blob: 489505f4865a87aaea0a1c8dbdf0488468651a54 [file] [log] [blame]
/*
* xmlsave.c: Implementation of the document serializer
*
* See Copyright for the status of this software.
*
* daniel@veillard.com
*/
#define IN_LIBXML
#include "libxml.h"
#include <string.h>
#include <libxml/xmlmemory.h>
#include <libxml/parserInternals.h>
#include <libxml/tree.h>
#include <libxml/xmlsave.h>
#define MAX_INDENT 60
#include <libxml/HTMLtree.h>
#include "buf.h"
#include "enc.h"
#include "save.h"
/************************************************************************
* *
* XHTML detection *
* *
************************************************************************/
#define XHTML_STRICT_PUBLIC_ID BAD_CAST \
"-//W3C//DTD XHTML 1.0 Strict//EN"
#define XHTML_STRICT_SYSTEM_ID BAD_CAST \
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"
#define XHTML_FRAME_PUBLIC_ID BAD_CAST \
"-//W3C//DTD XHTML 1.0 Frameset//EN"
#define XHTML_FRAME_SYSTEM_ID BAD_CAST \
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-frameset.dtd"
#define XHTML_TRANS_PUBLIC_ID BAD_CAST \
"-//W3C//DTD XHTML 1.0 Transitional//EN"
#define XHTML_TRANS_SYSTEM_ID BAD_CAST \
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"
#define XHTML_NS_NAME BAD_CAST "http://www.w3.org/1999/xhtml"
/**
* xmlIsXHTML:
* @systemID: the system identifier
* @publicID: the public identifier
*
* Try to find if the document correspond to an XHTML DTD
*
* Returns 1 if true, 0 if not and -1 in case of error
*/
int
xmlIsXHTML(const xmlChar *systemID, const xmlChar *publicID) {
if ((systemID == NULL) && (publicID == NULL))
return(-1);
if (publicID != NULL) {
if (xmlStrEqual(publicID, XHTML_STRICT_PUBLIC_ID)) return(1);
if (xmlStrEqual(publicID, XHTML_FRAME_PUBLIC_ID)) return(1);
if (xmlStrEqual(publicID, XHTML_TRANS_PUBLIC_ID)) return(1);
}
if (systemID != NULL) {
if (xmlStrEqual(systemID, XHTML_STRICT_SYSTEM_ID)) return(1);
if (xmlStrEqual(systemID, XHTML_FRAME_SYSTEM_ID)) return(1);
if (xmlStrEqual(systemID, XHTML_TRANS_SYSTEM_ID)) return(1);
}
return(0);
}
#ifdef LIBXML_OUTPUT_ENABLED
#define TODO \
xmlGenericError(xmlGenericErrorContext, \
"Unimplemented block at %s:%d\n", \
__FILE__, __LINE__);
struct _xmlSaveCtxt {
void *_private;
int type;
int fd;
const xmlChar *filename;
const xmlChar *encoding;
xmlCharEncodingHandlerPtr handler;
xmlOutputBufferPtr buf;
int options;
int level;
int format;
char indent[MAX_INDENT + 1]; /* array for indenting output */
int indent_nr;
int indent_size;
xmlCharEncodingOutputFunc escape; /* used for element content */
xmlCharEncodingOutputFunc escapeAttr;/* used for attribute content */
};
/************************************************************************
* *
* Output error handlers *
* *
************************************************************************/
/**
* xmlSaveErrMemory:
* @extra: extra information
*
* Handle an out of memory condition
*/
static void
xmlSaveErrMemory(const char *extra)
{
__xmlSimpleError(XML_FROM_OUTPUT, XML_ERR_NO_MEMORY, NULL, NULL, extra);
}
/**
* xmlSaveErr:
* @code: the error number
* @node: the location of the error.
* @extra: extra information
*
* Handle an out of memory condition
*/
static void
xmlSaveErr(int code, xmlNodePtr node, const char *extra)
{
const char *msg = NULL;
switch(code) {
case XML_SAVE_NOT_UTF8:
msg = "string is not in UTF-8\n";
break;
case XML_SAVE_CHAR_INVALID:
msg = "invalid character value\n";
break;
case XML_SAVE_UNKNOWN_ENCODING:
msg = "unknown encoding %s\n";
break;
case XML_SAVE_NO_DOCTYPE:
msg = "document has no DOCTYPE\n";
break;
default:
msg = "unexpected error number\n";
}
__xmlSimpleError(XML_FROM_OUTPUT, code, node, msg, extra);
}
/************************************************************************
* *
* Special escaping routines *
* *
************************************************************************/
static unsigned char *
xmlSerializeHexCharRef(unsigned char *out, int val) {
unsigned char *ptr;
*out++ = '&';
*out++ = '#';
*out++ = 'x';
if (val < 0x10) ptr = out;
else if (val < 0x100) ptr = out + 1;
else if (val < 0x1000) ptr = out + 2;
else if (val < 0x10000) ptr = out + 3;
else if (val < 0x100000) ptr = out + 4;
else ptr = out + 5;
out = ptr + 1;
while (val > 0) {
switch (val & 0xF) {
case 0: *ptr-- = '0'; break;
case 1: *ptr-- = '1'; break;
case 2: *ptr-- = '2'; break;
case 3: *ptr-- = '3'; break;
case 4: *ptr-- = '4'; break;
case 5: *ptr-- = '5'; break;
case 6: *ptr-- = '6'; break;
case 7: *ptr-- = '7'; break;
case 8: *ptr-- = '8'; break;
case 9: *ptr-- = '9'; break;
case 0xA: *ptr-- = 'A'; break;
case 0xB: *ptr-- = 'B'; break;
case 0xC: *ptr-- = 'C'; break;
case 0xD: *ptr-- = 'D'; break;
case 0xE: *ptr-- = 'E'; break;
case 0xF: *ptr-- = 'F'; break;
default: *ptr-- = '0'; break;
}
val >>= 4;
}
*out++ = ';';
*out = 0;
return(out);
}
/**
* xmlEscapeEntities:
* @out: a pointer to an array of bytes to store the result
* @outlen: the length of @out
* @in: a pointer to an array of unescaped UTF-8 bytes
* @inlen: the length of @in
*
* Take a block of UTF-8 chars in and escape them. Used when there is no
* encoding specified.
*
* Returns 0 if success, or -1 otherwise
* The value of @inlen after return is the number of octets consumed
* if the return value is positive, else unpredictable.
* The value of @outlen after return is the number of octets consumed.
*/
static int
xmlEscapeEntities(unsigned char* out, int *outlen,
const xmlChar* in, int *inlen) {
unsigned char* outstart = out;
const unsigned char* base = in;
unsigned char* outend = out + *outlen;
const unsigned char* inend;
int val;
inend = in + (*inlen);
while ((in < inend) && (out < outend)) {
if (*in == '<') {
if (outend - out < 4) break;
*out++ = '&';
*out++ = 'l';
*out++ = 't';
*out++ = ';';
in++;
continue;
} else if (*in == '>') {
if (outend - out < 4) break;
*out++ = '&';
*out++ = 'g';
*out++ = 't';
*out++ = ';';
in++;
continue;
} else if (*in == '&') {
if (outend - out < 5) break;
*out++ = '&';
*out++ = 'a';
*out++ = 'm';
*out++ = 'p';
*out++ = ';';
in++;
continue;
} else if (((*in >= 0x20) && (*in < 0x80)) ||
(*in == '\n') || (*in == '\t')) {
/*
* default case, just copy !
*/
*out++ = *in++;
continue;
} else if (*in >= 0x80) {
/*
* We assume we have UTF-8 input.
*/
if (outend - out < 11) break;
if (*in < 0xC0) {
xmlSaveErr(XML_SAVE_NOT_UTF8, NULL, NULL);
in++;
goto error;
} else if (*in < 0xE0) {
if (inend - in < 2) break;
val = (in[0]) & 0x1F;
val <<= 6;
val |= (in[1]) & 0x3F;
in += 2;
} else if (*in < 0xF0) {
if (inend - in < 3) break;
val = (in[0]) & 0x0F;
val <<= 6;
val |= (in[1]) & 0x3F;
val <<= 6;
val |= (in[2]) & 0x3F;
in += 3;
} else if (*in < 0xF8) {
if (inend - in < 4) break;
val = (in[0]) & 0x07;
val <<= 6;
val |= (in[1]) & 0x3F;
val <<= 6;
val |= (in[2]) & 0x3F;
val <<= 6;
val |= (in[3]) & 0x3F;
in += 4;
} else {
xmlSaveErr(XML_SAVE_CHAR_INVALID, NULL, NULL);
in++;
goto error;
}
if (!IS_CHAR(val)) {
xmlSaveErr(XML_SAVE_CHAR_INVALID, NULL, NULL);
in++;
goto error;
}
/*
* We could do multiple things here. Just save as a char ref
*/
out = xmlSerializeHexCharRef(out, val);
} else if (IS_BYTE_CHAR(*in)) {
if (outend - out < 6) break;
out = xmlSerializeHexCharRef(out, *in++);
} else {
xmlGenericError(xmlGenericErrorContext,
"xmlEscapeEntities : char out of range\n");
in++;
goto error;
}
}
*outlen = out - outstart;
*inlen = in - base;
return(0);
error:
*outlen = out - outstart;
*inlen = in - base;
return(-1);
}
/************************************************************************
* *
* Allocation and deallocation *
* *
************************************************************************/
/**
* xmlSaveCtxtInit:
* @ctxt: the saving context
*
* Initialize a saving context
*/
static void
xmlSaveCtxtInit(xmlSaveCtxtPtr ctxt)
{
int i;
int len;
if (ctxt == NULL) return;
if ((ctxt->encoding == NULL) && (ctxt->escape == NULL))
ctxt->escape = xmlEscapeEntities;
len = xmlStrlen((xmlChar *)xmlTreeIndentString);
if ((xmlTreeIndentString == NULL) || (len == 0)) {
memset(&ctxt->indent[0], 0, MAX_INDENT + 1);
} else {
ctxt->indent_size = len;
ctxt->indent_nr = MAX_INDENT / ctxt->indent_size;
for (i = 0;i < ctxt->indent_nr;i++)
memcpy(&ctxt->indent[i * ctxt->indent_size], xmlTreeIndentString,
ctxt->indent_size);
ctxt->indent[ctxt->indent_nr * ctxt->indent_size] = 0;
}
if (xmlSaveNoEmptyTags) {
ctxt->options |= XML_SAVE_NO_EMPTY;
}
}
/**
* xmlFreeSaveCtxt:
*
* Free a saving context, destroying the output in any remaining buffer
*/
static void
xmlFreeSaveCtxt(xmlSaveCtxtPtr ctxt)
{
if (ctxt == NULL) return;
if (ctxt->encoding != NULL)
xmlFree((char *) ctxt->encoding);
if (ctxt->buf != NULL)
xmlOutputBufferClose(ctxt->buf);
xmlFree(ctxt);
}
/**
* xmlNewSaveCtxt:
*
* Create a new saving context
*
* Returns the new structure or NULL in case of error
*/
static xmlSaveCtxtPtr
xmlNewSaveCtxt(const char *encoding, int options)
{
xmlSaveCtxtPtr ret;
ret = (xmlSaveCtxtPtr) xmlMalloc(sizeof(xmlSaveCtxt));
if (ret == NULL) {
xmlSaveErrMemory("creating saving context");
return ( NULL );
}
memset(ret, 0, sizeof(xmlSaveCtxt));
if (encoding != NULL) {
ret->handler = xmlFindCharEncodingHandler(encoding);
if (ret->handler == NULL) {
xmlSaveErr(XML_SAVE_UNKNOWN_ENCODING, NULL, encoding);
xmlFreeSaveCtxt(ret);
return(NULL);
}
ret->encoding = xmlStrdup((const xmlChar *)encoding);
ret->escape = NULL;
}
xmlSaveCtxtInit(ret);
/*
* Use the options
*/
/* Re-check this option as it may already have been set */
if ((ret->options & XML_SAVE_NO_EMPTY) && ! (options & XML_SAVE_NO_EMPTY)) {
options |= XML_SAVE_NO_EMPTY;
}
ret->options = options;
if (options & XML_SAVE_FORMAT)
ret->format = 1;
else if (options & XML_SAVE_WSNONSIG)
ret->format = 2;
return(ret);
}
/************************************************************************
* *
* Dumping XML tree content to a simple buffer *
* *
************************************************************************/
/**
* xmlAttrSerializeContent:
* @buf: the XML buffer output
* @doc: the document
* @attr: the attribute pointer
*
* Serialize the attribute in the buffer
*/
static void
xmlAttrSerializeContent(xmlOutputBufferPtr buf, xmlAttrPtr attr)
{
xmlNodePtr children;
children = attr->children;
while (children != NULL) {
switch (children->type) {
case XML_TEXT_NODE:
xmlBufAttrSerializeTxtContent(buf->buffer, attr->doc,
attr, children->content);
break;
case XML_ENTITY_REF_NODE:
xmlBufAdd(buf->buffer, BAD_CAST "&", 1);
xmlBufAdd(buf->buffer, children->name,
xmlStrlen(children->name));
xmlBufAdd(buf->buffer, BAD_CAST ";", 1);
break;
default:
/* should not happen unless we have a badly built tree */
break;
}
children = children->next;
}
}
/**
* xmlBufDumpNotationTable:
* @buf: an xmlBufPtr output
* @table: A notation table
*
* This will dump the content of the notation table as an XML DTD definition
*/
void
xmlBufDumpNotationTable(xmlBufPtr buf, xmlNotationTablePtr table) {
xmlBufferPtr buffer;
buffer = xmlBufferCreate();
if (buffer == NULL) {
/*
* TODO set the error in buf
*/
return;
}
xmlDumpNotationTable(buffer, table);
xmlBufMergeBuffer(buf, buffer);
}
/**
* xmlBufDumpElementDecl:
* @buf: an xmlBufPtr output
* @elem: An element table
*
* This will dump the content of the element declaration as an XML
* DTD definition
*/
void
xmlBufDumpElementDecl(xmlBufPtr buf, xmlElementPtr elem) {
xmlBufferPtr buffer;
buffer = xmlBufferCreate();
if (buffer == NULL) {
/*
* TODO set the error in buf
*/
return;
}
xmlDumpElementDecl(buffer, elem);
xmlBufMergeBuffer(buf, buffer);
}
/**
* xmlBufDumpAttributeDecl:
* @buf: an xmlBufPtr output
* @attr: An attribute declaration
*
* This will dump the content of the attribute declaration as an XML
* DTD definition
*/
void
xmlBufDumpAttributeDecl(xmlBufPtr buf, xmlAttributePtr attr) {
xmlBufferPtr buffer;
buffer = xmlBufferCreate();
if (buffer == NULL) {
/*
* TODO set the error in buf
*/
return;
}
xmlDumpAttributeDecl(buffer, attr);
xmlBufMergeBuffer(buf, buffer);
}
/**
* xmlBufDumpEntityDecl:
* @buf: an xmlBufPtr output
* @ent: An entity table
*
* This will dump the content of the entity table as an XML DTD definition
*/
void
xmlBufDumpEntityDecl(xmlBufPtr buf, xmlEntityPtr ent) {
xmlBufferPtr buffer;
buffer = xmlBufferCreate();
if (buffer == NULL) {
/*
* TODO set the error in buf
*/
return;
}
xmlDumpEntityDecl(buffer, ent);
xmlBufMergeBuffer(buf, buffer);
}
/************************************************************************
* *
* Dumping XML tree content to an I/O output buffer *
* *
************************************************************************/
static int xmlSaveSwitchEncoding(xmlSaveCtxtPtr ctxt, const char *encoding) {
xmlOutputBufferPtr buf = ctxt->buf;
if ((encoding != NULL) && (buf->encoder == NULL) && (buf->conv == NULL)) {
buf->encoder = xmlFindCharEncodingHandler((const char *)encoding);
if (buf->encoder == NULL) {
xmlSaveErr(XML_SAVE_UNKNOWN_ENCODING, NULL,
(const char *)encoding);
return(-1);
}
buf->conv = xmlBufCreate();
if (buf->conv == NULL) {
xmlCharEncCloseFunc(buf->encoder);
xmlSaveErrMemory("creating encoding buffer");
return(-1);
}
/*
* initialize the state, e.g. if outputting a BOM
*/
xmlCharEncOutput(buf, 1);
}
return(0);
}
static int xmlSaveClearEncoding(xmlSaveCtxtPtr ctxt) {
xmlOutputBufferPtr buf = ctxt->buf;
xmlOutputBufferFlush(buf);
xmlCharEncCloseFunc(buf->encoder);
xmlBufFree(buf->conv);
buf->encoder = NULL;
buf->conv = NULL;
return(0);
}
#ifdef LIBXML_HTML_ENABLED
static void
xhtmlNodeDumpOutput(xmlSaveCtxtPtr ctxt, xmlNodePtr cur);
#endif
static void xmlNodeDumpOutputInternal(xmlSaveCtxtPtr ctxt, xmlNodePtr cur);
void xmlNsListDumpOutput(xmlOutputBufferPtr buf, xmlNsPtr cur);
static int xmlDocContentDumpOutput(xmlSaveCtxtPtr ctxt, xmlDocPtr cur);
/**
* xmlOutputBufferWriteWSNonSig:
* @ctxt: The save context
* @extra: Number of extra indents to apply to ctxt->level
*
* Write out formatting for non-significant whitespace output.
*/
static void
xmlOutputBufferWriteWSNonSig(xmlSaveCtxtPtr ctxt, int extra)
{
int i;
if ((ctxt == NULL) || (ctxt->buf == NULL))
return;
xmlOutputBufferWrite(ctxt->buf, 1, "\n");
for (i = 0; i < (ctxt->level + extra); i += ctxt->indent_nr) {
xmlOutputBufferWrite(ctxt->buf, ctxt->indent_size *
((ctxt->level + extra - i) > ctxt->indent_nr ?
ctxt->indent_nr : (ctxt->level + extra - i)),
ctxt->indent);
}
}
/**
* xmlNsDumpOutput:
* @buf: the XML buffer output
* @cur: a namespace
* @ctxt: the output save context. Optional.
*
* Dump a local Namespace definition.
* Should be called in the context of attributes dumps.
* If @ctxt is supplied, @buf should be its buffer.
*/
static void
xmlNsDumpOutput(xmlOutputBufferPtr buf, xmlNsPtr cur, xmlSaveCtxtPtr ctxt) {
if ((cur == NULL) || (buf == NULL)) return;
if ((cur->type == XML_LOCAL_NAMESPACE) && (cur->href != NULL)) {
if (xmlStrEqual(cur->prefix, BAD_CAST "xml"))
return;
if (ctxt != NULL && ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 2);
else
xmlOutputBufferWrite(buf, 1, " ");
/* Within the context of an element attributes */
if (cur->prefix != NULL) {
xmlOutputBufferWrite(buf, 6, "xmlns:");
xmlOutputBufferWriteString(buf, (const char *)cur->prefix);
} else
xmlOutputBufferWrite(buf, 5, "xmlns");
xmlOutputBufferWrite(buf, 1, "=");
xmlBufWriteQuotedString(buf->buffer, cur->href);
}
}
/**
* xmlNsDumpOutputCtxt
* @ctxt: the save context
* @cur: a namespace
*
* Dump a local Namespace definition to a save context.
* Should be called in the context of attribute dumps.
*/
static void
xmlNsDumpOutputCtxt(xmlSaveCtxtPtr ctxt, xmlNsPtr cur) {
xmlNsDumpOutput(ctxt->buf, cur, ctxt);
}
/**
* xmlNsListDumpOutputCtxt
* @ctxt: the save context
* @cur: the first namespace
*
* Dump a list of local namespace definitions to a save context.
* Should be called in the context of attribute dumps.
*/
static void
xmlNsListDumpOutputCtxt(xmlSaveCtxtPtr ctxt, xmlNsPtr cur) {
while (cur != NULL) {
xmlNsDumpOutput(ctxt->buf, cur, ctxt);
cur = cur->next;
}
}
/**
* xmlNsListDumpOutput:
* @buf: the XML buffer output
* @cur: the first namespace
*
* Dump a list of local Namespace definitions.
* Should be called in the context of attributes dumps.
*/
void
xmlNsListDumpOutput(xmlOutputBufferPtr buf, xmlNsPtr cur) {
while (cur != NULL) {
xmlNsDumpOutput(buf, cur, NULL);
cur = cur->next;
}
}
/**
* xmlDtdDumpOutput:
* @buf: the XML buffer output
* @dtd: the pointer to the DTD
*
* Dump the XML document DTD, if any.
*/
static void
xmlDtdDumpOutput(xmlSaveCtxtPtr ctxt, xmlDtdPtr dtd) {
xmlOutputBufferPtr buf;
xmlNodePtr cur;
int format, level;
if (dtd == NULL) return;
if ((ctxt == NULL) || (ctxt->buf == NULL))
return;
buf = ctxt->buf;
xmlOutputBufferWrite(buf, 10, "<!DOCTYPE ");
xmlOutputBufferWriteString(buf, (const char *)dtd->name);
if (dtd->ExternalID != NULL) {
xmlOutputBufferWrite(buf, 8, " PUBLIC ");
xmlBufWriteQuotedString(buf->buffer, dtd->ExternalID);
xmlOutputBufferWrite(buf, 1, " ");
xmlBufWriteQuotedString(buf->buffer, dtd->SystemID);
} else if (dtd->SystemID != NULL) {
xmlOutputBufferWrite(buf, 8, " SYSTEM ");
xmlBufWriteQuotedString(buf->buffer, dtd->SystemID);
}
if ((dtd->entities == NULL) && (dtd->elements == NULL) &&
(dtd->attributes == NULL) && (dtd->notations == NULL) &&
(dtd->pentities == NULL)) {
xmlOutputBufferWrite(buf, 1, ">");
return;
}
xmlOutputBufferWrite(buf, 3, " [\n");
/*
* Dump the notations first they are not in the DTD children list
* Do this only on a standalone DTD or on the internal subset though.
*/
if ((dtd->notations != NULL) && ((dtd->doc == NULL) ||
(dtd->doc->intSubset == dtd))) {
xmlBufDumpNotationTable(buf->buffer,
(xmlNotationTablePtr) dtd->notations);
}
format = ctxt->format;
level = ctxt->level;
ctxt->format = 0;
ctxt->level = -1;
for (cur = dtd->children; cur != NULL; cur = cur->next) {
xmlNodeDumpOutputInternal(ctxt, cur);
}
ctxt->format = format;
ctxt->level = level;
xmlOutputBufferWrite(buf, 2, "]>");
}
/**
* xmlAttrDumpOutput:
* @buf: the XML buffer output
* @cur: the attribute pointer
*
* Dump an XML attribute
*/
static void
xmlAttrDumpOutput(xmlSaveCtxtPtr ctxt, xmlAttrPtr cur) {
xmlOutputBufferPtr buf;
if (cur == NULL) return;
buf = ctxt->buf;
if (buf == NULL) return;
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 2);
else
xmlOutputBufferWrite(buf, 1, " ");
if ((cur->ns != NULL) && (cur->ns->prefix != NULL)) {
xmlOutputBufferWriteString(buf, (const char *)cur->ns->prefix);
xmlOutputBufferWrite(buf, 1, ":");
}
xmlOutputBufferWriteString(buf, (const char *)cur->name);
xmlOutputBufferWrite(buf, 2, "=\"");
xmlAttrSerializeContent(buf, cur);
xmlOutputBufferWrite(buf, 1, "\"");
}
#ifdef LIBXML_HTML_ENABLED
/**
* htmlNodeDumpOutputInternal:
* @cur: the current node
*
* Dump an HTML node, recursive behaviour, children are printed too.
*/
static int
htmlNodeDumpOutputInternal(xmlSaveCtxtPtr ctxt, xmlNodePtr cur) {
const xmlChar *oldenc = NULL;
const xmlChar *oldctxtenc = ctxt->encoding;
const xmlChar *encoding = ctxt->encoding;
xmlOutputBufferPtr buf = ctxt->buf;
int switched_encoding = 0;
xmlDocPtr doc;
xmlInitParser();
doc = cur->doc;
if (doc != NULL) {
oldenc = doc->encoding;
if (ctxt->encoding != NULL) {
doc->encoding = BAD_CAST ctxt->encoding;
} else if (doc->encoding != NULL) {
encoding = doc->encoding;
}
}
if ((encoding != NULL) && (doc != NULL))
htmlSetMetaEncoding(doc, (const xmlChar *) encoding);
if ((encoding == NULL) && (doc != NULL))
encoding = htmlGetMetaEncoding(doc);
if (encoding == NULL)
encoding = BAD_CAST "HTML";
if ((encoding != NULL) && (oldctxtenc == NULL) &&
(buf->encoder == NULL) && (buf->conv == NULL)) {
if (xmlSaveSwitchEncoding(ctxt, (const char*) encoding) < 0) {
doc->encoding = oldenc;
return(-1);
}
switched_encoding = 1;
}
if (ctxt->options & XML_SAVE_FORMAT)
htmlNodeDumpFormatOutput(buf, doc, cur,
(const char *)encoding, 1);
else
htmlNodeDumpFormatOutput(buf, doc, cur,
(const char *)encoding, 0);
/*
* Restore the state of the saving context at the end of the document
*/
if ((switched_encoding) && (oldctxtenc == NULL)) {
xmlSaveClearEncoding(ctxt);
}
if (doc != NULL)
doc->encoding = oldenc;
return(0);
}
#endif
/**
* xmlNodeDumpOutputInternal:
* @cur: the current node
*
* Dump an XML node, recursive behaviour, children are printed too.
*/
static void
xmlNodeDumpOutputInternal(xmlSaveCtxtPtr ctxt, xmlNodePtr cur) {
int format = ctxt->format;
xmlNodePtr tmp, root, unformattedNode = NULL, parent;
xmlAttrPtr attr;
xmlChar *start, *end;
xmlOutputBufferPtr buf;
if (cur == NULL) return;
buf = ctxt->buf;
root = cur;
parent = cur->parent;
while (1) {
switch (cur->type) {
case XML_DOCUMENT_NODE:
case XML_HTML_DOCUMENT_NODE:
xmlDocContentDumpOutput(ctxt, (xmlDocPtr) cur);
break;
case XML_DTD_NODE:
xmlDtdDumpOutput(ctxt, (xmlDtdPtr) cur);
break;
case XML_DOCUMENT_FRAG_NODE:
/* Always validate cur->parent when descending. */
if ((cur->parent == parent) && (cur->children != NULL)) {
parent = cur;
cur = cur->children;
continue;
}
break;
case XML_ELEMENT_DECL:
xmlBufDumpElementDecl(buf->buffer, (xmlElementPtr) cur);
break;
case XML_ATTRIBUTE_DECL:
xmlBufDumpAttributeDecl(buf->buffer, (xmlAttributePtr) cur);
break;
case XML_ENTITY_DECL:
xmlBufDumpEntityDecl(buf->buffer, (xmlEntityPtr) cur);
break;
case XML_ELEMENT_NODE:
if ((cur != root) && (ctxt->format == 1) &&
(xmlIndentTreeOutput))
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level),
ctxt->indent);
/*
* Some users like lxml are known to pass nodes with a corrupted
* tree structure. Fall back to a recursive call to handle this
* case.
*/
if ((cur->parent != parent) && (cur->children != NULL)) {
xmlNodeDumpOutputInternal(ctxt, cur);
break;
}
xmlOutputBufferWrite(buf, 1, "<");
if ((cur->ns != NULL) && (cur->ns->prefix != NULL)) {
xmlOutputBufferWriteString(buf, (const char *)cur->ns->prefix);
xmlOutputBufferWrite(buf, 1, ":");
}
xmlOutputBufferWriteString(buf, (const char *)cur->name);
if (cur->nsDef)
xmlNsListDumpOutputCtxt(ctxt, cur->nsDef);
for (attr = cur->properties; attr != NULL; attr = attr->next)
xmlAttrDumpOutput(ctxt, attr);
if (cur->children == NULL) {
if ((ctxt->options & XML_SAVE_NO_EMPTY) == 0) {
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 0);
xmlOutputBufferWrite(buf, 2, "/>");
} else {
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 1);
xmlOutputBufferWrite(buf, 3, "></");
if ((cur->ns != NULL) && (cur->ns->prefix != NULL)) {
xmlOutputBufferWriteString(buf,
(const char *)cur->ns->prefix);
xmlOutputBufferWrite(buf, 1, ":");
}
xmlOutputBufferWriteString(buf, (const char *)cur->name);
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 0);
xmlOutputBufferWrite(buf, 1, ">");
}
} else {
if (ctxt->format == 1) {
tmp = cur->children;
while (tmp != NULL) {
if ((tmp->type == XML_TEXT_NODE) ||
(tmp->type == XML_CDATA_SECTION_NODE) ||
(tmp->type == XML_ENTITY_REF_NODE)) {
ctxt->format = 0;
unformattedNode = cur;
break;
}
tmp = tmp->next;
}
}
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 1);
xmlOutputBufferWrite(buf, 1, ">");
if (ctxt->format == 1) xmlOutputBufferWrite(buf, 1, "\n");
if (ctxt->level >= 0) ctxt->level++;
parent = cur;
cur = cur->children;
continue;
}
break;
case XML_TEXT_NODE:
if (cur->content == NULL)
break;
if (cur->name != xmlStringTextNoenc) {
xmlOutputBufferWriteEscape(buf, cur->content, ctxt->escape);
} else {
/*
* Disable escaping, needed for XSLT
*/
xmlOutputBufferWriteString(buf, (const char *) cur->content);
}
break;
case XML_PI_NODE:
if ((cur != root) && (ctxt->format == 1) && (xmlIndentTreeOutput))
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level),
ctxt->indent);
if (cur->content != NULL) {
xmlOutputBufferWrite(buf, 2, "<?");
xmlOutputBufferWriteString(buf, (const char *)cur->name);
if (cur->content != NULL) {
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 0);
else
xmlOutputBufferWrite(buf, 1, " ");
xmlOutputBufferWriteString(buf,
(const char *)cur->content);
}
xmlOutputBufferWrite(buf, 2, "?>");
} else {
xmlOutputBufferWrite(buf, 2, "<?");
xmlOutputBufferWriteString(buf, (const char *)cur->name);
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 0);
xmlOutputBufferWrite(buf, 2, "?>");
}
break;
case XML_COMMENT_NODE:
if ((cur != root) && (ctxt->format == 1) && (xmlIndentTreeOutput))
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level),
ctxt->indent);
if (cur->content != NULL) {
xmlOutputBufferWrite(buf, 4, "<!--");
xmlOutputBufferWriteString(buf, (const char *)cur->content);
xmlOutputBufferWrite(buf, 3, "-->");
}
break;
case XML_ENTITY_REF_NODE:
xmlOutputBufferWrite(buf, 1, "&");
xmlOutputBufferWriteString(buf, (const char *)cur->name);
xmlOutputBufferWrite(buf, 1, ";");
break;
case XML_CDATA_SECTION_NODE:
if (cur->content == NULL || *cur->content == '\0') {
xmlOutputBufferWrite(buf, 12, "<![CDATA[]]>");
} else {
start = end = cur->content;
while (*end != '\0') {
if ((*end == ']') && (*(end + 1) == ']') &&
(*(end + 2) == '>')) {
end = end + 2;
xmlOutputBufferWrite(buf, 9, "<![CDATA[");
xmlOutputBufferWrite(buf, end - start,
(const char *)start);
xmlOutputBufferWrite(buf, 3, "]]>");
start = end;
}
end++;
}
if (start != end) {
xmlOutputBufferWrite(buf, 9, "<![CDATA[");
xmlOutputBufferWriteString(buf, (const char *)start);
xmlOutputBufferWrite(buf, 3, "]]>");
}
}
break;
case XML_ATTRIBUTE_NODE:
xmlAttrDumpOutput(ctxt, (xmlAttrPtr) cur);
break;
case XML_NAMESPACE_DECL:
xmlNsDumpOutputCtxt(ctxt, (xmlNsPtr) cur);
break;
default:
break;
}
while (1) {
if (cur == root)
return;
if ((ctxt->format == 1) &&
(cur->type != XML_XINCLUDE_START) &&
(cur->type != XML_XINCLUDE_END))
xmlOutputBufferWrite(buf, 1, "\n");
if (cur->next != NULL) {
cur = cur->next;
break;
}
cur = parent;
/* cur->parent was validated when descending. */
parent = cur->parent;
if (cur->type == XML_ELEMENT_NODE) {
if (ctxt->level > 0) ctxt->level--;
if ((xmlIndentTreeOutput) && (ctxt->format == 1))
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level),
ctxt->indent);
xmlOutputBufferWrite(buf, 2, "</");
if ((cur->ns != NULL) && (cur->ns->prefix != NULL)) {
xmlOutputBufferWriteString(buf,
(const char *)cur->ns->prefix);
xmlOutputBufferWrite(buf, 1, ":");
}
xmlOutputBufferWriteString(buf, (const char *)cur->name);
if (ctxt->format == 2)
xmlOutputBufferWriteWSNonSig(ctxt, 0);
xmlOutputBufferWrite(buf, 1, ">");
if (cur == unformattedNode) {
ctxt->format = format;
unformattedNode = NULL;
}
}
}
}
}
/**
* xmlDocContentDumpOutput:
* @cur: the document
*
* Dump an XML document.
*/
static int
xmlDocContentDumpOutput(xmlSaveCtxtPtr ctxt, xmlDocPtr cur) {
#ifdef LIBXML_HTML_ENABLED
xmlDtdPtr dtd;
int is_xhtml = 0;
#endif
const xmlChar *oldenc = cur->encoding;
const xmlChar *oldctxtenc = ctxt->encoding;
const xmlChar *encoding = ctxt->encoding;
xmlCharEncodingOutputFunc oldescape = ctxt->escape;
xmlCharEncodingOutputFunc oldescapeAttr = ctxt->escapeAttr;
xmlOutputBufferPtr buf = ctxt->buf;
xmlCharEncoding enc;
int switched_encoding = 0;
xmlInitParser();
if ((cur->type != XML_HTML_DOCUMENT_NODE) &&
(cur->type != XML_DOCUMENT_NODE))
return(-1);
if (ctxt->encoding != NULL) {
cur->encoding = BAD_CAST ctxt->encoding;
} else if (cur->encoding != NULL) {
encoding = cur->encoding;
}
if (((cur->type == XML_HTML_DOCUMENT_NODE) &&
((ctxt->options & XML_SAVE_AS_XML) == 0) &&
((ctxt->options & XML_SAVE_XHTML) == 0)) ||
(ctxt->options & XML_SAVE_AS_HTML)) {
#ifdef LIBXML_HTML_ENABLED
if (encoding != NULL)
htmlSetMetaEncoding(cur, (const xmlChar *) encoding);
if (encoding == NULL)
encoding = htmlGetMetaEncoding(cur);
if (encoding == NULL)
encoding = BAD_CAST "HTML";
if ((encoding != NULL) && (oldctxtenc == NULL) &&
(buf->encoder == NULL) && (buf->conv == NULL)) {
if (xmlSaveSwitchEncoding(ctxt, (const char*) encoding) < 0) {
cur->encoding = oldenc;
return(-1);
}
}
if (ctxt->options & XML_SAVE_FORMAT)
htmlDocContentDumpFormatOutput(buf, cur,
(const char *)encoding, 1);
else
htmlDocContentDumpFormatOutput(buf, cur,
(const char *)encoding, 0);
if (ctxt->encoding != NULL)
cur->encoding = oldenc;
return(0);
#else
return(-1);
#endif
} else if ((cur->type == XML_DOCUMENT_NODE) ||
(ctxt->options & XML_SAVE_AS_XML) ||
(ctxt->options & XML_SAVE_XHTML)) {
enc = xmlParseCharEncoding((const char*) encoding);
if ((encoding != NULL) && (oldctxtenc == NULL) &&
(buf->encoder == NULL) && (buf->conv == NULL) &&
((ctxt->options & XML_SAVE_NO_DECL) == 0)) {
if ((enc != XML_CHAR_ENCODING_UTF8) &&
(enc != XML_CHAR_ENCODING_NONE) &&
(enc != XML_CHAR_ENCODING_ASCII)) {
/*
* we need to switch to this encoding but just for this
* document since we output the XMLDecl the conversion
* must be done to not generate not well formed documents.
*/
if (xmlSaveSwitchEncoding(ctxt, (const char*) encoding) < 0) {
cur->encoding = oldenc;
return(-1);
}
switched_encoding = 1;
}
if (ctxt->escape == xmlEscapeEntities)
ctxt->escape = NULL;
if (ctxt->escapeAttr == xmlEscapeEntities)
ctxt->escapeAttr = NULL;
}
/*
* Save the XML declaration
*/
if ((ctxt->options & XML_SAVE_NO_DECL) == 0) {
xmlOutputBufferWrite(buf, 14, "<?xml version=");
if (cur->version != NULL)
xmlBufWriteQuotedString(buf->buffer, cur->version);
else
xmlOutputBufferWrite(buf, 5, "\"1.0\"");
if (encoding != NULL) {
xmlOutputBufferWrite(buf, 10, " encoding=");
xmlBufWriteQuotedString(buf->buffer, (xmlChar *) encoding);
}
switch (cur->standalone) {
case 0:
xmlOutputBufferWrite(buf, 16, " standalone=\"no\"");
break;
case 1:
xmlOutputBufferWrite(buf, 17, " standalone=\"yes\"");
break;
}
xmlOutputBufferWrite(buf, 3, "?>\n");
}
#ifdef LIBXML_HTML_ENABLED
if (ctxt->options & XML_SAVE_XHTML)
is_xhtml = 1;
if ((ctxt->options & XML_SAVE_NO_XHTML) == 0) {
dtd = xmlGetIntSubset(cur);
if (dtd != NULL) {
is_xhtml = xmlIsXHTML(dtd->SystemID, dtd->ExternalID);
if (is_xhtml < 0) is_xhtml = 0;
}
}
#endif
if (cur->children != NULL) {
xmlNodePtr child = cur->children;
while (child != NULL) {
ctxt->level = 0;
#ifdef LIBXML_HTML_ENABLED
if (is_xhtml)
xhtmlNodeDumpOutput(ctxt, child);
else
#endif
xmlNodeDumpOutputInternal(ctxt, child);
if ((child->type != XML_XINCLUDE_START) &&
(child->type != XML_XINCLUDE_END))
xmlOutputBufferWrite(buf, 1, "\n");
child = child->next;
}
}
}
/*
* Restore the state of the saving context at the end of the document
*/
if ((switched_encoding) && (oldctxtenc == NULL)) {
xmlSaveClearEncoding(ctxt);
ctxt->escape = oldescape;
ctxt->escapeAttr = oldescapeAttr;
}
cur->encoding = oldenc;
return(0);
}
#ifdef LIBXML_HTML_ENABLED
/************************************************************************
* *
* Functions specific to XHTML serialization *
* *
************************************************************************/
/**
* xhtmlIsEmpty:
* @node: the node
*
* Check if a node is an empty xhtml node
*
* Returns 1 if the node is an empty node, 0 if not and -1 in case of error
*/
static int
xhtmlIsEmpty(xmlNodePtr node) {
if (node == NULL)
return(-1);
if (node->type != XML_ELEMENT_NODE)
return(0);
if ((node->ns != NULL) && (!xmlStrEqual(node->ns->href, XHTML_NS_NAME)))
return(0);
if (node->children != NULL)
return(0);
switch (node->name[0]) {
case 'a':
if (xmlStrEqual(node->name, BAD_CAST "area"))
return(1);
return(0);
case 'b':
if (xmlStrEqual(node->name, BAD_CAST "br"))
return(1);
if (xmlStrEqual(node->name, BAD_CAST "base"))
return(1);
if (xmlStrEqual(node->name, BAD_CAST "basefont"))
return(1);
return(0);
case 'c':
if (xmlStrEqual(node->name, BAD_CAST "col"))
return(1);
return(0);
case 'f':
if (xmlStrEqual(node->name, BAD_CAST "frame"))
return(1);
return(0);
case 'h':
if (xmlStrEqual(node->name, BAD_CAST "hr"))
return(1);
return(0);
case 'i':
if (xmlStrEqual(node->name, BAD_CAST "img"))
return(1);
if (xmlStrEqual(node->name, BAD_CAST "input"))
return(1);
if (xmlStrEqual(node->name, BAD_CAST "isindex"))
return(1);
return(0);
case 'l':
if (xmlStrEqual(node->name, BAD_CAST "link"))
return(1);
return(0);
case 'm':
if (xmlStrEqual(node->name, BAD_CAST "meta"))
return(1);
return(0);
case 'p':
if (xmlStrEqual(node->name, BAD_CAST "param"))
return(1);
return(0);
}
return(0);
}
/**
* xhtmlAttrListDumpOutput:
* @cur: the first attribute pointer
*
* Dump a list of XML attributes
*/
static void
xhtmlAttrListDumpOutput(xmlSaveCtxtPtr ctxt, xmlAttrPtr cur) {
xmlAttrPtr xml_lang = NULL;
xmlAttrPtr lang = NULL;
xmlAttrPtr name = NULL;
xmlAttrPtr id = NULL;
xmlNodePtr parent;
xmlOutputBufferPtr buf;
if (cur == NULL) return;
buf = ctxt->buf;
parent = cur->parent;
while (cur != NULL) {
if ((cur->ns == NULL) && (xmlStrEqual(cur->name, BAD_CAST "id")))
id = cur;
else
if ((cur->ns == NULL) && (xmlStrEqual(cur->name, BAD_CAST "name")))
name = cur;
else
if ((cur->ns == NULL) && (xmlStrEqual(cur->name, BAD_CAST "lang")))
lang = cur;
else
if ((cur->ns != NULL) && (xmlStrEqual(cur->name, BAD_CAST "lang")) &&
(xmlStrEqual(cur->ns->prefix, BAD_CAST "xml")))
xml_lang = cur;
else if ((cur->ns == NULL) &&
((cur->children == NULL) ||
(cur->children->content == NULL) ||
(cur->children->content[0] == 0)) &&
(htmlIsBooleanAttr(cur->name))) {
if (cur->children != NULL)
xmlFreeNode(cur->children);
cur->children = xmlNewText(cur->name);
if (cur->children != NULL)
cur->children->parent = (xmlNodePtr) cur;
}
xmlAttrDumpOutput(ctxt, cur);
cur = cur->next;
}
/*
* C.8
*/
if ((name != NULL) && (id == NULL)) {
if ((parent != NULL) && (parent->name != NULL) &&
((xmlStrEqual(parent->name, BAD_CAST "a")) ||
(xmlStrEqual(parent->name, BAD_CAST "p")) ||
(xmlStrEqual(parent->name, BAD_CAST "div")) ||
(xmlStrEqual(parent->name, BAD_CAST "img")) ||
(xmlStrEqual(parent->name, BAD_CAST "map")) ||
(xmlStrEqual(parent->name, BAD_CAST "applet")) ||
(xmlStrEqual(parent->name, BAD_CAST "form")) ||
(xmlStrEqual(parent->name, BAD_CAST "frame")) ||
(xmlStrEqual(parent->name, BAD_CAST "iframe")))) {
xmlOutputBufferWrite(buf, 5, " id=\"");
xmlAttrSerializeContent(buf, name);
xmlOutputBufferWrite(buf, 1, "\"");
}
}
/*
* C.7.
*/
if ((lang != NULL) && (xml_lang == NULL)) {
xmlOutputBufferWrite(buf, 11, " xml:lang=\"");
xmlAttrSerializeContent(buf, lang);
xmlOutputBufferWrite(buf, 1, "\"");
} else
if ((xml_lang != NULL) && (lang == NULL)) {
xmlOutputBufferWrite(buf, 7, " lang=\"");
xmlAttrSerializeContent(buf, xml_lang);
xmlOutputBufferWrite(buf, 1, "\"");
}
}
/**
* xhtmlNodeDumpOutput:
* @buf: the XML buffer output
* @doc: the XHTML document
* @cur: the current node
* @level: the imbrication level for indenting
* @format: is formatting allowed
* @encoding: an optional encoding string
*
* Dump an XHTML node, recursive behaviour, children are printed too.
*/
static void
xhtmlNodeDumpOutput(xmlSaveCtxtPtr ctxt, xmlNodePtr cur) {
int format = ctxt->format, addmeta;
xmlNodePtr tmp, root, unformattedNode = NULL;
xmlChar *start, *end;
xmlOutputBufferPtr buf = ctxt->buf;
if (cur == NULL) return;
root = cur;
while (1) {
switch (cur->type) {
case XML_DOCUMENT_NODE:
case XML_HTML_DOCUMENT_NODE:
xmlDocContentDumpOutput(ctxt, (xmlDocPtr) cur);
break;
case XML_NAMESPACE_DECL:
xmlNsDumpOutputCtxt(ctxt, (xmlNsPtr) cur);
break;
case XML_DTD_NODE:
xmlDtdDumpOutput(ctxt, (xmlDtdPtr) cur);
break;
case XML_DOCUMENT_FRAG_NODE:
if (cur->children) {
cur = cur->children;
continue;
}
break;
case XML_ELEMENT_DECL:
xmlBufDumpElementDecl(buf->buffer, (xmlElementPtr) cur);
break;
case XML_ATTRIBUTE_DECL:
xmlBufDumpAttributeDecl(buf->buffer, (xmlAttributePtr) cur);
break;
case XML_ENTITY_DECL:
xmlBufDumpEntityDecl(buf->buffer, (xmlEntityPtr) cur);
break;
case XML_ELEMENT_NODE:
addmeta = 0;
if ((cur != root) && (ctxt->format == 1) && (xmlIndentTreeOutput))
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level),
ctxt->indent);
xmlOutputBufferWrite(buf, 1, "<");
if ((cur->ns != NULL) && (cur->ns->prefix != NULL)) {
xmlOutputBufferWriteString(buf, (const char *)cur->ns->prefix);
xmlOutputBufferWrite(buf, 1, ":");
}
xmlOutputBufferWriteString(buf, (const char *)cur->name);
if (cur->nsDef)
xmlNsListDumpOutputCtxt(ctxt, cur->nsDef);
if ((xmlStrEqual(cur->name, BAD_CAST "html") &&
(cur->ns == NULL) && (cur->nsDef == NULL))) {
/*
* 3.1.1. Strictly Conforming Documents A.3.1.1 3/
*/
xmlOutputBufferWriteString(buf,
" xmlns=\"http://www.w3.org/1999/xhtml\"");
}
if (cur->properties != NULL)
xhtmlAttrListDumpOutput(ctxt, cur->properties);
if ((cur->parent != NULL) &&
(cur->parent->parent == (xmlNodePtr) cur->doc) &&
xmlStrEqual(cur->name, BAD_CAST"head") &&
xmlStrEqual(cur->parent->name, BAD_CAST"html")) {
tmp = cur->children;
while (tmp != NULL) {
if (xmlStrEqual(tmp->name, BAD_CAST"meta")) {
xmlChar *httpequiv;
httpequiv = xmlGetProp(tmp, BAD_CAST"http-equiv");
if (httpequiv != NULL) {
if (xmlStrcasecmp(httpequiv,
BAD_CAST"Content-Type") == 0) {
xmlFree(httpequiv);
break;
}
xmlFree(httpequiv);
}
}
tmp = tmp->next;
}
if (tmp == NULL)
addmeta = 1;
}
if (cur->children == NULL) {
if (((cur->ns == NULL) || (cur->ns->prefix == NULL)) &&
((xhtmlIsEmpty(cur) == 1) && (addmeta == 0))) {
/*
* C.2. Empty Elements
*/
xmlOutputBufferWrite(buf, 3, " />");
} else {
if (addmeta == 1) {
xmlOutputBufferWrite(buf, 1, ">");
if (ctxt->format == 1) {
xmlOutputBufferWrite(buf, 1, "\n");
if (xmlIndentTreeOutput)
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level + 1 > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level + 1),
ctxt->indent);
}
xmlOutputBufferWriteString(buf,
"<meta http-equiv=\"Content-Type\" "
"content=\"text/html; charset=");
if (ctxt->encoding) {
xmlOutputBufferWriteString(buf,
(const char *)ctxt->encoding);
} else {
xmlOutputBufferWrite(buf, 5, "UTF-8");
}
xmlOutputBufferWrite(buf, 4, "\" />");
if (ctxt->format == 1)
xmlOutputBufferWrite(buf, 1, "\n");
} else {
xmlOutputBufferWrite(buf, 1, ">");
}
/*
* C.3. Element Minimization and Empty Element Content
*/
xmlOutputBufferWrite(buf, 2, "</");
if ((cur->ns != NULL) && (cur->ns->prefix != NULL)) {
xmlOutputBufferWriteString(buf,
(const char *)cur->ns->prefix);
xmlOutputBufferWrite(buf, 1, ":");
}
xmlOutputBufferWriteString(buf, (const char *)cur->name);
xmlOutputBufferWrite(buf, 1, ">");
}
} else {
xmlOutputBufferWrite(buf, 1, ">");
if (addmeta == 1) {
if (ctxt->format == 1) {
xmlOutputBufferWrite(buf, 1, "\n");
if (xmlIndentTreeOutput)
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level + 1 > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level + 1),
ctxt->indent);
}
xmlOutputBufferWriteString(buf,
"<meta http-equiv=\"Content-Type\" "
"content=\"text/html; charset=");
if (ctxt->encoding) {
xmlOutputBufferWriteString(buf,
(const char *)ctxt->encoding);
} else {
xmlOutputBufferWrite(buf, 5, "UTF-8");
}
xmlOutputBufferWrite(buf, 4, "\" />");
}
if (ctxt->format == 1) {
tmp = cur->children;
while (tmp != NULL) {
if ((tmp->type == XML_TEXT_NODE) ||
(tmp->type == XML_ENTITY_REF_NODE)) {
unformattedNode = cur;
ctxt->format = 0;
break;
}
tmp = tmp->next;
}
}
if (ctxt->format == 1) xmlOutputBufferWrite(buf, 1, "\n");
if (ctxt->level >= 0) ctxt->level++;
cur = cur->children;
continue;
}
break;
case XML_TEXT_NODE:
if (cur->content == NULL)
break;
if ((cur->name == xmlStringText) ||
(cur->name != xmlStringTextNoenc)) {
xmlOutputBufferWriteEscape(buf, cur->content, ctxt->escape);
} else {
/*
* Disable escaping, needed for XSLT
*/
xmlOutputBufferWriteString(buf, (const char *) cur->content);
}
break;
case XML_PI_NODE:
if (cur->content != NULL) {
xmlOutputBufferWrite(buf, 2, "<?");
xmlOutputBufferWriteString(buf, (const char *)cur->name);
if (cur->content != NULL) {
xmlOutputBufferWrite(buf, 1, " ");
xmlOutputBufferWriteString(buf,
(const char *)cur->content);
}
xmlOutputBufferWrite(buf, 2, "?>");
} else {
xmlOutputBufferWrite(buf, 2, "<?");
xmlOutputBufferWriteString(buf, (const char *)cur->name);
xmlOutputBufferWrite(buf, 2, "?>");
}
break;
case XML_COMMENT_NODE:
if (cur->content != NULL) {
xmlOutputBufferWrite(buf, 4, "<!--");
xmlOutputBufferWriteString(buf, (const char *)cur->content);
xmlOutputBufferWrite(buf, 3, "-->");
}
break;
case XML_ENTITY_REF_NODE:
xmlOutputBufferWrite(buf, 1, "&");
xmlOutputBufferWriteString(buf, (const char *)cur->name);
xmlOutputBufferWrite(buf, 1, ";");
break;
case XML_CDATA_SECTION_NODE:
if (cur->content == NULL || *cur->content == '\0') {
xmlOutputBufferWrite(buf, 12, "<![CDATA[]]>");
} else {
start = end = cur->content;
while (*end != '\0') {
if (*end == ']' && *(end + 1) == ']' &&
*(end + 2) == '>') {
end = end + 2;
xmlOutputBufferWrite(buf, 9, "<![CDATA[");
xmlOutputBufferWrite(buf, end - start,
(const char *)start);
xmlOutputBufferWrite(buf, 3, "]]>");
start = end;
}
end++;
}
if (start != end) {
xmlOutputBufferWrite(buf, 9, "<![CDATA[");
xmlOutputBufferWriteString(buf, (const char *)start);
xmlOutputBufferWrite(buf, 3, "]]>");
}
}
break;
case XML_ATTRIBUTE_NODE:
xmlAttrDumpOutput(ctxt, (xmlAttrPtr) cur);
break;
default:
break;
}
while (1) {
if (cur == root)
return;
if (ctxt->format == 1)
xmlOutputBufferWrite(buf, 1, "\n");
if (cur->next != NULL) {
cur = cur->next;
break;
}
/*
* The parent should never be NULL here but we want to handle
* corrupted documents gracefully.
*/
if (cur->parent == NULL)
return;
cur = cur->parent;
if (cur->type == XML_ELEMENT_NODE) {
if (ctxt->level > 0) ctxt->level--;
if ((xmlIndentTreeOutput) && (ctxt->format == 1))
xmlOutputBufferWrite(buf, ctxt->indent_size *
(ctxt->level > ctxt->indent_nr ?
ctxt->indent_nr : ctxt->level),
ctxt->indent);
xmlOutputBufferWrite(buf, 2, "</");
if ((cur->ns != NULL) && (cur->ns->prefix != NULL)) {
xmlOutputBufferWriteString(buf,
(const char *)cur->ns->prefix);
xmlOutputBufferWrite(buf, 1, ":");
}
xmlOutputBufferWriteString(buf, (const char *)cur->name);
xmlOutputBufferWrite(buf, 1, ">");
if (cur == unformattedNode) {
ctxt->format = format;
unformattedNode = NULL;
}
}
}
}
}
#endif
/************************************************************************
* *
* Public entry points *
* *
************************************************************************/
/**
* xmlSaveToFd:
* @fd: a file descriptor number
* @encoding: the encoding name to use or NULL
* @options: a set of xmlSaveOptions
*
* Create a document saving context serializing to a file descriptor
* with the encoding and the options given.
*
* Returns a new serialization context or NULL in case of error.
*/
xmlSaveCtxtPtr
xmlSaveToFd(int fd, const char *encoding, int options)
{
xmlSaveCtxtPtr ret;
ret = xmlNewSaveCtxt(encoding, options);
if (ret == NULL) return(NULL);
ret->buf = xmlOutputBufferCreateFd(fd, ret->handler);
if (ret->buf == NULL) {
xmlCharEncCloseFunc(ret->handler);
xmlFreeSaveCtxt(ret);
return(NULL);
}
return(ret);
}
/**
* xmlSaveToFilename:
* @filename: a file name or an URL
* @encoding: the encoding name to use or NULL
* @options: a set of xmlSaveOptions
*
* Create a document saving context serializing to a filename or possibly
* to an URL (but this is less reliable) with the encoding and the options
* given.
*
* Returns a new serialization context or NULL in case of error.
*/
xmlSaveCtxtPtr
xmlSaveToFilename(const char *filename, const char *encoding, int options)
{
xmlSaveCtxtPtr ret;
int compression = 0; /* TODO handle compression option */
ret = xmlNewSaveCtxt(encoding, options);
if (ret == NULL) return(NULL);
ret->buf = xmlOutputBufferCreateFilename(filename, ret->handler,
compression);
if (ret->buf == NULL) {
xmlCharEncCloseFunc(ret->handler);
xmlFreeSaveCtxt(ret);
return(NULL);
}
return(ret);
}
/**
* xmlSaveToBuffer:
* @buffer: a buffer
* @encoding: the encoding name to use or NULL
* @options: a set of xmlSaveOptions
*
* Create a document saving context serializing to a buffer
* with the encoding and the options given
*
* Returns a new serialization context or NULL in case of error.
*/
xmlSaveCtxtPtr
xmlSaveToBuffer(xmlBufferPtr buffer, const char *encoding, int options)
{
xmlSaveCtxtPtr ret;
ret = xmlNewSaveCtxt(encoding, options);
if (ret == NULL) return(NULL);
ret->buf = xmlOutputBufferCreateBuffer(buffer, ret->handler);
if (ret->buf == NULL) {
xmlCharEncCloseFunc(ret->handler);
xmlFreeSaveCtxt(ret);
return(NULL);
}
return(ret);
}
/**
* xmlSaveToIO:
* @iowrite: an I/O write function
* @ioclose: an I/O close function
* @ioctx: an I/O handler
* @encoding: the encoding name to use or NULL
* @options: a set of xmlSaveOptions
*
* Create a document saving context serializing to a file descriptor
* with the encoding and the options given
*
* Returns a new serialization context or NULL in case of error.
*/
xmlSaveCtxtPtr
xmlSaveToIO(xmlOutputWriteCallback iowrite,
xmlOutputCloseCallback ioclose,
void *ioctx, const char *encoding, int options)
{
xmlSaveCtxtPtr ret;
ret = xmlNewSaveCtxt(encoding, options);
if (ret == NULL) return(NULL);
ret->buf = xmlOutputBufferCreateIO(iowrite, ioclose, ioctx, ret->handler);
if (ret->buf == NULL) {
xmlCharEncCloseFunc(ret->handler);
xmlFreeSaveCtxt(ret);
return(NULL);
}
return(ret);
}
/**
* xmlSaveDoc:
* @ctxt: a document saving context
* @doc: a document
*
* Save a full document to a saving context
* TODO: The function is not fully implemented yet as it does not return the
* byte count but 0 instead
*
* Returns the number of byte written or -1 in case of error
*/
long
xmlSaveDoc(xmlSaveCtxtPtr ctxt, xmlDocPtr doc)
{
long ret = 0;
if ((ctxt == NULL) || (doc == NULL)) return(-1);
if (xmlDocContentDumpOutput(ctxt, doc) < 0)
return(-1);
return(ret);
}
/**
* xmlSaveTree:
* @ctxt: a document saving context
* @node: the top node of the subtree to save
*
* Save a subtree starting at the node parameter to a saving context
* TODO: The function is not fully implemented yet as it does not return the
* byte count but 0 instead
*
* Returns the number of byte written or -1 in case of error
*/
long
xmlSaveTree(xmlSaveCtxtPtr ctxt, xmlNodePtr cur)
{
long ret = 0;
if ((ctxt == NULL) || (cur == NULL)) return(-1);
#ifdef LIBXML_HTML_ENABLED
if (ctxt->options & XML_SAVE_XHTML) {
xhtmlNodeDumpOutput(ctxt, cur);
return(ret);
}
if (((cur->type != XML_NAMESPACE_DECL) && (cur->doc != NULL) &&
(cur->doc->type == XML_HTML_DOCUMENT_NODE) &&
((ctxt->options & XML_SAVE_AS_XML) == 0)) ||
(ctxt->options & XML_SAVE_AS_HTML)) {
htmlNodeDumpOutputInternal(ctxt, cur);
return(ret);
}
#endif
xmlNodeDumpOutputInternal(ctxt, cur);
return(ret);
}
/**
* xmlSaveFlush:
* @ctxt: a document saving context
*
* Flush a document saving context, i.e. make sure that all bytes have
* been output.
*
* Returns the number of byte written or -1 in case of error.
*/
int
xmlSaveFlush(xmlSaveCtxtPtr ctxt)
{
if (ctxt == NULL) return(-1);
if (ctxt->buf == NULL) return(-1);
return(xmlOutputBufferFlush(ctxt->buf));
}
/**
* xmlSaveClose:
* @ctxt: a document saving context
*
* Close a document saving context, i.e. make sure that all bytes have
* been output and free the associated data.
*
* Returns the number of byte written or -1 in case of error.
*/
int
xmlSaveClose(xmlSaveCtxtPtr ctxt)
{
int ret;
if (ctxt == NULL) return(-1);
ret = xmlSaveFlush(ctxt);
xmlFreeSaveCtxt(ctxt);
return(ret);
}
/**
* xmlSaveSetEscape:
* @ctxt: a document saving context
* @escape: the escaping function
*
* Set a custom escaping function to be used for text in element content
*
* Returns 0 if successful or -1 in case of error.
*/
int
xmlSaveSetEscape(xmlSaveCtxtPtr ctxt, xmlCharEncodingOutputFunc escape)
{
if (ctxt == NULL) return(-1);
ctxt->escape = escape;
return(0);
}
/**
* xmlSaveSetAttrEscape:
* @ctxt: a document saving context
* @escape: the escaping function
*
* Set a custom escaping function to be used for text in attribute content
*
* Returns 0 if successful or -1 in case of error.
*/
int
xmlSaveSetAttrEscape(xmlSaveCtxtPtr ctxt, xmlCharEncodingOutputFunc escape)
{
if (ctxt == NULL) return(-1);
ctxt->escapeAttr = escape;
return(0);
}
/************************************************************************
* *
* Public entry points based on buffers *
* *
************************************************************************/
/**
* xmlBufAttrSerializeTxtContent:
* @buf: and xmlBufPtr output
* @doc: the document
* @attr: the attribute node
* @string: the text content
*
* Serialize text attribute values to an xmlBufPtr
*/
void
xmlBufAttrSerializeTxtContent(xmlBufPtr buf, xmlDocPtr doc,
xmlAttrPtr attr, const xmlChar * string)
{
xmlChar *base, *cur;
if (string == NULL)
return;
base = cur = (xmlChar *) string;
while (*cur != 0) {
if (*cur == '\n') {
if (base != cur)
xmlBufAdd(buf, base, cur - base);
xmlBufAdd(buf, BAD_CAST "&#10;", 5);
cur++;
base = cur;
} else if (*cur == '\r') {
if (base != cur)
xmlBufAdd(buf, base, cur - base);
xmlBufAdd(buf, BAD_CAST "&#13;", 5);
cur++;
base = cur;
} else if (*cur == '\t') {
if (base != cur)
xmlBufAdd(buf, base, cur - base);
xmlBufAdd(buf, BAD_CAST "&#9;", 4);
cur++;
base = cur;
} else if (*cur == '"') {
if (base != cur)
xmlBufAdd(buf, base, cur - base);
xmlBufAdd(buf, BAD_CAST "&quot;", 6);
cur++;
base = cur;
} else if (*cur == '<') {
if (base != cur)
xmlBufAdd(buf, base, cur - base);
xmlBufAdd(buf, BAD_CAST "&lt;", 4);
cur++;
base = cur;
} else if (*cur == '>') {
if (base != cur)
xmlBufAdd(buf, base, cur - base);
xmlBufAdd(buf, BAD_CAST "&gt;", 4);
cur++;
base = cur;
} else if (*cur == '&') {
if (base != cur)
xmlBufAdd(buf, base, cur - base);
xmlBufAdd(buf, BAD_CAST "&amp;", 5);
cur++;
base = cur;
} else if ((*cur >= 0x80) && (cur[1] != 0) &&
((doc == NULL) || (doc->encoding == NULL))) {
/*
* We assume we have UTF-8 content.
*/
unsigned char tmp[12];
int val = 0, l = 1;
if (base != cur)
xmlBufAdd(buf, base, cur - base);
if (*cur < 0xC0) {
xmlSaveErr(XML_SAVE_NOT_UTF8, (xmlNodePtr) attr, NULL);
xmlSerializeHexCharRef(tmp, *cur);
xmlBufAdd(buf, (xmlChar *) tmp, -1);
cur++;
base = cur;
continue;
} else if (*cur < 0xE0) {
val = (cur[0]) & 0x1F;
val <<= 6;
val |= (cur[1]) & 0x3F;
l = 2;
} else if ((*cur < 0xF0) && (cur [2] != 0)) {
val = (cur[0]) & 0x0F;
val <<= 6;
val |= (cur[1]) & 0x3F;
val <<= 6;
val |= (cur[2]) & 0x3F;
l = 3;
} else if ((*cur < 0xF8) && (cur [2] != 0) && (cur[3] != 0)) {
val = (cur[0]) & 0x07;
val <<= 6;
val |= (cur[1]) & 0x3F;
val <<= 6;
val |= (cur[2]) & 0x3F;
val <<= 6;
val |= (cur[3]) & 0x3F;
l = 4;
}
if ((l == 1) || (!IS_CHAR(val))) {
xmlSaveErr(XML_SAVE_CHAR_INVALID, (xmlNodePtr) attr, NULL);
xmlSerializeHexCharRef(tmp, *cur);
xmlBufAdd(buf, (xmlChar *) tmp, -1);
cur++;
base = cur;
continue;
}
/*
* We could do multiple things here. Just save
* as a char ref
*/
xmlSerializeHexCharRef(tmp, val);
xmlBufAdd(buf, (xmlChar *) tmp, -1);
cur += l;
base = cur;
} else {
cur++;
}
}
if (base != cur)
xmlBufAdd(buf, base, cur - base);
}
/**
* xmlAttrSerializeTxtContent:
* @buf: the XML buffer output
* @doc: the document
* @attr: the attribute node
* @string: the text content
*
* Serialize text attribute values to an xml simple buffer
*/
void
xmlAttrSerializeTxtContent(xmlBufferPtr buf, xmlDocPtr doc,
xmlAttrPtr attr, const xmlChar * string)
{
xmlBufPtr buffer;
if ((buf == NULL) || (string == NULL))
return;
buffer = xmlBufFromBuffer(buf);
if (buffer == NULL)
return;
xmlBufAttrSerializeTxtContent(buffer, doc, attr, string);
xmlBufBackToBuffer(buffer);
}
/**
* xmlNodeDump:
* @buf: the XML buffer output
* @doc: the document
* @cur: the current node
* @level: the imbrication level for indenting
* @format: is formatting allowed
*
* Dump an XML node, recursive behaviour,children are printed too.
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called.
* Since this is using xmlBuffer structures it is limited to 2GB and somehow
* deprecated, use xmlNodeDumpOutput() instead.
*
* Returns the number of bytes written to the buffer or -1 in case of error
*/
int
xmlNodeDump(xmlBufferPtr buf, xmlDocPtr doc, xmlNodePtr cur, int level,
int format)
{
xmlBufPtr buffer;
size_t ret;
if ((buf == NULL) || (cur == NULL))
return(-1);
buffer = xmlBufFromBuffer(buf);
if (buffer == NULL)
return(-1);
ret = xmlBufNodeDump(buffer, doc, cur, level, format);
xmlBufBackToBuffer(buffer);
if (ret > INT_MAX)
return(-1);
return((int) ret);
}
/**
* xmlBufNodeDump:
* @buf: the XML buffer output
* @doc: the document
* @cur: the current node
* @level: the imbrication level for indenting
* @format: is formatting allowed
*
* Dump an XML node, recursive behaviour,children are printed too.
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called
*
* Returns the number of bytes written to the buffer, in case of error 0
* is returned or @buf stores the error
*/
size_t
xmlBufNodeDump(xmlBufPtr buf, xmlDocPtr doc, xmlNodePtr cur, int level,
int format)
{
size_t use;
int ret;
xmlOutputBufferPtr outbuf;
int oldalloc;
xmlInitParser();
if (cur == NULL) {
#ifdef DEBUG_TREE
xmlGenericError(xmlGenericErrorContext,
"xmlNodeDump : node == NULL\n");
#endif
return (-1);
}
if (buf == NULL) {
#ifdef DEBUG_TREE
xmlGenericError(xmlGenericErrorContext,
"xmlNodeDump : buf == NULL\n");
#endif
return (-1);
}
outbuf = (xmlOutputBufferPtr) xmlMalloc(sizeof(xmlOutputBuffer));
if (outbuf == NULL) {
xmlSaveErrMemory("creating buffer");
return (-1);
}
memset(outbuf, 0, (size_t) sizeof(xmlOutputBuffer));
outbuf->buffer = buf;
outbuf->encoder = NULL;
outbuf->writecallback = NULL;
outbuf->closecallback = NULL;
outbuf->context = NULL;
outbuf->written = 0;
use = xmlBufUse(buf);
oldalloc = xmlBufGetAllocationScheme(buf);
xmlBufSetAllocationScheme(buf, XML_BUFFER_ALLOC_DOUBLEIT);
xmlNodeDumpOutput(outbuf, doc, cur, level, format, NULL);
xmlBufSetAllocationScheme(buf, oldalloc);
xmlFree(outbuf);
ret = xmlBufUse(buf) - use;
return (ret);
}
/**
* xmlElemDump:
* @f: the FILE * for the output
* @doc: the document
* @cur: the current node
*
* Dump an XML/HTML node, recursive behaviour, children are printed too.
*/
void
xmlElemDump(FILE * f, xmlDocPtr doc, xmlNodePtr cur)
{
xmlOutputBufferPtr outbuf;
xmlInitParser();
if (cur == NULL) {
#ifdef DEBUG_TREE
xmlGenericError(xmlGenericErrorContext,
"xmlElemDump : cur == NULL\n");
#endif
return;
}
#ifdef DEBUG_TREE
if (doc == NULL) {
xmlGenericError(xmlGenericErrorContext,
"xmlElemDump : doc == NULL\n");
}
#endif
outbuf = xmlOutputBufferCreateFile(f, NULL);
if (outbuf == NULL)
return;
if ((doc != NULL) && (doc->type == XML_HTML_DOCUMENT_NODE)) {
#ifdef LIBXML_HTML_ENABLED
htmlNodeDumpOutput(outbuf, doc, cur, NULL);
#else
xmlSaveErr(XML_ERR_INTERNAL_ERROR, cur, "HTML support not compiled in\n");
#endif /* LIBXML_HTML_ENABLED */
} else
xmlNodeDumpOutput(outbuf, doc, cur, 0, 1, NULL);
xmlOutputBufferClose(outbuf);
}
/************************************************************************
* *
* Saving functions front-ends *
* *
************************************************************************/
/**
* xmlNodeDumpOutput:
* @buf: the XML buffer output
* @doc: the document
* @cur: the current node
* @level: the imbrication level for indenting
* @format: is formatting allowed
* @encoding: an optional encoding string
*
* Dump an XML node, recursive behaviour, children are printed too.
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called
*/
void
xmlNodeDumpOutput(xmlOutputBufferPtr buf, xmlDocPtr doc, xmlNodePtr cur,
int level, int format, const char *encoding)
{
xmlSaveCtxt ctxt;
#ifdef LIBXML_HTML_ENABLED
xmlDtdPtr dtd;
int is_xhtml = 0;
#endif
xmlInitParser();
if ((buf == NULL) || (cur == NULL)) return;
if (encoding == NULL)
encoding = "UTF-8";
memset(&ctxt, 0, sizeof(ctxt));
ctxt.buf = buf;
ctxt.level = level;
ctxt.format = format ? 1 : 0;
ctxt.encoding = (const xmlChar *) encoding;
xmlSaveCtxtInit(&ctxt);
ctxt.options |= XML_SAVE_AS_XML;
#ifdef LIBXML_HTML_ENABLED
dtd = xmlGetIntSubset(doc);
if (dtd != NULL) {
is_xhtml = xmlIsXHTML(dtd->SystemID, dtd->ExternalID);
if (is_xhtml < 0)
is_xhtml = 0;
}
if (is_xhtml)
xhtmlNodeDumpOutput(&ctxt, cur);
else
#endif
xmlNodeDumpOutputInternal(&ctxt, cur);
}
/**
* xmlDocDumpFormatMemoryEnc:
* @out_doc: Document to generate XML text from
* @doc_txt_ptr: Memory pointer for allocated XML text
* @doc_txt_len: Length of the generated XML text
* @txt_encoding: Character encoding to use when generating XML text
* @format: should formatting spaces been added
*
* Dump the current DOM tree into memory using the character encoding specified
* by the caller. Note it is up to the caller of this function to free the
* allocated memory with xmlFree().
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called
*/
void
xmlDocDumpFormatMemoryEnc(xmlDocPtr out_doc, xmlChar **doc_txt_ptr,
int * doc_txt_len, const char * txt_encoding,
int format) {
xmlSaveCtxt ctxt;
int dummy = 0;
xmlOutputBufferPtr out_buff = NULL;
xmlCharEncodingHandlerPtr conv_hdlr = NULL;
if (doc_txt_len == NULL) {
doc_txt_len = &dummy; /* Continue, caller just won't get length */
}
if (doc_txt_ptr == NULL) {
*doc_txt_len = 0;
return;
}
*doc_txt_ptr = NULL;
*doc_txt_len = 0;
if (out_doc == NULL) {
/* No document, no output */
return;
}
/*
* Validate the encoding value, if provided.
* This logic is copied from xmlSaveFileEnc.
*/
if (txt_encoding == NULL)
txt_encoding = (const char *) out_doc->encoding;
if (txt_encoding != NULL) {
conv_hdlr = xmlFindCharEncodingHandler(txt_encoding);
if ( conv_hdlr == NULL ) {
xmlSaveErr(XML_SAVE_UNKNOWN_ENCODING, (xmlNodePtr) out_doc,
txt_encoding);
return;
}
}
if ((out_buff = xmlAllocOutputBuffer(conv_hdlr)) == NULL ) {
xmlSaveErrMemory("creating buffer");
return;
}
memset(&ctxt, 0, sizeof(ctxt));
ctxt.buf = out_buff;
ctxt.level = 0;
ctxt.format = format ? 1 : 0;
ctxt.encoding = (const xmlChar *) txt_encoding;
xmlSaveCtxtInit(&ctxt);
ctxt.options |= XML_SAVE_AS_XML;
xmlDocContentDumpOutput(&ctxt, out_doc);
xmlOutputBufferFlush(out_buff);
if (out_buff->conv != NULL) {
*doc_txt_len = xmlBufUse(out_buff->conv);
*doc_txt_ptr = xmlStrndup(xmlBufContent(out_buff->conv), *doc_txt_len);
} else {
*doc_txt_len = xmlBufUse(out_buff->buffer);
*doc_txt_ptr = xmlStrndup(xmlBufContent(out_buff->buffer),*doc_txt_len);
}
(void)xmlOutputBufferClose(out_buff);
if ((*doc_txt_ptr == NULL) && (*doc_txt_len > 0)) {
*doc_txt_len = 0;
xmlSaveErrMemory("creating output");
}
return;
}
/**
* xmlDocDumpMemory:
* @cur: the document
* @mem: OUT: the memory pointer
* @size: OUT: the memory length
*
* Dump an XML document in memory and return the #xmlChar * and it's size
* in bytes. It's up to the caller to free the memory with xmlFree().
* The resulting byte array is zero terminated, though the last 0 is not
* included in the returned size.
*/
void
xmlDocDumpMemory(xmlDocPtr cur, xmlChar**mem, int *size) {
xmlDocDumpFormatMemoryEnc(cur, mem, size, NULL, 0);
}
/**
* xmlDocDumpFormatMemory:
* @cur: the document
* @mem: OUT: the memory pointer
* @size: OUT: the memory length
* @format: should formatting spaces been added
*
*
* Dump an XML document in memory and return the #xmlChar * and it's size.
* It's up to the caller to free the memory with xmlFree().
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called
*/
void
xmlDocDumpFormatMemory(xmlDocPtr cur, xmlChar**mem, int *size, int format) {
xmlDocDumpFormatMemoryEnc(cur, mem, size, NULL, format);
}
/**
* xmlDocDumpMemoryEnc:
* @out_doc: Document to generate XML text from
* @doc_txt_ptr: Memory pointer for allocated XML text
* @doc_txt_len: Length of the generated XML text
* @txt_encoding: Character encoding to use when generating XML text
*
* Dump the current DOM tree into memory using the character encoding specified
* by the caller. Note it is up to the caller of this function to free the
* allocated memory with xmlFree().
*/
void
xmlDocDumpMemoryEnc(xmlDocPtr out_doc, xmlChar **doc_txt_ptr,
int * doc_txt_len, const char * txt_encoding) {
xmlDocDumpFormatMemoryEnc(out_doc, doc_txt_ptr, doc_txt_len,
txt_encoding, 0);
}
/**
* xmlDocFormatDump:
* @f: the FILE*
* @cur: the document
* @format: should formatting spaces been added
*
* Dump an XML document to an open FILE.
*
* returns: the number of bytes written or -1 in case of failure.
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called
*/
int
xmlDocFormatDump(FILE *f, xmlDocPtr cur, int format) {
xmlSaveCtxt ctxt;
xmlOutputBufferPtr buf;
const char * encoding;
xmlCharEncodingHandlerPtr handler = NULL;
int ret;
if (cur == NULL) {
#ifdef DEBUG_TREE
xmlGenericError(xmlGenericErrorContext,
"xmlDocDump : document == NULL\n");
#endif
return(-1);
}
encoding = (const char *) cur->encoding;
if (encoding != NULL) {
handler = xmlFindCharEncodingHandler(encoding);
if (handler == NULL) {
xmlFree((char *) cur->encoding);
cur->encoding = NULL;
encoding = NULL;
}
}
buf = xmlOutputBufferCreateFile(f, handler);
if (buf == NULL) return(-1);
memset(&ctxt, 0, sizeof(ctxt));
ctxt.buf = buf;
ctxt.level = 0;
ctxt.format = format ? 1 : 0;
ctxt.encoding = (const xmlChar *) encoding;
xmlSaveCtxtInit(&ctxt);
ctxt.options |= XML_SAVE_AS_XML;
xmlDocContentDumpOutput(&ctxt, cur);
ret = xmlOutputBufferClose(buf);
return(ret);
}
/**
* xmlDocDump:
* @f: the FILE*
* @cur: the document
*
* Dump an XML document to an open FILE.
*
* returns: the number of bytes written or -1 in case of failure.
*/
int
xmlDocDump(FILE *f, xmlDocPtr cur) {
return(xmlDocFormatDump (f, cur, 0));
}
/**
* xmlSaveFileTo:
* @buf: an output I/O buffer
* @cur: the document
* @encoding: the encoding if any assuming the I/O layer handles the transcoding
*
* Dump an XML document to an I/O buffer.
* Warning ! This call xmlOutputBufferClose() on buf which is not available
* after this call.
*
* returns: the number of bytes written or -1 in case of failure.
*/
int
xmlSaveFileTo(xmlOutputBufferPtr buf, xmlDocPtr cur, const char *encoding) {
xmlSaveCtxt ctxt;
int ret;
if (buf == NULL) return(-1);
if (cur == NULL) {
xmlOutputBufferClose(buf);
return(-1);
}
memset(&ctxt, 0, sizeof(ctxt));
ctxt.buf = buf;
ctxt.level = 0;
ctxt.format = 0;
ctxt.encoding = (const xmlChar *) encoding;
xmlSaveCtxtInit(&ctxt);
ctxt.options |= XML_SAVE_AS_XML;
xmlDocContentDumpOutput(&ctxt, cur);
ret = xmlOutputBufferClose(buf);
return(ret);
}
/**
* xmlSaveFormatFileTo:
* @buf: an output I/O buffer
* @cur: the document
* @encoding: the encoding if any assuming the I/O layer handles the transcoding
* @format: should formatting spaces been added
*
* Dump an XML document to an I/O buffer.
* Warning ! This call xmlOutputBufferClose() on buf which is not available
* after this call.
*
* returns: the number of bytes written or -1 in case of failure.
*/
int
xmlSaveFormatFileTo(xmlOutputBufferPtr buf, xmlDocPtr cur,
const char *encoding, int format)
{
xmlSaveCtxt ctxt;
int ret;
if (buf == NULL) return(-1);
if ((cur == NULL) ||
((cur->type != XML_DOCUMENT_NODE) &&
(cur->type != XML_HTML_DOCUMENT_NODE))) {
xmlOutputBufferClose(buf);
return(-1);
}
memset(&ctxt, 0, sizeof(ctxt));
ctxt.buf = buf;
ctxt.level = 0;
ctxt.format = format ? 1 : 0;
ctxt.encoding = (const xmlChar *) encoding;
xmlSaveCtxtInit(&ctxt);
ctxt.options |= XML_SAVE_AS_XML;
xmlDocContentDumpOutput(&ctxt, cur);
ret = xmlOutputBufferClose(buf);
return (ret);
}
/**
* xmlSaveFormatFileEnc:
* @filename: the filename or URL to output
* @cur: the document being saved
* @encoding: the name of the encoding to use or NULL.
* @format: should formatting spaces be added.
*
* Dump an XML document to a file or an URL.
*
* Returns the number of bytes written or -1 in case of error.
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called
*/
int
xmlSaveFormatFileEnc( const char * filename, xmlDocPtr cur,
const char * encoding, int format ) {
xmlSaveCtxt ctxt;
xmlOutputBufferPtr buf;
xmlCharEncodingHandlerPtr handler = NULL;
int ret;
if (cur == NULL)
return(-1);
if (encoding == NULL)
encoding = (const char *) cur->encoding;
if (encoding != NULL) {
handler = xmlFindCharEncodingHandler(encoding);
if (handler == NULL)
return(-1);
}
#ifdef LIBXML_ZLIB_ENABLED
if (cur->compression < 0) cur->compression = xmlGetCompressMode();
#endif
/*
* save the content to a temp buffer.
*/
buf = xmlOutputBufferCreateFilename(filename, handler, cur->compression);
if (buf == NULL) return(-1);
memset(&ctxt, 0, sizeof(ctxt));
ctxt.buf = buf;
ctxt.level = 0;
ctxt.format = format ? 1 : 0;
ctxt.encoding = (const xmlChar *) encoding;
xmlSaveCtxtInit(&ctxt);
ctxt.options |= XML_SAVE_AS_XML;
xmlDocContentDumpOutput(&ctxt, cur);
ret = xmlOutputBufferClose(buf);
return(ret);
}
/**
* xmlSaveFileEnc:
* @filename: the filename (or URL)
* @cur: the document
* @encoding: the name of an encoding (or NULL)
*
* Dump an XML document, converting it to the given encoding
*
* returns: the number of bytes written or -1 in case of failure.
*/
int
xmlSaveFileEnc(const char *filename, xmlDocPtr cur, const char *encoding) {
return ( xmlSaveFormatFileEnc( filename, cur, encoding, 0 ) );
}
/**
* xmlSaveFormatFile:
* @filename: the filename (or URL)
* @cur: the document
* @format: should formatting spaces been added
*
* Dump an XML document to a file. Will use compression if
* compiled in and enabled. If @filename is "-" the stdout file is
* used. If @format is set then the document will be indented on output.
* Note that @format = 1 provide node indenting only if xmlIndentTreeOutput = 1
* or xmlKeepBlanksDefault(0) was called
*
* returns: the number of bytes written or -1 in case of failure.
*/
int
xmlSaveFormatFile(const char *filename, xmlDocPtr cur, int format) {
return ( xmlSaveFormatFileEnc( filename, cur, NULL, format ) );
}
/**
* xmlSaveFile:
* @filename: the filename (or URL)
* @cur: the document
*
* Dump an XML document to a file. Will use compression if
* compiled in and enabled. If @filename is "-" the stdout file is
* used.
* returns: the number of bytes written or -1 in case of failure.
*/
int
xmlSaveFile(const char *filename, xmlDocPtr cur) {
return(xmlSaveFormatFileEnc(filename, cur, NULL, 0));
}
#endif /* LIBXML_OUTPUT_ENABLED */
#define bottom_xmlsave
#include "elfgcchack.h"