Up

module Nethtml

: sig

Parsing of HTML

#
type document =
# | Element of (string * (string * string) list * document list)
# | Data of string

The type document represents parsed HTML documents:

  • Element (name, args, subnodes) is an element node for an element of type name (i.e. written <name ...>...</name>) with arguments args and subnodes subnodes (the material within the element). The arguments are simply name/value pairs. Entity references (something like &xy;) occuring in the values are not resolved.
    Arguments without values (e.g. <select name="x" multiple>: here, multiple is such an argument) are represented as (name,name), i.e. the name is also returned as value.
    As argument names are case-insensitive, the names are all lowercase.
  • Data s is a character data node. Again, entity references are contained as such and not as what they mean.

Character encodings: The parser is restricted to ASCII-compatible encodings (see the function Netconversion.is_ascii_compatible for a definition). In order to read other encodings, the text must be first recoded to an ASCII-compatible encoding (example below). Names of elements and attributes must additionally be ASCII-only.

We also need a type that declares how to handle the various tags. This is called a "simplified DTD", as it is derived from SGML DTDs, but simplified to the extent used in the HTML definition.

The HTML specification (http://www.w3.org/TR/1999/REC-html401-19991224) is the reference for the HTML DTD. You can see there that most HTML elements are either an inline element, a block element, or both ("flow" element). The grammar of HTML is described in terms of these classes. For instance, a P tag (paragraph) is a block element and contains block elements whereas B (bold) is an inline element and contains inline elements. From this follows that you cannot put a P inside a B: <B><P>something</P></B> is illegal.

The parser needs this information to resolve such input, i.e. do something with bad HTML. As HTML allows tag minimization (many end tags can be omitted), the parser can read this as: <B></B><P>something</P> (and the </B> in the input is ignored).

If all start and all end tags are written out, changing the simplified_dtd does not make any difference.

There is no normative text that says how to read bad HTML. Because of this, it is - to a large degree - an interpretation of HTML what you put into simplified_dtd. We provide two versions:

  • html40_dtd: tries to be close to the official spec
  • relaxed_html40_dtd: tries to be close to what common web browsers implement
#
type element_class = [
| `Inline
| `Block
| `Essential_block
| `None
| `Everywhere
]

Element classes are a property used in the HTML DTD. For our purposes, we define element classes simply as an enumeration:

  • `Inline is the class of inline HTML elements
  • `Block is the class of block HTML elements
  • `Essential_block is a sub-class of `Block with the additional property that every start tag must be explicitly ended
  • `None means that the members of the class are neither block nor inline elements, but have to be handled specially
  • `Everywhere means that the members of the class can occur everywhere, regardless of whether a constraint allows it or not.
#
type model_constraint = [
| `Inline
| `Block
| `Flow
| `Empty
| `Any
| `Special
| `Elements of string list
| `Sub_exclusions of string list * model_constraint
]

Model constraints define the possible sub elements of an element:

  • `Inline: The sub elements must belong to the class `Inline
  • `Block: The sub elements must be members of the classes `Block or `Essential_block
  • `Flow: The sub elements must belong to the classes `Inline, `Block, or `Essential_block
  • `Empty: There are no sub elements
  • `Any: Any sub element is allowed
  • `Special: The element has special content (e.g. <script>). Functionally equivalent to `Empty
  • `Elements l: Only these enumerated elements may occur as sub elements
  • `Or(m1,m2): One of the constraints m1 or m2 must hold
  • `Except(m1,m2): The constraint m1 must hold, and m2 must not hold
  • `Sub_exclusions(l,m): The constraint m must hold; furthermore, the elements enumerated in list l are not allowed as direct or indirect subelements, even if m or the model of a subelement would allow them. The difference to `Except(m, `Elements l) is that the exclusion is inherited to the subelements. The `Sub_exclusions expression must be toplevel, i.e. it must not occur within an `Or, `Except, or another 'Sub_exclusions expression.

Note that the members of the class `Everywhere are allowed everywhere, regardless of whether the model constraint allows them or not.

Note that certain aspects are not modeled:

  • #PCDATA: We do not specify where PCDATA is allowed and where not.
  • Order, Number: We do neither specify in which order the sub elements must occur nor how often they can occur
  • Inclusions: DTDs may describe that an element extraordinarily allows a list of elements in all sub elements.
  • Optional tags: Whether start or end tags can be omitted (to some extent, this can be expressed with `Essential_block, however)
#
type simplified_dtd = (string * (element_class * model_constraint)) list

A simplified_dtd is an associative list of tuples (element_name, (element_class, constraint)): For every element_name it is declared that it is a member of element_class, and that the sub elements must satisfy constraint.

It is not allowed to have several entries for the same element.

#
val html40_dtd : simplified_dtd

The (transitional) HTML 4.0 DTD, expressed as simplified_dtd

#
val relaxed_html40_dtd : simplified_dtd

A relaxed version of the HTML 4.0 DTD that matches better common practice. In particular, this DTD additionally allows that inline elements may span blocks. For example,

<B>text1 <P>text2

is parsed as

<B>text1 <P>text2</P></B>

and not as

<B>text1 </B><P>text2</P>

\- the latter is more correct (and parsed by html40_dtd), but is not what users expect.

Note that this is still not what many browsers implement. For example, Netscape treats most inline tags specially: <B> switches bold on, </B> switches bold off. For example,

<A href='a'>text1<B>text2<A href='b'>text3

is parsed as

<A href='a'>text1<B>text2</B></A><B><A href='b'>text3</A></B>

\- there is an extra B element around the second anchor! (You can see what Netscape parses by loading a page into the "Composer".) IMHO it is questionable to consider inline tags as switches because this is totally outside of the HTML specification, and browsers may differ in that point.

Furthermore, several elements are turned into essential blocks: TABLE, UL, OL, and DL. David Fox reported a problem with structures like:

<TABLE><TR><TD><TABLE><TR><TD>x</TD></TD></TR></TABLE>y</TD></TR></TABLE>

i.e. the TD of the inner table has two end tags. Without additional help, the second </TD> would close the outer table cell. Because of this problem, tables are now essential meaning that it is not allowed to implicitly add a missing </TABLE>; every table element has to be explicitly ended. This rule seems to be what many browsers implement.

#
val parse_document : ?dtd:simplified_dtd -> ?return_declarations:bool -> ?return_pis:bool -> ?return_comments:bool -> Lexing.lexbuf -> document list

Parses the HTML document from a lexbuf and returns it.

dtd specifies the DTD to use. By default, html40_dtd is used which bases on the transitional HTML 4.0 DTD
return_declarations if set, the parser returns <!...> declarations as Element("!",["contents",c],[]) nodes, where c is the string inside <! and >. - By default, declarations are skipped.
return_pis if set, the parser returns <?...> (or <?...?>) processing instructions as Element("?",["contents",c],[]) nodes, where c is the string inside <? and > (or ?>). - By default, processing instructions are skipped.
return_comments if set, the parser returns <!-- .... --> comments as Element("--",["contents",c],[]) nodes, where c is the string inside <!-- and -->. - By default, comments are skipped.
#
val parse : ?dtd:simplified_dtd -> ?return_declarations:bool -> ?return_pis:bool -> ?return_comments:bool -> Netchannels.in_obj_channel -> document list

Parses the HTML document from an object channel and returns it. For example, to parse the HTML string s:

     let ch = Netchannels.input_string s in
     let doc = parse ch

Arguments are the same as in parse_document.

Note on XHTML

The parser can read XHTML, as long as the following XML features are not used:

  • Internal DTD subset, i.e. <!DOCTYPE html ... [ ... ]>
  • External entities
  • <![CDATA[
  • <![INCLUDE[
  • <![IGNORE[

The following XML features are ok:

  • Processing instructions
  • Empty elements (e.g. <br/>) as long as the element is declared as `Empty.

Note on Character Encodings

The parser can only read character streams that are encoded in an ASCII- compatible way. For example, it is possible to read a UTF-8-encoded stream, but not a UTF-16-encoded stream. All bytes between 1 and 127 are taken as ASCII, and other bytes are ignored (copied from input to output).

Non-ASCII-compatible streams must be recoded first. For example, to read a UTF-16-encoded netchannel ch, use:

   let p = 
     new Netconversion.recoding_pipe ~in_enc:`Enc_utf16 ~out_enc:`Enc_utf8 () in
   let ch' =
     new Netchannels.input_filter ch p in
   let doc =
     Nethtml.parse ch' in
   ch' # close_in();
   ch # close_in();
#
val decode : ?enc:Netconversion.encoding -> ?subst:(int -> string) -> ?entity_base:[
| `Html
| `Xml
| `Empty
] -> ?lookup:(string -> string) -> ?dtd:simplified_dtd -> document list -> document list

Converts entities &name; and &#num; into the corresponding characters. The argument enc must indicate the character set of the document (by default ISO-8859-1 for backwards compatibility). If a character cannot be represented in this encoding, the function subst is called (input is the Unicode code point, output is the substituted string). By default, the function fails if such a character is found.

The arg entity_base selects which entities can be converted (see Netencoding.Html.decode). The function lookup is called for all unknown &name; entities. By default, this function fails.

Declarations, processing instructions, and comments are not decoded. The same also applies to elements declared as `Special in the DTD. The dtd argument determines the DTD, by default html40_dtd is assumed.

#
val encode : ?enc:Netconversion.encoding -> ?prefer_name:bool -> ?dtd:simplified_dtd -> document list -> document list

Converts problematic characters to their corresponding entities. The argument enc must indicate the character set of the document (by default ISO-8859-1 for backwards compatibility). If prefer_name, the algorithm tries to find the named entities (&name;); otherwise only numeric entities (&#num;) are generated. Names are preferred by default.

Declarations, processing instructions, and comments are not encoded. The same also applies to elements declared as `Special in the DTD. The dtd argument determines the DTD, by default html40_dtd is assumed.

#
val map_list : (string -> string) -> document list -> document list

map_list f doclst: Applies f to all attribute values and data strings (except the attributes of "?", "!", or "--" nodes).

This can be used to change the text encoding of a parsed document:

     let doc' = map_list String.lowercase doc

converts all text data to lowercase characters.

#
type xmap_value =
# | Xmap_attribute of string * string * string
# | Xmap_data of string option * string
#
val xmap_list : (xmap_value -> string) -> string option -> document list -> document list

xmap_list f surrounding_element_opt doclst: Similar to map_list, the function f is applied to all attribute values and data strings. Unlike map_list, more information is passed to the callback function f. This function is called with an xmap_value argument:

  • Xmap_attribute(ename,aname,aval): The function is called for an attribute value of element ename. The attribute is aname and has the value aval. The function must return the new value of the attribute (i.e. aval').
  • Xmap_data(ename_opt,data): The function is called for a data node surrounded by an element ename_opt (which is None if the data node is the outermost node). The string data is the value of the data node. The function must return the new value of the data node (i.e. data').

xmap_list is invoked with surrounding_element_opt which is the name of the surrounding element, or None if such an element does not exist, or is unknown.

#
val write : ?dtd:simplified_dtd -> ?xhtml:bool -> Netchannels.out_obj_channel -> document list -> unit

Writes the document to the output channel. No additional encoding or decoding happens.

Empty elements are written without end tag (see also optional argument xhtml); the rest is written unabbreviated.

Example: To write the document to a file:

     let f = open_out "filename" in
     let ch = new Netchannels.output_channel f in
     write ch doc;
     ch # close_out()
dtd The assumed simplified DTD, by default html40_dtd
xhtml makes the output compatible with XHTML 1.0 Strict by closing `Empty tags with "/>" (true by default).
end