Performance, Implementation, and Design Notes

The following notes are informative, not normative. Despite the appearance of words such as "must" and "should", all requirements in this section appear elsewhere in the specification.

B.1 Notes on invalid documents

This specification does not define how conforming user agents handle general error conditions, including how user agents behave when they encounter elements, attributes, attribute values, or entities not specified in this document.

However, to facilitate experimentation and interoperability between implementations of various versions of HTML, we recommend the following behavior:

We also recommend that user agents provide support for notifying the user of such errors.

Since user agents may vary in how they handle error conditions, authors and users must not rely on specific error recovery behavior.

The HTML 2.0 specification ([RFC1866]) observes that many HTML 2.0 user agents assume that a document that does not begin with a document type declaration refers to the HTML 2.0 specification. As experience shows that this is a poor assumption, the current specification does not recommend this behavior.

For reasons of interoperability, authors must not "extend" HTML through the available SGML mechanisms (e.g., extending the DTD, adding a new set of entity definitions, etc.).

B.2 Special characters in URI attribute values

B.2.1 Non-ASCII characters in URI attribute values

Although URIs do not contain non-ASCII values (see [URI], section 2.1) authors sometimes specify them in attribute values expecting URIs (i.e., defined with %URI; in the DTD). For instance, the following href value is illegal:

We recommend that user agents adopt the following convention for handling non-ASCII characters in such cases:

This procedure results in a syntactically legal URI (as defined in [RFC1738], section 2.2 or [RFC2141], section 2) that is independent of the character encoding to which the HTML document carrying the URI may have been transcoded.

Note. Some older user agents trivially process URIs in HTML using the bytes of the character encoding in which the document was received. Some older HTML documents rely on this practice and break when transcoded. User agents that want to handle these older documents should, on receiving a URI containing characters outside the legal set, first use the conversion based on UTF-8. Only if the resulting URI does not resolve should they try constructing a URI based on the bytes of the character encoding in which the document was received.

B.2.2 Ampersands in URI attribute values

The URI that is constructed when a form is submitted may be used as an anchor-style link (e.g., the href attribute for the A element). Unfortunately, the use of the "&" character to separate form fields interacts with its use in SGML attribute values to delimit character entity references. For example, to use the URI "http://host/?x=1&y=2" as a linking URI, it must be written <A href="http://host/?x=1&y=2"> or <A href="http://host/?x=1&y=2">.

We recommend that HTTP server implementors, and in particular, CGI implementors support the use of ";" in place of "&" to save authors the trouble of escaping "&" characters in this manner.

B.3 SGML implementation notes

B.3.1 Line breaks

SGML (see [ISO8879], section 7.6.1) specifies that a line break immediately following a start tag must be ignored, as must a line break immediately before an end tag. This applies to all HTML elements without exception.

B.3.2 Specifying non-HTML data

Script and style data may appear as element content or attribute values. The following sections describe the boundary between HTML markup and foreign data.

Note. The DTD defines script and style data to be CDATA for both element content and attribute values. SGML rules do not allow character references in CDATA element content but do allow them in CDATA attribute values. Authors should pay particular attention when cutting and pasting script and style data between element content and attribute values.

This asymmetry also means that when transcoding from a richer to a poorer character encoding, the transcoder cannot simply replace unconvertible characters in script or style data with the corresponding numeric character references; it must parse the HTML document and know about each script and style language's syntax in order to process the data correctly.

Element content

When script or style data is the content of an element (SCRIPT and STYLE), the data begins immediately after the element start tag and ends at the first ETAGO ("</") delimiter followed by a name start character ([a-zA-Z]); note that this may not be the element's end tag. Authors should therefore escape "</" within the content. Escape mechanisms are specific to each scripting or style sheet language.

In JavaScript, this code can be expressed legally by hiding the ETAGO delimiter before an SGML name start character:

    <SCRIPT type="text/javascript">
      document.write ("<EM>This will work<\/EM>")
    </SCRIPT>

In Tcl, one may accomplish this as follows:

    <SCRIPT type="text/tcl">
      document write "<EM>This will work<\/EM>"
    </SCRIPT>

In VBScript, the problem may be avoided with the Chr() function:

    "<EM>This will work<" & Chr(47) & "EM>"

Attribute values

When script or style data is the value of an attribute (either style or the intrinsic event attributes), authors should escape occurrences of the delimiting single or double quotation mark within the value according to the script or style language convention. Authors should also escape occurrences of "&" if the "&" is not meant to be the beginning of a character reference.

B.3.3 SGML features with limited support

SGML systems conforming to [ISO8879] are expected to recognize a number of features that aren't widely supported by HTML user agents. We recommend that authors avoid using all of these features.

B.3.4 Boolean attributes

Authors should be aware that many user agents only recognize the minimized form of boolean attributes and not the full form.

B.3.5 Marked Sections

Marked sections play a role similar to the #ifdef construct recognized by C preprocessors.

SGML also defines the use of marked sections for CDATA content, within which "<" is not treated as the start of a tag, e.g.,

The telltale sign that a user agent doesn't recognize a marked section is the appearance of "]]>", which is seen when the user agent mistakenly uses the first ">" character as the end of the tag starting with "<![".

B.3.6 Processing Instructions

Processing instructions are a mechanism to capture platform-specific idioms. A processing instruction begins with <? and ends with >

Authors should be aware that many user agents render processing instructions as part of the document's text.

B.3.7 Shorthand markup

Some SGML SHORTTAG constructs save typing but add no expressive capability to the SGML application. Although these constructs technically introduce no ambiguity, they reduce the robustness of documents, especially when the language is enhanced to include new elements. Thus, while SHORTTAG constructs of SGML related to attributes are widely used and implemented, those related to elements are not. Documents that use them are conforming SGML documents, but are unlikely to work with many existing HTML tools.

NET tags:
```
<name/.../
```
closed Start Tag:
```
<name1<name2>
```
Empty Start Tag:
```
<>
```
Empty End Tag:
```
</>
```

B.4 Notes on helping search engines index your Web site

This section provides some simple suggestions that will make your documents more accessible to search engines.

B.4.1 Search robots

The robots.txt file

When a Robot visits a Web site, say http://www.foobar.com/, it firsts checks for http://www.foobar.com/robots.txt. If it can find this document, it will analyze its contents to see if it is allowed to retrieve the document. You can customize the robots.txt file to apply only to specific robots, and to disallow access to specific directories or files.

Here is a sample robots.txt file that prevents all robots from visiting the entire site

The Robot will simply look for a "/robots.txt" URI on your site, where a site is defined as a HTTP server running on a particular host and port number. Here are some sample locations for robots.txt:

There can only be a single "/robots.txt" on a site. Specifically, you should not put "robots.txt" files in user directories, because a robot will never look at them. If you want your users to be able to create their own "robots.txt", you will need to merge them all into a single "/robots.txt". If you don't want to do this your users might want to use the Robots META Tag instead.

Some tips: URI's are case-sensitive, and "/robots.txt" string must be all lower-case. Blank lines are not permitted within a single record in the "robots.txt" file.

There must be exactly one "User-agent" field per record. The robot should be liberal in interpreting this field. A case-insensitive substring match of the name without version information is recommended.

If the value is "*", the record describes the default access policy for any robot that has not matched any of the other records. It is not allowed to have multiple such records in the "/robots.txt" file.

The "Disallow" field specifies a partial URI that is not to be visited. This can be a full path, or a partial path; any URI that starts with this value will not be retrieved. For example,

An empty value for "Disallow", indicates that all URIs can be retrieved. At least one "Disallow" field must be present in the robots.txt file.

Site URI	URI for `robots.txt`
./	./robots.txt
.:80/	.:80/robots.txt
.:1234/	.:1234/robots.txt
http://w3.org/	http://w3.org/robots.txt

Robots and the META element

The META element allows HTML authors to tell visiting robots whether a document may be indexed, or used to harvest more links. No server administrator action is required.

B.5 Notes on tables

B.5.1 Design rationale

The HTML table model has evolved from studies of existing SGML tables models, the treatment of tables in common word processing packages, and a wide range of tabular layout techniques in magazines, books and other paper-based documents. The model was chosen to allow simple tables to be expressed simply with extra complexity available when needed. This makes it practical to create the markup for HTML tables with everyday text editors and reduces the learning curve for getting started. This feature has been very important to the success of HTML to date.

Increasingly, people are creating tables by converting from other document formats or by creating them directly with WYSIWYG editors. It is important that the HTML table model fit well with these authoring tools. This affects how the cells that span multiple rows or columns are represented, and how alignment and other presentation properties are associated with groups of cells.

Dynamic reformatting

A major consideration for the HTML table model is that the author does not control how a user will size a table, what fonts he or she will use, etc. This makes it risky to rely on column widths specified in terms of absolute pixel units. Instead, tables must be able to change sizes dynamically to match the current window size and fonts. Authors can provide guidance as to the relative widths of columns, but user agents should ensure that columns are wide enough to render the width of the largest element of the cell's content. If the author's specification must be overridden, relative widths of individual columns should not be changed drastically.

Incremental display

For large tables or slow network connections, incremental table display is important to user satisfaction. User agents should be able to begin displaying a table before all of the data has been received. The default window width for most user agents shows about 80 characters, and the graphics for many HTML pages are designed with these defaults in mind. By specifying the number of columns, and including provision for control of table width and the widths of different columns, authors can give hints to user agents that allow the incremental display of table contents.

For incremental display, the browser needs the number of columns and their widths. The default width of the table is the current window size (width="100%"). This can be altered by setting the width attribute of the TABLE element. By default, all columns have the same width, but you can specify column widths with one or more COL elements before the table data starts.

The remaining issue is the number of columns. Some people have suggested waiting until the first row of the table has been received, but this could take a long time if the cells have a lot of content. On the whole it makes more sense, when incremental display is desired, to get authors to explicitly specify the number of columns in the TABLE element.

Authors still need a way of telling user agents whether to use incremental display or to size the table automatically to fit the cell contents. In the two pass auto-sizing mode, the number of columns is determined by the first pass. In the incremental mode, the number of columns must be stated up front (with COL or COLGROUP elements).

Structure and presentation

HTML distinguishes structural markup such as paragraphs and quotations from rendering idioms such as margins, fonts, colors, etc. How does this distinction affect tables? From the purist's point of view, the alignment of text within table cells and the borders between cells is a rendering issue, not one of structure. In practice, though, it is useful to group these with the structural information, as these features are highly portable from one application to the next. The HTML table model leaves most rendering information to associated style sheets. The model presented in this specification is designed to take advantage of such style sheets but not to require them.

Current desktop publishing packages provide very rich control over the rendering of tables, and it would be impractical to reproduce this in HTML, without making HTML into a bulky rich text format like RTF or MIF. This specification does, however, offer authors the ability to choose from a set of commonly used classes of border styles. The frame attribute controls the appearance of the border frame around the table while the rules attribute determines the choice of rulings within the table. A finer level of control will be supported via rendering annotations. The style attribute can be used for specifying rendering information for individual elements. Further rendering information can be given with the STYLE element in the document head or via linked style sheets.

During the development of this specification, a number of avenues were investigated for specifying the ruling patterns for tables. One issue concerns the kinds of statements that can be made. Including support for edge subtraction as well as edge addition leads to relatively complex algorithms. For instance, work on allowing the full set of table elements to include the frame and rules attributes led to an algorithm involving some 24 steps to determine whether a particular edge of a cell should be ruled or not. Even this additional complexity doesn't provide enough rendering control to meet the full range of needs for tables. The current specification deliberately sticks to a simple intuitive model, sufficient for most purposes. Further experimental work is needed before a more complex approach is standardized.

Row and column groups

This specification provides a superset of the simpler model presented in earlier work on HTML+. Tables are considered as being formed from an optional caption together with a sequence of rows, which in turn consist of a sequence of table cells. The model further differentiates header and data cells, and allows cells to span multiple rows and columns.

Following the CALS table model (see [CALS]), this specification allows table rows to be grouped into head and body and foot sections. This simplifies the representation of rendering information and can be used to repeat table head and foot rows when breaking tables across page boundaries, or to provide fixed headers above a scrollable body panel. In the markup, the foot section is placed before the body sections. This is an optimization shared with CALS for dealing with very long tables. It allows the foot to be rendered without having to wait for the entire table to be processed.

Accessibility

For the visually impaired, HTML offers the hope of setting to rights the damage caused by the adoption of windows based graphical user interfaces. The HTML table model includes attributes for labeling each cell, to support high quality text to speech conversion. The same attributes can also be used to support automated import and export of table data to databases or spreadsheets.

B.5.2 Recommended Layout Algorithms

If COL or COLGROUP elements are present, they specify the number of columns and the table may be rendered using a fixed layout. Otherwise the autolayout algorithm described below should be used.

If the width attribute is not specified, visual user agents should assume a default value of 100% for formatting.

It is recommended that user agents increase table widths beyond the value specified by width in cases when cell contents would otherwise overflow. User agents that override the specified width should do so within reason. User agents may elect to split words across lines to avoid the need for excessive horizontal scrolling or when such scrolling is impractical or undesired.

For the purposes of layout, user agents should consider that table captions (specified by the CAPTION element) behave like cells. Each caption is a cell that spans all of the table's columns if at the top or bottom of the table, and rows if at the left or right side of the table.

Fixed Layout Algorithm

For this algorithm, it is assumed that the number of columns is known. The column widths by default should be set to the same size. Authors may override this by specifying relative or absolute column widths, using the COLGROUP or COL elements. The default table width is the space between the current left and right margins, but may be overridden by the width attribute on the TABLE element, or determined from absolute column widths. To deal with mixtures of absolute and relative column widths, the first step is to allocate space from the table width to columns with absolute widths. After this, the space remaining is divided up between the columns with relative widths.

The table syntax alone is insufficient to guarantee the consistency of attribute values. For instance, the number of COL and COLGROUP elements may be inconsistent with the number of columns implied by the table cells. A further problem occurs when the columns are too narrow to avoid overflow of cell contents. The width of the table as specified by the TABLE element or COL elements may result in overflow of cell contents. It is recommended that user agents attempt to recover gracefully from these situations, e.g., by hyphenating words and resorting to splitting words if hyphenation points are unknown.

In the event that an indivisible element causes cell overflow, the user agent may consider adjusting column widths and re-rendering the table. In the worst case, clipping may be considered if column width adjustments and/or scrollable cell content are not feasible. In any case, if cell content is split or clipped this should be indicated to the user in an appropriate manner.

Autolayout Algorithm

If the number of columns is not specified with COL and COLGROUP elements, then the user agent should use the following autolayout algorithm. It uses two passes through the table data and scales linearly with the size of the table.

In the first pass, line wrapping is disabled, and the user agent keeps track of the minimum and maximum width of each cell. The maximum width is given by the widest line. Since line wrap has been disabled, paragraphs are treated as long lines unless broken by BR elements. The minimum width is given by the widest text element (word, image, etc.) taking into account leading indents and list bullets, etc. In other words, it is necessary to determine the minimum width a cell would require in a window of its own before the cell begins to overflow. Allowing user agents to split words will minimize the need for horizontal scrolling or in the worst case, clipping the cell contents.

This process also applies to any nested tables occurring in cell content. The minimum and maximum widths for cells in nested tables are used to determine the minimum and maximum widths for these tables and hence for the parent table cell itself. The algorithm is linear with aggregate cell content, and broadly speaking, independent of the depth of nesting.

To cope with character alignment of cell contents, the algorithm keeps three running min/max totals for each column: Left of align char, right of align char and unaligned. The minimum width for a column is then: max(min_left + min_right, min_non-aligned).

The minimum and maximum cell widths are then used to determine the corresponding minimum and maximum widths for the columns. These in turn, are used to find the minimum and maximum width for the table. Note that cells can contain nested tables, but this doesn't complicate the code significantly. The next step is to assign column widths according to the available space (i.e., the space between the current left and right margins).

For cells that span multiple columns, a simple approach consists of apportioning the min/max widths evenly to each of the constituent columns. A slightly more complex approach is to use the min/max widths of unspanned cells to weight how spanned widths are apportioned. Experiments suggest that a blend of the two approaches gives good results for a wide range of tables.

The table borders and intercell margins need to be included in assigning column widths. There are three cases:

This assignment step is then repeated for nested tables using the minimum and maximum widths derived for all such tables in the first pass. In this case, the width of the parent table cell plays the role of the current window size in the above description. This process is repeated recursively for all nested tables. The topmost table is then rendered using the assigned widths. Nested tables are subsequently rendered as part of the parent table's cell contents.

If the table width is specified with the width attribute, the user agent attempts to set column widths to match. The width attribute is not binding if this results in columns having less than their minimum (i.e., indivisible) widths.

If relative widths are specified with the COL element, the algorithm is modified to increase column widths over the minimum width to meet the relative width constraints. The COL elements should be taken as hints only, so columns shouldn't be set to less than their minimum width. Similarly, columns shouldn't be made so wide that the table stretches well beyond the extent of the window. If a COL element specifies a relative width of zero, the column should always be set to its minimum width.

When using the two pass layout algorithm, the default alignment position in the absence of an explicit or inherited charoff attribute can be determined by choosing the position that would center lines for which the widths before and after the alignment character are at the maximum values for any of the lines in the column for which align="char". For incremental table layout the suggested default is charoff="50%". If several cells in different rows for the same column use character alignment, then by default, all such cells should line up, regardless of which character is used for alignment. Rules for handling objects too large for a column apply when the explicit or implied alignment results in a situation where the data exceeds the assigned width of the column.

Choice of attribute names. It would have been preferable to choose values for the frame attribute consistent with the rules attribute and the values used for alignment. For instance: none, top, bottom, topbot, left, right, leftright, all. Unfortunately, SGML requires enumerated attribute values to be unique for each element, independent of the attribute name. This causes immediate problems for "none", "left", "right" and "all". The values for the frame attribute have been chosen to avoid clashes with the rules, align, and valign attributes. This provides a measure of future proofing, as it is anticipated that the frame and rules attributes will be added to other table elements in future revisions to this specification. An alternative would be to make frame a CDATA attribute. The consensus of the W3C HTML Working Group was that the benefits of being able to use SGML validation tools to check attributes based on enumerated values outweighs the need for consistent names.

B.6 Notes on forms

B.6.1 Incremental display

The incremental display of documents being received from the network gives rise to certain problems with respect to forms. User agents should prevent forms from being submitted until all of the form's elements have been received.

The incremental display of documents raises some issues with respect to tabbing navigation. The heuristic of giving focus to the lowest valued tabindex in the document seems reasonable enough at first glance. However this implies having to wait until all of the document's text is received, since until then, the lowest valued tabindex may still change. If the user hits the tab key before then, it is reasonable for user agents to move the focus to the lowest currently available tabindex.

If forms are associated with client-side scripts, there is further potential for problems. For instance, a script handler for a given field may refer to a field that doesn't yet exist.

B.6.2 Future projects

This specification defines a set of elements and attributes powerful enough to fulfill the general need for producing forms. However there is still room for many possible improvements. For instance the following problems could be addressed in the future:

Another possible extension would be to add the usemap attribute to INPUT for use as client-side image map when "type=image". The AREA element corresponding to the location clicked would contribute the value to be passed to the server. To avoid the need to modify server scripts, it may be appropriate to extend AREA to provide x and y values for use with the INPUT element.

B.7 Notes on scripting

B.7.1 Reserved syntax for future script macros

This specification reserves syntax for the future support of script macros in HTML CDATA attributes. The intention is to allow attributes to be set depending on the properties of objects that appear earlier on the page. The syntax is:

Current Practice for Script Macros

The macro body is made up of one or more statements in the default scripting language (as per intrinsic event attributes). The semicolon following the right brace is always needed, as otherwise the right brace character "}" is treated as being part of the macro body. Its also worth noting that quote marks are always needed for attributes containing script macros.

Macro processing takes place when the document is loaded (or reloaded) but does not take place again when the document is resized, repainted, etc.

DEPRECATED EXAMPLE:
Here are some examples using JavaScript. The first one randomizes the document background color:

    
<BODY bgcolor='&{randomrgb};'>

Perhaps you want to dim the background for evening viewing:

    
<BODY bgcolor='&{if(Date.getHours > 18)...};'>

The next example uses JavaScript to set the coordinates for a client-side image map:

    
<MAP NAME=foo>
   <AREA shape="rect" coords="&{myrect(imageuri)};" href="&{myuri};" alt="">
 </MAP>

This example sets the size of an image based upon document properties:

    
<IMG src="bar.gif" width='&{document.banner.width/2};' height='50%' alt="banner">

You can set the URI for a link or image by script:

 <SCRIPT type="text/javascript">
   function manufacturer(widget) {
       ...
   }
   function location(manufacturer) {
       ...
   }
   function logo(manufacturer) {
       ...
   }
 </SCRIPT>
  <A href='&{location(manufacturer("widget"))};'>widget</A>
  <IMG src='&{logo(manufacturer("widget"))};' alt="logo">

This last example shows how SGML CDATA attributes can be quoted using single or double quote marks. If you use single quotes around the attribute string then you can include double quote marks as part of the attribute string. Another approach is use " for double quote marks:

   
<IMG src="&{logo(manufacturer(&quot;widget&quot;))};" alt="logo">

B.8 Notes on frames

Since there is no guarantee that a frame target name is unique, it is appropriate to describe the current practice in finding a frame given a target name:

B.9 Notes on accessibility

The W3C Web Accessibility Initiative ([WAI]) is producing a series of guidelines to improve Web accessibility for people with disabilities. There are three sets of guidelines:

B.10 Notes on security

Anchors, embedded images, and all other elements that contain URIs as parameters may cause the URI to be dereferenced in response to user input. In this case, the security issues of [RFC1738], section 6, should be considered. The widely deployed methods for submitting form requests -- HTTP and SMTP -- provide little assurance of confidentiality. Information providers who request sensitive information via forms -- especially with the INPUT element, type="password" -- should be aware and make their users aware of the lack of confidentiality.

B.10.1 Security issues for forms

A user agent should not send any file that the user has not explicitly asked to be sent. Thus, HTML user agents are expected to confirm any default file names that might be suggested by the value attribute of the INPUT element. Hidden controls must not specify files.

This specification does not contain a mechanism for encryption of the data; this should be handled by whatever other mechanisms are in place for secure transmission of data.

Once a file is uploaded, the processing agent should process and store it appropriately.

Appendix B: Performance, Implementation, and Design Notes