Files
StreamDevice/doc/formatconverter.html
2014-05-02 15:07:06 +00:00

222 lines
7.6 KiB
HTML

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"
"http://www.w3.org/TR/html4/loose.dtd">
<html>
<head>
<title>StreamDevice: Format Converter API</title>
<link rel="shortcut icon" href="favicon.ico">
<link rel="stylesheet" type="text/css" href="stream.css">
<meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1">
<meta name="author" content="Dirk Zimoch">
</head>
<body>
<iframe src="nav.html" id="navleft"></iframe>
<h1>Format Converter API</h1>
<a name="class"></a>
<h2>Converter Class</h2>
<p>
A user defined converter class inherits public from
<i>StreamFormatConverter</i> and handles one or more conversion characters.
It is not necessary that a given conversion character supports both,
printing and scanning.
But if it does, both must be handled by the same converter class.
</p>
<p>
Any conversion corresponds to one data type. The converter class must
implement print and/or scan methods for this data type. It must also
implement a parse method to analyse the format string.
</p>
<p>
A converter class must be registered with a call to RegisterConverter()
in the global file context.
</p>
<p>
The converter must not contain any class variables, because there will be
only one global instance for each conversion character - not one for each
format string!
</p>
<h3>Example: LONG converter for <code>%Q</code></h3>
<pre>
#include "StreamFormatConverter.h"
class MyConverter : public StreamFormatConverter
{
int parse(const StreamFormat&, StreamBuffer&, const char*&, bool);
bool printLong(const StreamFormat&, StreamBuffer&, long);
int scanLong(const StreamFormat&, const char*, long&);
};
RegisterConverter(MyConverter,"Q");
// ... (implementation)
</pre>
<a name="theory"></a>
<h2>Theroy of Operation</h2>
<a name="registration"></a>
<h3>Registration </h3>
<div class="indent"><code>
RegisterConverter(<i>converterClass</i>, "<i>characters</i>");
</code></div>
<p>
This macro registers the converter class for all given conversion
characters. In most cases, you will give only one character.
The macro must be called once for each class in the global file context.
</p>
<p>
HINT: Do not branch depending on the conversion character.
Provide multiple classes, that's more efficient.
</p>
<a name="parsing"></a>
<h3>Parsing</h3>
<div class="indent"><code>
int parse (const&nbsp;StreamFormat&&nbsp;fmt, StreamBuffer&&nbsp;info,
const&nbsp;char*&&nbsp;source, bool&nbsp;scanFormat);
</code></div>
<div class="indent"><code>
struct&nbsp;StreamFormat {
char&nbsp;conv;
StreamFormatType&nbsp;type;
unsigned&nbsp;char&nbsp;flags;
short&nbsp;prec;
unsigned&nbsp;short&nbsp;width;
unsigned&nbsp;short&nbsp;infolen;
const&nbsp;char*&nbsp;info;
};
</code></div>
<p>
During initialization, <code>parse()</code> is called whenever one of the
conversion characters handled by your converter class is found in a protocol.
The fields <code>fmt.conv</code>, <code>fmt.flags</code>,
<code>fmt.prec</code>, and <code>fmt.width</code> have
already been filled in. If a scan format is parsed, <code>scanFormat</code>
is <code>true</code>. If a print format is parsed, <code>scanFormat</code>
is <code>false</code>.
</p>
<p>
The <code>fmt.flags</code> field is a bitset and can have any of the following
flags set:
</p>
<ul>
<li><code>left_flag</code>: the format contained a <code>-</code>.
This is normaly used to indicate that the value should be printed
left-aligned.
</li>
<li><code>sign_flag</code>: the format contained a <code>+</code>.
This normaly requests to print a sign even for positive numbers.
</li>
<li><code>space_flag</code>: the format contained a '<code>&nbsp;</code>' (space).
This normaly requests to print a space instead of a sign for positive numbers.
</li>
<li><code>alt_flag</code>: the format contained a <code>#</code>.
This indicated the request to use an alternative format. For example in
<code>%#x</code> the hex number is preficed with <code>0x</code>.
</li>
<li><code>zero_flag</code>: the format contained a <code>0</code>.
This normaly requests to pad a numerical value with leading zeros
instead of leading spaces.
<li><code>skip_flag</code>: the format contained a <code>*</code>.
The value is parsed and checked but then discarded.
</ul>
<p>
It is not necessary that these flags have exactly the same meaning in your
formats, but a similar and intuitive meaning helpful for the user.
</p>
<p>
There are two additional flags, <code>default_flag</code> indicating a
<code>?</code> and <code>compare_flag</code> indicating a <code>=</code>
int the format, that are handled internally by <em>StreamDevice</em> and
are not of interest to the converter class.
</p>
<p>
The <code>source</code> pointer points to the character of the format string
just after the conversion character. You can parse additional characters if
they belong to the format string handled by your class.
Move the <code>source</code> pointer so that is points to the first character
after your format string.
This is done for example in the builtin formats
<code>%[charset]</code> or <code>%{enum0|enum1}</code>.
However, many formats don't need additional characters.
</p>
<h4>Example</h4>
<pre>
source source
before after
parse() parse()
| |
"%39[0-9a-zA-Z]constant text"
|
conversion
character
</pre>
<p>
You can write any data you may need later in <code>print*()</code> or
<code>scan*()</code> to the
<a href="streambuffer.html"><i>Streambuffer</i></a> <code>info</code>.
This will probably be necessary if you have parsed additional characters
from the format string as in the above example<br>
</p>
<p>
Return <code>long_format</code>, <em>double_format</em>,
<em>string_format</em>, or <code>enum_format</code> depending on the
datatype associated with the conversion character.
It is not necessary to return the same value for print and for scan
formats.
You can even return different values depending on the format string,
but I can't imagine why anyone should do that.
</p>
<p>
If the format is not a real data conversion but does other things with
the data (append or check a checksum, encode or decode the data,...),
return <code>pseudo_format</code>.
</p>
<p>
Return <code>false</code> if there is any parse error or if print or scan
is requested but not supported by this conversion.
</p>
<a name="printing_scanning"></a>
<h3>Printing and Scanning</h3>
<p>
Provide a <code>print[Long|Double|String|Pseudo]()</code> and/or
<code>scan[Long|Double|String|Pseudo]()</code> method appropriate
for the data type you have returned in the <code>parse()</code> method.
That method is called whenever the conversion appears in an output or input,
respectively.
You only need to implement the flavour of print and/or scan suitable for
the datatype returned by <code>parse()</code>.
</p>
<p>
Now, <code>fmt.type</code> contains the value returned by <code>parse()</code>.
With <code>fmt.info()</code> get access to the string you have written to
<code>info</code> in <code>parse()</code> (null terminated).
</p>
<p>
The length of the info string can be found in <code>fmt.infolen</code>.
</p>
<p>
In <code>print*()</code>, append the converted value to <code>output</code>.
Do not modify what is already in output (unless you really know what you're
doing).
Return <code>true</code> on success, <code>false</code> on failure.
</p>
<p>
In <code>scan*()</code>, read the value from input and return the number of
consumed bytes.
In the string version, don't write more bytes than <code>maxlen</code>!
If the <code>skip_flag</code> is set, you don't need to write to
<code>value</code>, since the value will be discarded anyway.
Return <code>-1</code> on failure.
</p>
<hr>
<p><small>Dirk Zimoch, 2007</small></p>
</body>
</html>