ReadXmlExample
Cet exemple lit xml nœud par nœud.
ReadXml.cpp
//------------------------------------------------------------
// Copyright (C) Microsoft. All rights reserved.
//------------------------------------------------------------
#ifndef UNICODE
#define UNICODE
#endif
#include <windows.h>
#include <stdio.h>
#include "WebServices.h"
#include "process.h"
#include "string.h"
#pragma comment(lib, "WebServices.lib")
#pragma comment(lib, "Rpcrt4.lib")
// Print out rich error info
void PrintError(HRESULT errorCode, WS_ERROR* error)
{
wprintf(L"Failure: errorCode=0x%lx\n", errorCode);
if (errorCode == E_INVALIDARG || errorCode == WS_E_INVALID_OPERATION)
{
// Correct use of the APIs should never generate these errors
wprintf(L"The error was due to an invalid use of an API. This is likely due to a bug in the program.\n");
DebugBreak();
}
HRESULT hr = NOERROR;
if (error != NULL)
{
ULONG errorCount;
hr = WsGetErrorProperty(error, WS_ERROR_PROPERTY_STRING_COUNT, &errorCount, sizeof(errorCount));
if (FAILED(hr))
{
goto Exit;
}
for (ULONG i = 0; i < errorCount; i++)
{
WS_STRING string;
hr = WsGetErrorString(error, i, &string);
if (FAILED(hr))
{
goto Exit;
}
wprintf(L"%.*s\n", string.length, string.chars);
}
}
Exit:
if (FAILED(hr))
{
wprintf(L"Could not get error string (errorCode=0x%lx)\n", hr);
}
}
void PrintString(const WS_XML_STRING* string)
{
if (string->length == 0)
{
return;
}
WCHAR* chars = NULL;
int charCount = 0;
charCount =
MultiByteToWideChar(
CP_UTF8,
MB_ERR_INVALID_CHARS,
(char*)string->bytes,
string->length, NULL, 0);
if (charCount == 0)
{
printf("<error: MultiByteToWideChar failed>");
goto Exit;
}
chars = (WCHAR*)HeapAlloc(
GetProcessHeap(),
0,
charCount * sizeof(WCHAR));
if (chars == NULL)
{
printf("<error: HeapAlloc failed>");
goto Exit;
}
if (MultiByteToWideChar(
CP_UTF8,
MB_ERR_INVALID_CHARS,
(char*)string->bytes, string->length,
chars, charCount) != charCount)
{
printf("<error: MultiByteToWideChar failed>");
goto Exit;
}
printf("%.*S", charCount, chars);
Exit:
if (chars != NULL)
{
HeapFree(GetProcessHeap(), 0, chars);
}
}
void PrintText(const WS_XML_TEXT* text)
{
switch (text->textType)
{
case WS_XML_TEXT_TYPE_UTF8:
{
const WS_XML_UTF8_TEXT* utf8Text = (const WS_XML_UTF8_TEXT*) text;
printf("WS_XML_TEXT_TYPE_UTF8(value='");
PrintString(&utf8Text->value);
printf("')");
}
break;
case WS_XML_TEXT_TYPE_UTF16:
{
const WS_XML_UTF16_TEXT* utf16Text = (const WS_XML_UTF16_TEXT*) text;
printf("WS_XML_TEXT_TYPE_UTF16(value='%.*S')", utf16Text->byteCount / sizeof(WCHAR), (WCHAR*)utf16Text->bytes);
}
break;
case WS_XML_TEXT_TYPE_BASE64:
{
const WS_XML_BASE64_TEXT* base64Text = (const WS_XML_BASE64_TEXT*) text;
printf("WS_XML_TEXT_TYPE_BASE64(value={ ");
for (ULONG i = 0; i < base64Text->length; i++)
{
printf("%02x", base64Text->bytes[i]);
}
printf(" })");
}
break;
case WS_XML_TEXT_TYPE_BOOL:
{
const WS_XML_BOOL_TEXT* boolText = (const WS_XML_BOOL_TEXT*) text;
printf("WS_XML_TEXT_TYPE_BOOL(value='%s')", boolText->value ? "true" : "false");
}
break;
case WS_XML_TEXT_TYPE_INT32:
{
const WS_XML_INT32_TEXT* int32Text = (const WS_XML_INT32_TEXT*) text;
printf("WS_XML_TEXT_TYPE_INT32(value='%d')", int32Text->value);
}
break;
case WS_XML_TEXT_TYPE_INT64:
{
const WS_XML_INT64_TEXT* int64Text = (const WS_XML_INT64_TEXT*) text;
printf("WS_XML_TEXT_TYPE_INT64(value='%I64d')", int64Text->value);
}
break;
case WS_XML_TEXT_TYPE_UINT64:
{
const WS_XML_UINT64_TEXT* uint64Text = (const WS_XML_UINT64_TEXT*) text;
printf("WS_XML_TEXT_TYPE_UINT64(value='%I64u')", uint64Text->value);
}
break;
case WS_XML_TEXT_TYPE_FLOAT:
{
const WS_XML_FLOAT_TEXT* floatText = (const WS_XML_FLOAT_TEXT*) text;
printf("WS_XML_TEXT_TYPE_FLOAT(value='%f')", floatText->value);
}
break;
case WS_XML_TEXT_TYPE_DOUBLE:
{
const WS_XML_DOUBLE_TEXT* doubleText = (const WS_XML_DOUBLE_TEXT*) text;
printf("WS_XML_TEXT_TYPE_DOUBLE(value='%f')", doubleText->value);
}
break;
case WS_XML_TEXT_TYPE_DECIMAL:
{
const WS_XML_DECIMAL_TEXT* decimalText = (const WS_XML_DECIMAL_TEXT*) text;
const DECIMAL* dec = &decimalText->value;
printf("WS_XML_TEXT_TYPE_DECIMAL(value={%x %x %x, %I64x})", dec->wReserved, dec->signscale, dec->Hi32, dec->Lo64);
}
break;
case WS_XML_TEXT_TYPE_GUID:
{
WS_XML_GUID_TEXT* guidText = (WS_XML_GUID_TEXT*) text;
RPC_WSTR s;
if (UuidToString(&guidText->value, &s) == RPC_S_OK)
{
printf("WS_XML_TEXT_TYPE_GUID(value='%ls')", s);
RpcStringFree(&s);
}
else
{
printf("WS_XML_TEXT_TYPE_GUID(value='<error: UuidToString failed>')");
}
}
break;
case WS_XML_TEXT_TYPE_UNIQUE_ID:
{
WS_XML_UNIQUE_ID_TEXT* uniqueIdText = (WS_XML_UNIQUE_ID_TEXT*) text;
RPC_WSTR s;
if (UuidToString(&uniqueIdText->value, &s) == RPC_S_OK)
{
printf("WS_XML_TEXT_TYPE_UNIQUEID(value='%ls')", s);
RpcStringFree(&s);
}
else
{
printf("WS_XML_TEXT_TYPE_UNIQUEID(value='<error: UuidToString failed>')");
}
}
break;
case WS_XML_TEXT_TYPE_DATETIME:
{
const WS_XML_DATETIME_TEXT* dateTimeText = (const WS_XML_DATETIME_TEXT*)text;
WS_DATETIME value = dateTimeText->value;
printf("WS_XML_DATETIME_TEXT(ticks='%I64u',format='%d')", value.ticks, value.format);
}
break;
case WS_XML_TEXT_TYPE_TIMESPAN:
{
const WS_XML_TIMESPAN_TEXT* timeSpanText = (const WS_XML_TIMESPAN_TEXT*)text;
printf("WS_XML_TIMESPAN_TEXT(value='%I64u')", timeSpanText->value.ticks);
}
break;
case WS_XML_TEXT_TYPE_QNAME:
{
const WS_XML_QNAME_TEXT* qnameText = (const WS_XML_QNAME_TEXT*)text;
printf("WS_XML_QNAME_TEXT(prefix='");
PrintString(qnameText->prefix);
printf("', localName='");
PrintString(qnameText->localName);
printf("', ns='");
PrintString(qnameText->ns);
printf("')");
}
break;
case WS_XML_TEXT_TYPE_LIST:
{
const WS_XML_LIST_TEXT* listText = (const WS_XML_LIST_TEXT*)text;
printf("WS_XML_LIST_TEXT(value={");
for (ULONG i = 0; i < listText->itemCount; i++)
{
PrintText(listText->items[i]);
printf(" ");
}
printf("}");
}
break;
default:
printf("<error: Unexpected textType>");
break;
}
}
// Main entry point
int __cdecl wmain(int argc, __in_ecount(argc) wchar_t **argv)
{
UNREFERENCED_PARAMETER(argc);
UNREFERENCED_PARAMETER(argv);
HRESULT hr = NOERROR;
WS_ERROR* error = NULL;
WS_XML_READER* xmlReader = NULL;
// Create an error object for storing rich error information
hr = WsCreateError(
NULL,
0,
&error);
if (FAILED(hr))
{
goto Exit;
}
// Create an XML reader
hr = WsCreateReader(
NULL,
0,
&xmlReader,
error);
if (FAILED(hr))
{
goto Exit;
}
char* xml =
"<?xml version='1.0' encoding='UTF-8' standalone='yes'?>"
"<Orders xmlns='http://example.com'>"
"<!-- Order #1 -->"
"<PurchaseOrder id='1'>"
"<Quantity>42</Quantity>"
"<ProductName>Toaster</ProductName>"
"</PurchaseOrder>"
"<!-- Order #2 -->"
"<PurchaseOrder id='2'>"
"<Quantity>5</Quantity>"
"<ProductName><![CDATA[Block&Tackle]></ProductName>"
"</PurchaseOrder>"
"</Orders>";
BYTE* bytes = (BYTE*) xml;
ULONG byteCount = (ULONG)strlen(xml);
// Setup the source input
WS_XML_READER_BUFFER_INPUT bufferInput;
ZeroMemory(&bufferInput, sizeof(bufferInput));
bufferInput.input.inputType = WS_XML_READER_INPUT_TYPE_BUFFER;
bufferInput.encodedData = bytes;
bufferInput.encodedDataSize = byteCount;
// Setup the source encoding
WS_XML_READER_TEXT_ENCODING textEncoding;
ZeroMemory(&textEncoding, sizeof(textEncoding));
textEncoding.encoding.encodingType = WS_XML_READER_ENCODING_TYPE_TEXT;
textEncoding.charSet = WS_CHARSET_AUTO;
// Setup the reader
hr = WsSetInput(xmlReader, &textEncoding.encoding, &bufferInput.input, NULL, 0, error);
if (FAILED(hr))
{
goto Exit;
}
WS_CHARSET charSet;
hr = WsGetReaderProperty(xmlReader, WS_XML_READER_PROPERTY_CHARSET, &charSet, sizeof(charSet), error);
if (FAILED(hr))
{
goto Exit;
}
switch (charSet)
{
case WS_CHARSET_UTF8:
printf("charSet=utf-8\n");
break;
case WS_CHARSET_UTF16LE:
printf("charSet=utf-16LE\n");
break;
case WS_CHARSET_UTF16BE:
printf("charSet=utf-16BE\n");
break;
default:
printf("<error: Unexpected charset>\n");
hr = E_FAIL;
goto Exit;
}
ULONG depth = 0;
for (;;)
{
// Get the current node of the reader
const WS_XML_NODE* node;
hr = WsGetReaderNode(xmlReader, &node, error);
if (FAILED(hr))
{
goto Exit;
}
for (ULONG i = 0; i < depth; i++)
{
printf(" ");
}
// Print out the type of node and its contents
switch (node->nodeType)
{
case WS_XML_NODE_TYPE_ELEMENT:
{
const WS_XML_ELEMENT_NODE* elementNode = (const WS_XML_ELEMENT_NODE*) node;
printf("WS_XML_NODE_TYPE_ELEMENT(prefix='");
PrintString(elementNode->prefix);
printf("', localName='");
PrintString(elementNode->localName);
printf("', ns='");
PrintString(elementNode->ns);
printf("', attributes={");
for (ULONG i = 0; i < elementNode->attributeCount; i++)
{
if (i != 0)
{
printf(", ");
}
const WS_XML_ATTRIBUTE* attribute = elementNode->attributes[i];
if (attribute->isXmlNs)
{
printf("(isXmlNs=TRUE, prefix='");
PrintString(attribute->prefix);
printf("', ns='");
PrintString(attribute->ns);
printf("')");
}
else
{
printf("(isXmlNs=FALSE, prefix='");
PrintString(attribute->prefix);
printf("', localName='");
PrintString(attribute->localName);
printf("', ns='");
PrintString(attribute->ns);
printf("', value=");
PrintText(attribute->value);
printf(")");
}
}
printf("})\n");
}
depth++;
break;
case WS_XML_NODE_TYPE_END_ELEMENT:
printf("WS_XML_NODE_TYPE_END_ELEMENT\n");
depth--;
break;
case WS_XML_NODE_TYPE_TEXT:
{
const WS_XML_TEXT_NODE* textNode = (const WS_XML_TEXT_NODE*) node;
printf("WS_XML_NODE_TYPE_TEXT(value=");
PrintText(textNode->text);
printf(")\n");
}
break;
case WS_XML_NODE_TYPE_CDATA:
printf("WS_XML_NODE_TYPE_CDATA\n");
depth++;
break;
case WS_XML_NODE_TYPE_END_CDATA:
printf("WS_XML_NODE_TYPE_END_CDATA\n");
depth--;
break;
case WS_XML_NODE_TYPE_COMMENT:
{
const WS_XML_COMMENT_NODE* commentNode = (const WS_XML_COMMENT_NODE*) node;
printf("WS_XML_NODE_TYPE_COMMENT(value='");
PrintString(&commentNode->value);
printf("')\n");
}
break;
case WS_XML_NODE_TYPE_BOF:
printf("WS_XML_NODE_TYPE_BOF\n");
depth++;
break;
case WS_XML_NODE_TYPE_EOF:
printf("WS_XML_NODE_TYPE_EOF\n");
depth--;
break;
default:
printf("<error: Unexpected nodeType>\n");
hr = E_FAIL;
goto Exit;
}
// See if we've reached the end of the document
if (node->nodeType == WS_XML_NODE_TYPE_EOF)
{
break;
}
// Advance the reader
hr = WsReadNode(xmlReader, error);
if (FAILED(hr))
{
goto Exit;
}
}
Exit:
if (FAILED(hr))
{
// Print out the error
PrintError(hr, error);
}
fflush(
stdout);
if (xmlReader != NULL)
{
WsFreeReader(xmlReader);
}
if (error != NULL)
{
WsFreeError(error);
}
fflush(stdout);
return SUCCEEDED(hr) ? 0 : -1;
}