the internal representation of the XML nodes. This kind of prefix is called normalized prefix, or normprefix for short. It is possible to configure different prefixes for the external representation, i.e. when the XML tree is printed to a file. This other kind of prefix is called display prefix. We will have a look at them later. (For a more detailed discussion of namespaces, see Intro_namespaces.)

Next, we must tell the DTD object that we have a namespace manager:

let dtd = Pxp_dtd.create_dtd `Enc_iso88591
dtd # set_namespace_manager mng

For pxp_evlist and pxp_evpull we are now prepared (note that we need now a dtd variable, as only the DTD object knows the namespace manager). For pxp_tree and pxp_vtree, it is required to use a namespace-aware specification:

let spec = Pxp_tree_parser.default_namespace_spec 

(Normal specifications do not work, you would get "Namespace method not applicable" errors if you tried to use them.)

Using <:autoscope>

The special notation <:autoscope> enables namespace mode in this example:

let list =
        [ <html:li>"Item1"

In particular, <:autoscope> defines a new O'Caml variable for its subexpression: scope. This variable contains the namespace scope object, which contains the namespace declarations for the subexpression. <:autoscope> initialises this variable from the namespace manager such that it contains now a declaration for the html prefix. scope has type Pxp_dtd.namespace_scope.

In general, the namespace scope object contains the prefixes to use for the external representation (as opposed to the namespace manager which defines the prefixes for the internal representation). If the external prefixes can be the same as the internal ones, <:autoscope> is the right directive, as it initalizes the scope object with the prefixes from the namespace manager, so that both views are the same.

Print the tree by

list # display (`Out_channel stdout) `Enc_iso88591

Note that there is a display and a write method. The difference is that display prints the external prefixes (from scope), and that write prints the internal prefixes (from the namespace manager). In this introduction we prefer display.

Using <:scope> in its basic form

Alternatively, we can also create the scope variable manually:

let scope = Pxp_dtd.create_namespace_scope
              ~decl:[ """" ]
let list =
        [ <html:li>"Item1"

Note that we now use <:scope>. In this simple form, this construct just enables namespace mode, and takes the scope variable from the environment.

Furthermore, the namespace scope contains now a different namespace declaration: The display prefix "" is used for HTML. The empty prefix just means to declare a default prefix (by xmlns="URI"). The effect can be seen when the XML tree is printed by calling the display method.

If we had called create_namespace_scope with the decl argument

  ~decl:[ "foo""" ]

the displayed tree would use the foo prefix, and declare it as xmlns:foo="".

Using <:scope> to set display prefixes

Here is a third variant of the same example:

let scope = Pxp_dtd.create_namespace_scope mng
let list =
    <:scope ("")="">
        [ <html:li>"Item1"

The scope is now initially empty. The <:scope> notation is used to extend the scope for the time the subexpression is evaluated.

There is also a notation <:emptyscope> that creates an empty scope object, so one could even write

let list =
      <:scope ("")="">
          [ <html:li>"Item1"

The <:scope> notation can be used in any subexpression, and it modifies the display prefix to use in that subexpression. For example, here a different prefix foo is used for the second item:

let list =
      <:scope ("")="">
          [ <html:li>"Item1"
            <:scope foo="">

It is recommended to create the scope variable manually with a reasonable initial declaration, and to use <:scope> to enable namespace processing, and to extend the scope where necessary. The advantage of this approach is that the same scope object can be shared by many XML nodes, so you need less memory.

One tip: To get a namespace scope that is initialised with all prefixes of the namespace manager (as <:autoscope> does it), define

let scope = create_namespace_scope ~decl: mng#as_declaration mng

For event-based processing of XML, the namespace mode works in the same way as described here, there is no difference.

This web site is published by Informatikbüro Gerd Stolpmann
Powered by Caml