package xmlm

  1. Overview
  2. Docs

Streaming XML codec.

A well-formed sequence of signals represents an XML document tree traversal in depth first order (this has nothing to do with XML well-formedness). Input pulls a well-formed sequence of signals from a data source and output pushes a well-formed sequence of signals to a data destination. Functions are provided to easily transform sequences of signals to/from arborescent data structures.

Consult the features and limitations and examples of use.

References

Basic types and values

type encoding = [
  1. | `UTF_8
  2. | `UTF_16
    (*

    Endianness determined from the BOM.

    *)
  3. | `UTF_16BE
  4. | `UTF_16LE
  5. | `ISO_8859_1
  6. | `US_ASCII
]

The type for character encodings. For `UTF_16, endianness is determined from the BOM.

type dtd = string option

The type for the optional DTD.

type name = string * string

The type for attribute and element's expanded names (uri,local). An empty uri represents a name without a namespace name, i.e. an unprefixed name that is not under the scope of a default namespace.

type attribute = name * string

The type for attributes. Name and attribute data.

type tag = name * attribute list

The type for an element tag. Tag name and attribute list.

type signal = [
  1. | `Dtd of dtd
  2. | `El_start of tag
  3. | `El_end
  4. | `Data of string
]

The type for signals. A well-formed sequence of signals belongs to the language of the doc grammar :

doc ::= `Dtd tree
tree ::= `El_start child `El_end
child ::= `Data trees | trees
trees ::= tree child | epsilon

The trees production is used to expresses the fact that there will never be two consecutive `Data signals in the children of an element.

Input and output deal only with well-formed sequences or exceptions are raised. However on output consecutive `Data signals are allowed.

val ns_xml : string

Namespace name value bound to the reserved "xml" prefix.

val ns_xmlns : string

Namespace name value bound to the reserved "xmlns" prefix.

val pp_dtd : Format.formatter -> dtd -> unit

pp_dtd ppf dtd prints an unspecified representation of dtd on ppf.

val pp_name : Format.formatter -> name -> unit

pp_name ppf name prints an unspecified representation of name on ppf.

val pp_attribute : Format.formatter -> attribute -> unit

pp_attribute ppf att prints an unspecified representation of att on ppf.

val pp_tag : Format.formatter -> tag -> unit

pp_tag ppf tag prints an unspecified representation of tag on ppf.

val pp_signal : Format.formatter -> signal -> unit

pp_signal ppf s prints an unspecified representation of s on ppf.

Input

type pos = int * int

The type for input positions. Line and column number, both start with 1.

type error = [
  1. | `Max_buffer_size
    (*

    Maximal buffer size exceeded (Sys.max_string_length).

    *)
  2. | `Unexpected_eoi
    (*

    Unexpected end of input.

    *)
  3. | `Malformed_char_stream
    (*

    Malformed underlying character stream.

    *)
  4. | `Unknown_encoding of string
    (*

    Unknown encoding.

    *)
  5. | `Unknown_entity_ref of string
    (*

    Unknown entity reference, details.

    *)
  6. | `Unknown_ns_prefix of string
    (*

    Unknown namespace prefix details

    *)
  7. | `Illegal_char_ref of string
    (*

    Illegal character reference.

    *)
  8. | `Illegal_char_seq of string
    (*

    Illegal character sequence.

    *)
  9. | `Expected_char_seqs of string list * string
    (*

    Expected one of the character sequences in the list but found another.

    *)
  10. | `Expected_root_element
    (*

    Expected the document's root element.

    *)
]

The type for input errors.

val error_message : error -> string

Converts the error to an english error message.

exception Error of pos * error

Raised on input errors.

type source = [
  1. | `Channel of in_channel
  2. | `String of int * string
  3. | `Fun of unit -> int
]

The type for input sources. For `String starts reading at the given integer position. For `Fun the function must return the next byte as an int and raise End_of_file if there is no such byte.

type input

The type for input abstractions.

val make_input : ?enc:encoding option -> ?strip:bool -> ?ns:(string -> string option) -> ?entity:(string -> string option) -> source -> input

Returns a new input abstraction reading from the given source.

  • enc, character encoding of the document, details. Defaults to None.
  • strip, strips whitespace in character data, details. Defaults to false.
  • ns is called to bind undeclared namespace prefixes, details. Default returns always None.
  • entity is called to resolve non predefined entity references, details. Default returns always None.
val input : input -> signal

Inputs a signal. Repeated invocation of the function with the same input abstraction will generate a well-formed sequence of signals or an Error is raised. Furthermore there will be no two consecutive `Data signals in the sequence and their string is always non empty.

This behaviour is deprecated: after a well-formed sequence was input another may be input, see eoi and details.

Raises Error on input errors.

val input_tree : el:(tag -> 'a list -> 'a) -> data:(string -> 'a) -> input -> 'a

If the next signal is a :

  • `Data signal, inputs it and invokes data with the character data.
  • `El_start signal, inputs the sequence of signals until its matching `El_end and invokes el and data as follows

    • el, is called on each `El_end signals with the corresponding `El_start tag and the result of the callback invocation for the element's children.
    • data, is called on each `Data signals with the character data. This function won't be called twice consecutively or with the empty string.
  • Other signals, raises Invalid_argument.

Raises Error on input errors and Invalid_argument if the next signal is not `El_start or `Data.

val input_doc_tree : el:(tag -> 'a list -> 'a) -> data:(string -> 'a) -> input -> dtd * 'a

Same as input_tree but reads a complete well-formed sequence of signals.

Raises Error on input errors and Invalid_argument if the next signal is not `Dtd.

val peek : input -> signal

Same as input but doesn't remove the signal from the sequence.

Raises Error on input errors.

val eoi : input -> bool

Returns true if the end of input is reached. See details.

Raises Error on input errors.

val pos : input -> pos

Current position in the input abstraction.

Output

type 'a frag = [
  1. | `El of tag * 'a list
  2. | `Data of string
]

The type for deconstructing data structures of type 'a.

type dest = [
  1. | `Channel of out_channel
  2. | `Buffer of Buffer.t
  3. | `Fun of int -> unit
]

The type for output destinations. For `Buffer, the buffer won't be cleared. For `Fun the function is called with the output bytes as ints.

type output

The type for output abstractions.

val make_output : ?decl:bool -> ?nl:bool -> ?indent:int option -> ?ns_prefix:(string -> string option) -> dest -> output

Returns a new output abstraction writing to the given destination.

  • decl, if true the XML declaration is output (defaults to true).
  • nl, if true a newline is output when the root's element `El_end signal is output. Defaults to false.
  • indent, identation behaviour, see details. Defaults to None.
  • ns_prefix, undeclared namespace prefix bindings, see details. Default returns always None.
val output : output -> signal -> unit

Outputs a signal.

This behaviour is deprecated: after a well-formed sequence of signals was output a new well-formed sequence can be output.

Raises Invalid_argument if the resulting signal sequence on the output abstraction is not well-formed or if a namespace name could not be bound to a prefix.

val output_depth : output -> int

output_depth o is o's current element nesting level (undefined before the first `El_start and after the last `El_end).

val output_tree : ('a -> 'a frag) -> output -> 'a -> unit

Outputs signals corresponding to a value by recursively applying the given value deconstructor.

Raises see output.

val output_doc_tree : ('a -> 'a frag) -> output -> (dtd * 'a) -> unit

Same as output_tree but outputs a complete well-formed sequence of signals.

Raises see output.

Functorial interface (deprecated)

WARNING. The functioral interface is deprecated and will be removed.

Make allows client to specify types for strings and internal buffers. Among other things this can be used to perform hash-consing or to process the character stream, e.g. to normalize unicode characters or to convert to a custom encoding.

type std_string = string
type std_buffer = Buffer.t
module type String = sig ... end

Input signature for strings.

module type Buffer = sig ... end

Input signature for internal buffers.

module type S = sig ... end

Output signature of Make.

module Make (String : String) (Buffer : Buffer with type string = String.t) : S with type string = String.t

Functor building streaming XML IO with the given strings and buffers.

Features and limitations

The module assumes strings are immutable, thus strings the client gives or receives during the input and output process must not be modified.

Input

Encoding

The parser supports ASCII, US-ASCII, UTF-8, UTF-16, UTF-16LE, UTF-16BE and ISO-8559-1 (Latin-1) encoded documents. But strings returned by the library are always UTF-8 encoded.

The encoding can be specified explicitly using the optional argument enc. Otherwise the parser uses UTF-16 or UTF-8 if there is a BOM at the beginning of the document. If there is no BOM it uses the encoding specified in the XML declaration. Finally, if there is no XML declaration UTF-8 is assumed.

White space handling

The parser performs attribute data normalization on every attribute data. This means that attribute data does not have leading and trailling white space and that any white space is collapsed and transformed to a single space character (U+0020).

White space handling of character data depends on the strip argument. If strip is true, character data is treated like attribute data, white space before and after elements is removed and any white space is collapsed and transformed to a single space character (U+0020), except if the data is under the scope of a xml:space attribute whose value is preserve. If strip is false all white space data is preserved as present in the document (however all kinds of line ends are translated to the newline character (U+000A).

Namespaces

Xmlm's names are expanded names. The parser automatically handles the document's namespace declarations. Undeclared namespace prefixes can be bound via the callback ns, which must return a namespace name. If ns returns None an `Unknown_ns_prefix error is raised.

Attributes used for namespace declarations are preserved by the parser. They are in the ns_xmlns namespace. Default namespace declarations made with xmlns have the attribute name (Xmlm.ns_xmlns, "xmlns"). Prefix declarations have the prefix as the local name, for example xmlns:ex results in the attribute name (Xmlm.ns_xmlns, "ex").

Regarding constraints on the usage of the xml and xmlns prefixes by documents, the parser does not report errors on violations of the must constraints listed in this paragraph.

Character and entity references

Character references and predefined entities are automatically resolved. Other entity references can be resolved by the callback entity, which must return an UTF-8 string corresponding to the replacement character data. The replacement data is not analysed for further references, it is added to the data as such modulo white space stripping. If entity returns None the error `Unknown_entity_ref is returned.

Sequences of documents (deprecated)

WARNING. This feature is deprecated and will be removed.

When a well-formed sequence of signals is input, no data is consumed beyond the closing '>' of the document's root element.

If you want to parse a document as defined in the XML specification, call eoi after a well-formed sequence of signals, it must return true. If you expect another document on the same input abstraction a new well-formed sequence of signals can be input. Use eoi to check if a document follows (this may consume data).

Invoking eoi after a well-formed sequence of signals skips whitespaces, comments and processing instructions until it gets to either an XML declaration or a DTD or the start of a new element or the end of input (in which case eoi returns true). If there is a new document but there is no XML declaration or the declaration specifies UTF-16, the same encoding as for the previous document is used.

Miscellaneous

  • Parses the more liberal and simpler XML 1.1 Name definition (minus ':' because of namespaces).
  • The DTD is parsed roughly (no guarantee it is well formed) and its information is ignored.
  • The parser drops comments, processing instructions, and standalone declaration.
  • Element attributes are not checked for uniqueness.
  • Attribute and character data chunks are limited by Sys.max_string_length. The error `Max_buffer_size is raised if the limit is hit.
  • Tail recursive.
  • Non validating.

Output

Encoding

Outputs only UTF-8 encoded documents. Strings given to output functions must be UTF-8 encoded, no checks are performed. Unicode characters that are not legal XML characters are replaced by the Unicode replacement character.

Namespaces

Xmlm's names are expanded names. Expanded names are automatically converted to qualified names by the output abstraction. There is no particular api to specify prefixes and default namespaces, the actual result depends solely on the output of attributes belonging to the ns_xmlns namespace. For example to set the default namespace of an element to http://example.org/myns, use the following attribute :

(* xmlns='http://example.org/myns' *)
let default_ns = (Xmlm.ns_xmlns, "xmlns"), "http://example.org/myns"

To bind the prefix "ex" to http://example.org/ex, use the following attribute :

(* xmlns:ex='http://example.org/ex' *)
let ex_ns = (Xmlm.ns_xmlns, "ex"), "http://example.org/ex"

Note that outputing input signals without touching namespace declaration attributes will preserve existing prefixes and bindings provided the same namespace name is not bound to different prefixes in a given context.

The callback ns_prefix of an output abstraction can be used to give a prefix to a namespace name lacking a prefix binding in the current output scope. Given a namespace name the function must return the prefix to use. Note that this will not add any namespace declaration attribute to the output. If the function returns None, output will raise Invalid_argument. The default function returns always None.

Indentation

Output can be indented by specifying the indent argument when an output abstraction is created. If indent is None (default) signal output does not introduce any extra white space. If ident is Some c, each signal is output on its own line (for empty elements `El_start and `El_end are collapsed on a single line) and nested elements are indented with c space characters.

Sequences of documents (deprecated)

WARNING. This feature is deprecated and will be removed.

After a well-formed sequence of signals was output, the output abstraction can be reused to output a new well-formed sequence of signals.

Miscellaneous

  • Output on a channel does not flush it.
  • In attribute and character data you provide, markup delimiters '<','>','&', and '\"' are automatically escaped to predefined entities.
  • No checks are peformed on the prefix and local part of output names to verify they are NCNames. For example using the tag name ("","dip d") will produce a non well-formed document because of the space character.
  • Tail recursive.

Tips

  • The best options to do an input/output round trip and preserve as much information as possible is to input with strip = false and output with indent = None.
  • Complete whitespace control on output is achieved with indent = None and suitable `Data signals

Examples

Sequential processing

Sequential processing has the advantage that you don't need to get the whole document tree in memory to process it.

The following function reads a single document on an input channel and outputs it.

let id ic oc =
let i = Xmlm.make_input (`Channel ic) in
let o = Xmlm.make_output (`Channel oc) in
let rec pull i o depth =
  Xmlm.output o (Xmlm.peek i);
  match Xmlm.input i with
  | `El_start _ -> pull i o (depth + 1)
  | `El_end -> if depth = 1 then () else pull i o (depth - 1)
  | `Data _ -> pull i o depth
  | `Dtd _ -> assert false
in
Xmlm.output o (Xmlm.input i); (* `Dtd *)
pull i o 0;
if not (Xmlm.eoi i) then invalid_arg "document not well-formed"

The following function reads a sequence of documents on an input channel and outputs it.

let id_seq ic oc =
let i = Xmlm.make_input (`Channel ic) in
let o = Xmlm.make_output ~nl:true (`Channel oc) in
while not (Xmlm.eoi i) do Xmlm.output o (Xmlm.input i) done

The following function reads a sequence of documents on the input channel. In each document's tree it prunes non root elements whose name belongs to prune_list.

let prune_docs prune_list ic oc =
let i = Xmlm.make_input (`Channel ic) in
let o = Xmlm.make_output ~nl:true (`Channel oc) in
let copy i o = Xmlm.output o (Xmlm.input i) in
let prune (name, _) = List.mem name prune_list in
let rec process i o d =
  let rec skip i d = match Xmlm.input i with
  | `El_start _ -> skip i (d + 1)
  | `El_end -> if d = 1 then () else skip i (d - 1)
  | s -> skip i d
  in
  match Xmlm.peek i with
  | `El_start tag when prune tag -> skip i 0; process i o d
  | `El_start _ -> copy i o; process i o (d + 1)
  | `El_end -> copy i o; if d = 0 then () else process i o (d - 1)
  | `Data _ -> copy i o; process i o d
  | `Dtd _ -> assert false
in
let rec docs i o =
  copy i o; (* `Dtd *)
  copy i o; (* root start *)
  process i o 0;
  if Xmlm.eoi i then () else docs i o
in
docs i o

Tree processing

A document's sequence of signals can be easily converted to an arborescent data structure. Assume your trees are defined by :

type tree = E of Xmlm.tag * tree list | D of string

The following functions input/output xml documents from/to abstractions as value of type tree.

let in_tree i =
  let el tag childs = E (tag, childs)  in
  let data d = D d in
  Xmlm.input_doc_tree ~el ~data i

let out_tree o t =
  let frag = function
  | E (tag, childs) -> `El (tag, childs)
  | D d -> `Data d
  in
  Xmlm.output_doc_tree frag o t

Tabular data processing

We show how to process XML data that represents tabular data (some people like do that).

The file we need to deal with represents nominal data about W3C bureaucrats. There are no namespaces and attributes are ignored. The element structure of the document is :

  • <list>

    • <bureaucrat> represents a W3C bureaucrat (zero or more).

      A bureaucrat contains the following elements, in order.

      • <name> its name (mandatory, string).
      • <surname> its surname (mandatory, string).
      • <honest> present iff he implemented one of its spec (optional, empty).
      • <obfuscation_level> its grade on the open scale of obfuscation (mandatory, float).
      • <tr> (zero or more, string), technical reports he worked on.

In OCaml we represent a W3C bureaucrat by this type :

type w3c_bureaucrat = {
  name : string;
  surname : string;
  honest : bool;
  obfuscation_level : float;
  trs : string list; }

The following functions input and output W3C bureaucrats as lists of values of type w3c_bureaucrat.

let in_w3c_bureaucrats src =
  let i = Xmlm.make_input ~strip:true src in
  let tag n = ("", n), [] in
  let error () = invalid_arg "parse error" in
  let accept s i = if Xmlm.input i = s then () else error () in
  let rec i_seq el acc i = match Xmlm.peek i with
  | `El_start _ -> i_seq el ((el i) :: acc) i
  | `El_end -> List.rev acc
  | _ -> error ()
  in
  let i_el n i =
    accept (`El_start (tag n)) i;
    let d = match Xmlm.peek i with
    | `Data d -> ignore (Xmlm.input i); d
    | `El_end -> ""
    | _ -> error ()
    in
    accept (`El_end) i;
    d
  in
  let i_bureaucrat i =
    try
      accept (`El_start (tag "bureaucrat")) i;
      let name = i_el "name" i in
      let surname = i_el "surname" i in
      let honest = match Xmlm.peek i with
      | `El_start (("", "honest"), []) -> ignore (i_el "honest" i); true
      | _ -> false
      in
      let obf = float_of_string (i_el "obfuscation_level" i) in
      let trs = i_seq (i_el "tr") [] i in
      accept (`El_end) i;
      { name = name; surname = surname; honest = honest;
        obfuscation_level = obf; trs = trs }
    with
    | Failure _ -> error () (* float_of_string *)
  in
  accept (`Dtd None) i;
  accept (`El_start (tag "list")) i;
  let bl = i_seq i_bureaucrat [] i in
  accept (`El_end) i;
  if not (Xmlm.eoi i) then invalid_arg "more than one document";
  bl

let out_w3c_bureaucrats dst bl =
  let tag n = ("", n), [] in
  let o = Xmlm.make_output ~nl:true ~indent:(Some 2) dst in
  let out = Xmlm.output o in
  let o_el n d =
    out (`El_start (tag n));
    if d <> "" then out (`Data d);
    out `El_end
  in
  let o_bureaucrat b =
    out (`El_start (tag "bureaucrat"));
    o_el "name" b.name;
    o_el "surname" b.surname;
    if b.honest then o_el "honest" "";
    o_el "obfuscation_level" (string_of_float b.obfuscation_level);
    List.iter (o_el "tr") b.trs;
    out `El_end
  in
  out (`Dtd None);
  out (`El_start (tag "list"));
  List.iter o_bureaucrat bl;
  out (`El_end)
OCaml

Innovation. Community. Security.