/*
* xpointer.c : Code to handle XML Pointer
*
* Base implementation was made accordingly to
* W3C Candidate Recommendation 7 June 2000
* http://www.w3.org/TR/2000/CR-xptr-20000607
*
* Added support for the element() scheme described in:
* W3C Proposed Recommendation 13 November 2002
* http://www.w3.org/TR/2002/PR-xptr-element-20021113/
*
* See Copyright for the status of this software.
*
* daniel@veillard.com
*/
#define IN_LIBXML
#include "libxml.h"
/*
* TODO: better handling of error cases, the full expression should
* be parsed beforehand instead of a progressive evaluation
* TODO: Access into entities references are not supported now ...
* need a start to be able to pop out of entities refs since
* parent is the endity declaration, not the ref.
*/
#include <string.h>
#include <libxml/xpointer.h>
#include <libxml/xmlmemory.h>
#include <libxml/parserInternals.h>
#include <libxml/uri.h>
#include <libxml/xpath.h>
#include <libxml/xpathInternals.h>
#include <libxml/xmlerror.h>
#include <libxml/globals.h>
#ifdef LIBXML_XPTR_ENABLED
/* Add support of the xmlns() xpointer scheme to initialize the namespaces */
#define XPTR_XMLNS_SCHEME
/* #define DEBUG_RANGES */
#ifdef DEBUG_RANGES
#ifdef LIBXML_DEBUG_ENABLED
#include <libxml/debugXML.h>
#endif
#endif
#define TODO \
xmlGenericError(xmlGenericErrorContext, \
"Unimplemented block at %s:%d\n", \
__FILE__, __LINE__);
#define STRANGE \
xmlGenericError(xmlGenericErrorContext, \
"Internal error at %s:%d\n", \
__FILE__, __LINE__);
/************************************************************************
* *
* Some factorized error routines *
* *
************************************************************************/
/**
* xmlXPtrErrMemory:
* @extra: extra informations
*
* Handle a redefinition of attribute error
*/
static void
xmlXPtrErrMemory(const char *extra)
{
__xmlRaiseError(NULL, NULL, NULL, NULL, NULL, XML_FROM_XPOINTER,
XML_ERR_NO_MEMORY, XML_ERR_ERROR, NULL, 0, extra,
NULL, NULL, 0, 0,
"Memory allocation failed : %s\n", extra);
}
/**
* xmlXPtrErr:
* @ctxt: an XPTR evaluation context
* @extra: extra informations
*
* Handle a redefinition of attribute error
*/
static void
xmlXPtrErr(xmlXPathParserContextPtr ctxt, int error,
const char * msg, const xmlChar *extra)
{
if (ctxt != NULL)
ctxt->error = error;
if ((ctxt == NULL) || (ctxt->context == NULL)) {
__xmlRaiseError(NULL, NULL, NULL,
NULL, NULL, XML_FROM_XPOINTER, error,
XML_ERR_ERROR, NULL, 0,
(const char *) extra, NULL, NULL, 0, 0,
msg, extra);
return;
}
ctxt->context->lastError.domain = XML_FROM_XPOINTER;
ctxt->context->lastError.code = error;
ctxt->context->lastError.level = XML_ERR_ERROR;
ctxt->context->lastError.str1 = (char *) xmlStrdup(ctxt->base);
ctxt->context->lastError.int1 = ctxt->cur - ctxt->base;
ctxt->context->lastError.node = ctxt->context->debugNode;
if (ctxt->context->error != NULL) {
ctxt->context->error(ctxt->context->userData,
&ctxt->context->lastError);
} else {
__xmlRaiseError(NULL, NULL, NULL,
NULL, ctxt->context->debugNode, XML_FROM_XPOINTER,
error, XML_ERR_ERROR, NULL, 0,
(const char *) extra, (const char *) ctxt->base, NULL,
ctxt->cur - ctxt->base, 0,
msg, extra);
}
}
/************************************************************************
* *
* A few helper functions for child sequences *
* *
************************************************************************/
/* xmlXPtrAdvanceNode is a private function, but used by xinclude.c */
xmlNodePtr xmlXPtrAdvanceNode(xmlNodePtr cur, int *level);
/**
* xmlXPtrGetArity:
* @cur: the node
*
* Returns the number of child for an element, -1 in case of error
*/
static int
xmlXPtrGetArity(xmlNodePtr cur) {
int i;
if (cur == NULL)
return(-1);
cur = cur->children;
for (i = 0;cur != NULL;cur = cur->next) {
if ((cur->type == XML_ELEMENT_NODE) ||
(cur->type == XML_DOCUMENT_NODE) ||
(cur->type == XML_HTML_DOCUMENT_NODE)) {
i++;
}
}
return(i);
}
/**
* xmlXPtrGetIndex:
* @cur: the node
*
* Returns the index of the node in its parent children list, -1
* in case of error
*/
static int
xmlXPtrGetIndex(xmlNodePtr cur) {
int i;
if (cur == NULL)
return(-1);
for (i = 1;cur != NULL;cur = cur->prev) {
if ((cur->type == XML_ELEMENT_NODE) ||
(cur->type == XML_DOCUMENT_NODE) ||
(cur->type == XML_HTML_DOCUMENT_NODE)) {
i++;
}
}
return(i);
}
/**
* xmlXPtrGetNthChild:
* @cur: the node
* @no: the child number
*
* Returns the @no'th element child of @cur or NULL
*/
static xmlNodePtr
xmlXPtrGetNthChild(xmlNodePtr cur, int no) {
int i;
if (cur == NULL)
return(cur);
cur = cur->children;
for (i = 0;i <= no;cur = cur->next) {
if (cur == NULL)
return(cur);
if ((cur->type == XML_ELEMENT_NODE) ||
(cur->type == XML_DOCUMENT_NODE) ||
(cur->type == XML_HTML_DOCUMENT_NODE)) {
i++;
if (i == no)
break;
}
}
return(cur);
}
/************************************************************************
* *
* Handling of XPointer specific types *
* *
************************************************************************/
/**
* xmlXPtrCmpPoints:
* @node1: the first node
* @index1: the first index
* @node2: the second node
* @index2: the second index
*
* Compare two points w.r.t document order
*
* Returns -2 in case of error 1 if first point < second point, 0 if
* that's the same point, -1 otherwise
*/
static int
xmlXPtrCmpPoints(xmlNodePtr node1, int index1, xmlNodePtr node2, int index2) {
if ((node1 == NULL) || (node2 == NULL))
return(-2);
/*
* a couple of optimizations which will avoid computations in most cases
*/
if (node1 == node2) {
if (index1 < index2)
return(1);
if (index1 > index2)
return(-1);
return(0);
}
return(xmlXPathCmpNodes(node1, node2));
}
/**
* xmlXPtrNewPoint:
* @node: the xmlNodePtr
* @indx: the indx within the node
*
* Create a new xmlXPathObjectPtr of type point
*
* Returns the newly created object.
*/
static xmlXPathObjectPtr
xmlXPtrNewPoint(xmlNodePtr node, int indx) {
xmlXPathObjectPtr ret;
if (node == NULL)
return(NULL);
if (indx < 0)
return(NULL);
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating point");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_POINT;
ret->user = (void *) node;
ret->index = indx;
return(ret);
}
/**
* xmlXPtrRangeCheckOrder:
* @range: an object range
*
* Make sure the points in the range are in the right order
*/
static void
xmlXPtrRangeCheckOrder(xmlXPathObjectPtr range) {
int tmp;
xmlNodePtr tmp2;
if (range == NULL)
return;
if (range->type != XPATH_RANGE)
return;
if (range->user2 == NULL)
return;
tmp = xmlXPtrCmpPoints(range->user, range->index,
range->user2, range->index2);
if (tmp == -1) {
tmp2 = range->user;
range->user = range->user2;
range->user2 = tmp2;
tmp = range->index;
range->index = range->index2;
range->index2 = tmp;
}
}
/**
* xmlXPtrRangesEqual:
* @range1: the first range
* @range2: the second range
*
* Compare two ranges
*
* Returns 1 if equal, 0 otherwise
*/
static int
xmlXPtrRangesEqual(xmlXPathObjectPtr range1, xmlXPathObjectPtr range2) {
if (range1 == range2)
return(1);
if ((range1 == NULL) || (range2 == NULL))
return(0);
if (range1->type != range2->type)
return(0);
if (range1->type != XPATH_RANGE)
return(0);
if (range1->user != range2->user)
return(0);
if (range1->index != range2->index)
return(0);
if (range1->user2 != range2->user2)
return(0);
if (range1->index2 != range2->index2)
return(0);
return(1);
}
/**
* xmlXPtrNewRange:
* @start: the starting node
* @startindex: the start index
* @end: the ending point
* @endindex: the ending index
*
* Create a new xmlXPathObjectPtr of type range
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewRange(xmlNodePtr start, int startindex,
xmlNodePtr end, int endindex) {
xmlXPathObjectPtr ret;
if (start == NULL)
return(NULL);
if (end == NULL)
return(NULL);
if (startindex < 0)
return(NULL);
if (endindex < 0)
return(NULL);
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating range");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_RANGE;
ret->user = start;
ret->index = startindex;
ret->user2 = end;
ret->index2 = endindex;
xmlXPtrRangeCheckOrder(ret);
return(ret);
}
/**
* xmlXPtrNewRangePoints:
* @start: the starting point
* @end: the ending point
*
* Create a new xmlXPathObjectPtr of type range using 2 Points
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewRangePoints(xmlXPathObjectPtr start, xmlXPathObjectPtr end) {
xmlXPathObjectPtr ret;
if (start == NULL)
return(NULL);
if (end == NULL)
return(NULL);
if (start->type != XPATH_POINT)
return(NULL);
if (end->type != XPATH_POINT)
return(NULL);
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating range");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_RANGE;
ret->user = start->user;
ret->index = start->index;
ret->user2 = end->user;
ret->index2 = end->index;
xmlXPtrRangeCheckOrder(ret);
return(ret);
}
/**
* xmlXPtrNewRangePointNode:
* @start: the starting point
* @end: the ending node
*
* Create a new xmlXPathObjectPtr of type range from a point to a node
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewRangePointNode(xmlXPathObjectPtr start, xmlNodePtr end) {
xmlXPathObjectPtr ret;
if (start == NULL)
return(NULL);
if (end == NULL)
return(NULL);
if (start->type != XPATH_POINT)
return(NULL);
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating range");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_RANGE;
ret->user = start->user;
ret->index = start->index;
ret->user2 = end;
ret->index2 = -1;
xmlXPtrRangeCheckOrder(ret);
return(ret);
}
/**
* xmlXPtrNewRangeNodePoint:
* @start: the starting node
* @end: the ending point
*
* Create a new xmlXPathObjectPtr of type range from a node to a point
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewRangeNodePoint(xmlNodePtr start, xmlXPathObjectPtr end) {
xmlXPathObjectPtr ret;
if (start == NULL)
return(NULL);
if (end == NULL)
return(NULL);
if (start->type != XPATH_POINT)
return(NULL);
if (end->type != XPATH_POINT)
return(NULL);
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating range");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_RANGE;
ret->user = start;
ret->index = -1;
ret->user2 = end->user;
ret->index2 = end->index;
xmlXPtrRangeCheckOrder(ret);
return(ret);
}
/**
* xmlXPtrNewRangeNodes:
* @start: the starting node
* @end: the ending node
*
* Create a new xmlXPathObjectPtr of type range using 2 nodes
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewRangeNodes(xmlNodePtr start, xmlNodePtr end) {
xmlXPathObjectPtr ret;
if (start == NULL)
return(NULL);
if (end == NULL)
return(NULL);
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating range");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_RANGE;
ret->user = start;
ret->index = -1;
ret->user2 = end;
ret->index2 = -1;
xmlXPtrRangeCheckOrder(ret);
return(ret);
}
/**
* xmlXPtrNewCollapsedRange:
* @start: the starting and ending node
*
* Create a new xmlXPathObjectPtr of type range using a single nodes
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewCollapsedRange(xmlNodePtr start) {
xmlXPathObjectPtr ret;
if (start == NULL)
return(NULL);
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating range");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_RANGE;
ret->user = start;
ret->index = -1;
ret->user2 = NULL;
ret->index2 = -1;
return(ret);
}
/**
* xmlXPtrNewRangeNodeObject:
* @start: the starting node
* @end: the ending object
*
* Create a new xmlXPathObjectPtr of type range from a not to an object
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewRangeNodeObject(xmlNodePtr start, xmlXPathObjectPtr end) {
xmlXPathObjectPtr ret;
if (start == NULL)
return(NULL);
if (end == NULL)
return(NULL);
switch (end->type) {
case XPATH_POINT:
case XPATH_RANGE:
break;
case XPATH_NODESET:
/*
* Empty set ...
*/
if (end->nodesetval->nodeNr <= 0)
return(NULL);
break;
default:
/* TODO */
return(NULL);
}
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating range");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_RANGE;
ret->user = start;
ret->index = -1;
switch (end->type) {
case XPATH_POINT:
ret->user2 = end->user;
ret->index2 = end->index;
break;
case XPATH_RANGE:
ret->user2 = end->user2;
ret->index2 = end->index2;
break;
case XPATH_NODESET: {
ret->user2 = end->nodesetval->nodeTab[end->nodesetval->nodeNr - 1];
ret->index2 = -1;
break;
}
default:
STRANGE
return(NULL);
}
xmlXPtrRangeCheckOrder(ret);
return(ret);
}
#define XML_RANGESET_DEFAULT 10
/**
* xmlXPtrLocationSetCreate:
* @val: an initial xmlXPathObjectPtr, or NULL
*
* Create a new xmlLocationSetPtr of type double and of value @val
*
* Returns the newly created object.
*/
xmlLocationSetPtr
xmlXPtrLocationSetCreate(xmlXPathObjectPtr val) {
xmlLocationSetPtr ret;
ret = (xmlLocationSetPtr) xmlMalloc(sizeof(xmlLocationSet));
if (ret == NULL) {
xmlXPtrErrMemory("allocating locationset");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlLocationSet));
if (val != NULL) {
ret->locTab = (xmlXPathObjectPtr *) xmlMalloc(XML_RANGESET_DEFAULT *
sizeof(xmlXPathObjectPtr));
if (ret->locTab == NULL) {
xmlXPtrErrMemory("allocating locationset");
xmlFree(ret);
return(NULL);
}
memset(ret->locTab, 0 ,
XML_RANGESET_DEFAULT * (size_t) sizeof(xmlXPathObjectPtr));
ret->locMax = XML_RANGESET_DEFAULT;
ret->locTab[ret->locNr++] = val;
}
return(ret);
}
/**
* xmlXPtrLocationSetAdd:
* @cur: the initial range set
* @val: a new xmlXPathObjectPtr
*
* add a new xmlXPathObjectPtr to an existing LocationSet
* If the location already exist in the set @val is freed.
*/
void
xmlXPtrLocationSetAdd(xmlLocationSetPtr cur, xmlXPathObjectPtr val) {
int i;
if ((cur == NULL) || (val == NULL)) return;
/*
* check against doublons
*/
for (i = 0;i < cur->locNr;i++) {
if (xmlXPtrRangesEqual(cur->locTab[i], val)) {
xmlXPathFreeObject(val);
return;
}
}
/*
* grow the locTab if needed
*/
if (cur->locMax == 0) {
cur->locTab = (xmlXPathObjectPtr *) xmlMalloc(XML_RANGESET_DEFAULT *
sizeof(xmlXPathObjectPtr));
if (cur->locTab == NULL) {
xmlXPtrErrMemory("adding location to set");
return;
}
memset(cur->locTab, 0 ,
XML_RANGESET_DEFAULT * (size_t) sizeof(xmlXPathObjectPtr));
cur->locMax = XML_RANGESET_DEFAULT;
} else if (cur->locNr == cur->locMax) {
xmlXPathObjectPtr *temp;
cur->locMax *= 2;
temp = (xmlXPathObjectPtr *) xmlRealloc(cur->locTab, cur->locMax *
sizeof(xmlXPathObjectPtr));
if (temp == NULL) {
xmlXPtrErrMemory("adding location to set");
return;
}
cur->locTab = temp;
}
cur->locTab[cur->locNr++] = val;
}
/**
* xmlXPtrLocationSetMerge:
* @val1: the first LocationSet
* @val2: the second LocationSet
*
* Merges two rangesets, all ranges from @val2 are added to @val1
*
* Returns val1 once extended or NULL in case of error.
*/
xmlLocationSetPtr
xmlXPtrLocationSetMerge(xmlLocationSetPtr val1, xmlLocationSetPtr val2) {
int i;
if (val1 == NULL) return(NULL);
if (val2 == NULL) return(val1);
/*
* !!!!! this can be optimized a lot, knowing that both
* val1 and val2 already have unicity of their values.
*/
for (i = 0;i < val2->locNr;i++)
xmlXPtrLocationSetAdd(val1, val2->locTab[i]);
return(val1);
}
/**
* xmlXPtrLocationSetDel:
* @cur: the initial range set
* @val: an xmlXPathObjectPtr
*
* Removes an xmlXPathObjectPtr from an existing LocationSet
*/
void
xmlXPtrLocationSetDel(xmlLocationSetPtr cur, xmlXPathObjectPtr val) {
int i;
if (cur == NULL) return;
if (val == NULL) return;
/*
* check against doublons
*/
for (i = 0;i < cur->locNr;i++)
if (cur->locTab[i] == val) break;
if (i >= cur->locNr) {
#ifdef DEBUG
xmlGenericError(xmlGenericErrorContext,
"xmlXPtrLocationSetDel: Range wasn't found in RangeList\n");
#endif
return;
}
cur->locNr--;
for (;i < cur->locNr;i++)
cur->locTab[i] = cur->locTab[i + 1];
cur->locTab[cur->locNr] = NULL;
}
/**
* xmlXPtrLocationSetRemove:
* @cur: the initial range set
* @val: the index to remove
*
* Removes an entry from an existing LocationSet list.
*/
void
xmlXPtrLocationSetRemove(xmlLocationSetPtr cur, int val) {
if (cur == NULL) return;
if (val >= cur->locNr) return;
cur->locNr--;
for (;val < cur->locNr;val++)
cur->locTab[val] = cur->locTab[val + 1];
cur->locTab[cur->locNr] = NULL;
}
/**
* xmlXPtrFreeLocationSet:
* @obj: the xmlLocationSetPtr to free
*
* Free the LocationSet compound (not the actual ranges !).
*/
void
xmlXPtrFreeLocationSet(xmlLocationSetPtr obj) {
int i;
if (obj == NULL) return;
if (obj->locTab != NULL) {
for (i = 0;i < obj->locNr; i++) {
xmlXPathFreeObject(obj->locTab[i]);
}
xmlFree(obj->locTab);
}
xmlFree(obj);
}
/**
* xmlXPtrNewLocationSetNodes:
* @start: the start NodePtr value
* @end: the end NodePtr value or NULL
*
* Create a new xmlXPathObjectPtr of type LocationSet and initialize
* it with the single range made of the two nodes @start and @end
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewLocationSetNodes(xmlNodePtr start, xmlNodePtr end) {
xmlXPathObjectPtr ret;
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating locationset");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_LOCATIONSET;
if (end == NULL)
ret->user = xmlXPtrLocationSetCreate(xmlXPtrNewCollapsedRange(start));
else
ret->user = xmlXPtrLocationSetCreate(xmlXPtrNewRangeNodes(start,end));
return(ret);
}
/**
* xmlXPtrNewLocationSetNodeSet:
* @set: a node set
*
* Create a new xmlXPathObjectPtr of type LocationSet and initialize
* it with all the nodes from @set
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrNewLocationSetNodeSet(xmlNodeSetPtr set) {
xmlXPathObjectPtr ret;
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating locationset");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_LOCATIONSET;
if (set != NULL) {
int i;
xmlLocationSetPtr newset;
newset = xmlXPtrLocationSetCreate(NULL);
if (newset == NULL)
return(ret);
for (i = 0;i < set->nodeNr;i++)
xmlXPtrLocationSetAdd(newset,
xmlXPtrNewCollapsedRange(set->nodeTab[i]));
ret->user = (void *) newset;
}
return(ret);
}
/**
* xmlXPtrWrapLocationSet:
* @val: the LocationSet value
*
* Wrap the LocationSet @val in a new xmlXPathObjectPtr
*
* Returns the newly created object.
*/
xmlXPathObjectPtr
xmlXPtrWrapLocationSet(xmlLocationSetPtr val) {
xmlXPathObjectPtr ret;
ret = (xmlXPathObjectPtr) xmlMalloc(sizeof(xmlXPathObject));
if (ret == NULL) {
xmlXPtrErrMemory("allocating locationset");
return(NULL);
}
memset(ret, 0 , (size_t) sizeof(xmlXPathObject));
ret->type = XPATH_LOCATIONSET;
ret->user = (void *) val;
return(ret);
}
/************************************************************************
* *
* The parser *
* *
************************************************************************/
static void xmlXPtrEvalChildSeq(xmlXPathParserContextPtr ctxt, xmlChar *name);
/*
* Macros for accessing the content. Those should be used only by the parser,
* and not exported.
*
* Dirty macros, i.e. one need to make assumption on the context to use them
*
* CUR_PTR return the current pointer to the xmlChar to be parsed.
* CUR returns the current xmlChar value, i.e. a 8 bit value
* in ISO-Latin or UTF-8.
* This should be used internally by the parser
* only to compare to ASCII values otherwise it would break when
* running with UTF-8 encoding.
* NXT(n) returns the n'th next xmlChar. Same as CUR is should be used only
* to compare on ASCII based substring.
* SKIP(n) Skip n xmlChar, and must also be used only to skip ASCII defined
* strings within the parser.
* CURRENT Returns the current char value, with the full decoding of
* UTF-8 if we are using this mode. It returns an int.
* NEXT Skip to the next character, this does the proper decoding
* in UTF-8 mode. It also pop-up unfinished entities on the fly.
* It returns the pointer to the current xmlChar.
*/
#define CUR (*ctxt->cur)
#define SKIP(val) ctxt->cur += (val)
#define NXT(val) ctxt->cur[(val)]
#define CUR_PTR ctxt->cur
#define SKIP_BLANKS \
while (IS_BLANK_CH(*(ctxt->cur))) NEXT
#define CURRENT (*ctxt->cur)
#define NEXT ((*ctxt->cur) ? ctxt->cur++: ctxt->cur)
/*
* xmlXPtrGetChildNo:
* @ctxt: the XPointer Parser context
* @index: the child number
*
* Move the current node of the nodeset on the stack to the
* given child if found
*/
static void
xmlXPtrGetChildNo(xmlXPathParserContextPtr ctxt, int indx) {
xmlNodePtr cur = NULL;
xmlXPathObjectPtr obj;
xmlNodeSetPtr oldset;
CHECK_TYPE(XPATH_NODESET);
obj = valuePop(ctxt);
oldset = obj->nodesetval;
if ((indx <= 0) || (oldset == NULL) || (oldset->nodeNr != 1)) {
xmlXPathFreeObject(obj);
valuePush(ctxt, xmlXPathNewNodeSet(NULL));
return;
}
cur = xmlXPtrGetNthChild(oldset->nodeTab[0], indx);
if (cur == NULL) {
xmlXPathFreeObject(obj);
valuePush(ctxt, xmlXPathNewNodeSet(NULL));
return;
}
oldset->nodeTab[0] = cur;
valuePush(ctxt, obj);
}
/**
* xmlXPtrEvalXPtrPart:
* @ctxt: the XPointer Parser context
* @name: the preparsed Scheme for the XPtrPart
*
* XPtrPart ::= 'xpointer' '(' XPtrExpr ')'
* | Scheme '(' SchemeSpecificExpr ')'
*
* Scheme ::= NCName - 'xpointer' [VC: Non-XPointer schemes]
*
* SchemeSpecificExpr ::= StringWithBalancedParens
*
* StringWithBalancedParens ::=
* [^()]* ('(' StringWithBalancedParens ')' [^()]*)*
* [VC: Parenthesis escaping]
*
* XPtrExpr ::= Expr [VC: Parenthesis escaping]
*
* VC: Parenthesis escaping:
* The end of an XPointer part is signaled by the right parenthesis ")"
* character that is balanced with the left parenthesis "(" character
* that began the part. Any unbalanced parenthesis character inside the
* expression, even within literals, must be escaped with a circumflex (^)
* character preceding it. If the expression contains any literal
* occurrences of the circumflex, each must be escaped with an additional
* circumflex (that is, ^^). If the unescaped parentheses in the expression
* are not balanced, a syntax error results.
*
* Parse and evaluate an XPtrPart. Basically it generates the unescaped
* string and if the scheme is 'xpointer' it will call the XPath interpreter.
*
* TODO: there is no new scheme registration mechanism
*/
static void
xmlXPtrEvalXPtrPart(xmlXPathParserContextPtr ctxt, xmlChar *name) {
xmlChar *buffer, *cur;
int len;
int level;
if (name == NULL)
name = xmlXPathParseName(ctxt);
if (name == NULL)
XP_ERROR(XPATH_EXPR_ERROR);
if (CUR != '(')
XP_ERROR(XPATH_EXPR_ERROR);
NEXT;
level = 1;
len = xmlStrlen(ctxt->cur);
len++;
buffer = (xmlChar *) xmlMallocAtomic(len * sizeof (xmlChar));
if (buffer == NULL) {
xmlXPtrErrMemory("allocating buffer");
return;
}
cur = buffer;
while (CUR != 0) {
if (CUR == ')') {
level--;
if (level == 0) {
NEXT;
break;
}
*cur++ = CUR;
} else if (CUR == '(') {
level++;
*cur++ = CUR;
} else if (CUR == '^') {
NEXT;
if ((CUR == ')') || (CUR == '(') || (CUR == '^')) {
*cur++ = CUR;
} else {
*cur++ = '^';
*cur++ = CUR;
}
} else {
*cur++ = CUR;
}
NEXT;
}
*cur = 0;
if ((level != 0) && (CUR == 0)) {
xmlFree(buffer);
XP_ERROR(XPTR_SYNTAX_ERROR);
}
if (xmlStrEqual(name, (xmlChar *) "xpointer")) {
const xmlChar *left = CUR_PTR;
CUR_PTR = buffer;
/*
* To evaluate an xpointer scheme element (4.3) we need:
* context initialized to the root
* context position initalized to 1
* context size initialized to 1
*/
ctxt->context->node = (xmlNodePtr)ctxt->context->doc;
ctxt->context->proximityPosition = 1;
ctxt->context->contextSize = 1;
xmlXPathEvalExpr(ctxt);
CUR_PTR=left;
} else if (xmlStrEqual(name, (xmlChar *) "element")) {
const xmlChar *left = CUR_PTR;
xmlChar *name2;
CUR_PTR = buffer;
if (buffer[0] == '/') {
xmlXPathRoot(ctxt);
xmlXPtrEvalChildSeq(ctxt, NULL);
} else {
name2 = xmlXPathParseName(ctxt);
if (name2 == NULL) {
CUR_PTR = left;
xmlFree(buffer);
XP_ERROR(XPATH_EXPR_ERROR);
}
xmlXPtrEvalChildSeq(ctxt, name2);
}
CUR_PTR = left;
#ifdef XPTR_XMLNS_SCHEME
} else if (xmlStrEqual(name, (xmlChar *) "xmlns")) {
const xmlChar *left = CUR_PTR;
xmlChar *prefix;
xmlChar *URI;
xmlURIPtr value;
CUR_PTR = buffer;
prefix = xmlXPathParseNCName(ctxt);
if (prefix == NULL) {
xmlFree(buffer);
xmlFree(name);
XP_ERROR(XPTR_SYNTAX_ERROR);
}
SKIP_BLANKS;
if (CUR != '=') {
xmlFree(prefix);
xmlFree(buffer);
xmlFree(name);
XP_ERROR(XPTR_SYNTAX_ERROR);
}
NEXT;
SKIP_BLANKS;
/* @@ check escaping in the XPointer WD */
value = xmlParseURI((const char *)ctxt->cur);
if (value == NULL) {
xmlFree(prefix);
xmlFree(buffer);
xmlFree(name);
XP_ERROR(XPTR_SYNTAX_ERROR);
}
URI = xmlSaveUri(value);
xmlFreeURI(value);
if (URI == NULL) {
xmlFree(prefix);
xmlFree(buffer);
xmlFree(name);
XP_ERROR(XPATH_MEMORY_ERROR);
}
xmlXPathRegisterNs(ctxt->context, prefix, URI);
CUR_PTR = left;
xmlFree(URI);
xmlFree(prefix);
#endif /* XPTR_XMLNS_SCHEME */
} else {
xmlXPtrErr(ctxt, XML_XPTR_UNKNOWN_SCHEME,
"unsupported scheme '%s'\n", name);
}
xmlFree(buffer);
xmlFree(name);
}
/**
* xmlXPtrEvalFullXPtr:
* @ctxt: the XPointer Parser context
* @name: the preparsed Scheme for the first XPtrPart
*
* FullXPtr ::= XPtrPart (S? XPtrPart)*
*
* As the specs says:
* -----------
* When multiple XPtrParts are provided, they must be evaluated in
* left-to-right order. If evaluation of one part fails, the nexti
* is evaluated. The following conditions cause XPointer part failure:
*
* - An unknown scheme
* - A scheme that does not locate any sub-resource present in the resource
* - A scheme that is not applicable to the media type of the resource
*
* The XPointer application must consume a failed XPointer part and
* attempt to evaluate the next one, if any. The result of the first
* XPointer part whose evaluation succeeds is taken to be the fragment
* located by the XPointer as a whole. If all the parts fail, the result
* for the XPointer as a whole is a sub-resource error.
* -----------
*
* Parse and evaluate a Full XPtr i.e. possibly a cascade of XPath based
* expressions or other schemes.
*/
static void
xmlXPtrEvalFullXPtr(xmlXPathParserContextPtr ctxt, xmlChar *name) {
if (name == NULL)
name = xmlXPathParseName(ctxt);
if (name == NULL)
XP_ERROR(XPATH_EXPR_ERROR);
while (name != NULL) {
ctxt->error = XPATH_EXPRESSION_OK;
xmlXPtrEvalXPtrPart(ctxt, name);
/* in case of syntax error, break here */
if ((ctxt->error != XPATH_EXPRESSION_OK) &&
(ctxt->error != XML_XPTR_UNKNOWN_SCHEME))
return;
/*
* If the returned value is a non-empty nodeset
* or location set, return here.
*/
if (ctxt->value != NULL) {
xmlXPathObjectPtr obj = ctxt->value;
switch (obj->type) {
case XPATH_LOCATIONSET: {
xmlLocationSetPtr loc = ctxt->value->user;
if ((loc != NULL) && (loc->locNr > 0))
return;
break;
}
case XPATH_NODESET: {
xmlNodeSetPtr loc = ctxt->value->nodesetval;
if ((loc != NULL) && (loc->nodeNr > 0))
return;
break;
}
default:
break;
}
/*
* Evaluating to improper values is equivalent to
* a sub-resource error, clean-up the stack
*/
do {
obj = valuePop(ctxt);
if (obj != NULL) {
xmlXPathFreeObject(obj);
}
} while (obj != NULL);
}
/*
* Is there another XPointer part.
*/
SKIP_BLANKS;
name = xmlXPathParseName(ctxt);
}
}
/**
* xmlXPtrEvalChildSeq:
* @ctxt: the XPointer Parser context
* @name: a possible ID name of the child sequence
*
* ChildSeq ::= '/1' ('/' [0-9]*)*
* | Name ('/' [0-9]*)+
*
* Parse and evaluate a Child Sequence. This routine also handle the
* case of a Bare Name used to get a document ID.
*/
static void
xmlXPtrEvalChildSeq(xmlXPathParserContextPtr ctxt, xmlChar *name) {
/*
* XPointer don't allow by syntax to address in mutirooted trees
* this might prove useful in some cases, warn about it.
*/
if ((name == NULL) && (CUR == '/') && (NXT(1) != '1')) {
xmlXPtrErr(ctxt, XML_XPTR_CHILDSEQ_START,
"warning: ChildSeq not starting by /1\n", NULL);
}
if (name != NULL) {
valuePush(ctxt, xmlXPathNewString(name));
xmlFree(name);
xmlXPathIdFunction(ctxt, 1);
CHECK_ERROR;
}
while (CUR == '/') {
int child = 0;
NEXT;
while ((CUR >= '0') && (CUR <= '9')) {
child = child * 10 + (CUR - '0');
NEXT;
}
xmlXPtrGetChildNo(ctxt, child);
}
}
/**
* xmlXPtrEvalXPointer:
* @ctxt: the XPointer Parser context
*
* XPointer ::= Name
* | ChildSeq
* | FullXPtr
*
* Parse and evaluate an XPointer
*/
static void
xmlXPtrEvalXPointer(xmlXPathParserContextPtr ctxt) {
if (ctxt->valueTab == NULL) {
/* Allocate the value stack */
ctxt->valueTab = (xmlXPathObjectPtr *)
xmlMalloc(10 * sizeof(xmlXPathObjectPtr));
if (ctxt->valueTab == NULL) {
xmlXPtrErrMemory("allocating evaluation context");
return;
}
ctxt->valueNr = 0;
ctxt->valueMax = 10;
ctxt->value = NULL;
ctxt->valueFrame = 0;
}
SKIP_BLANKS;
if (CUR == '/') {
xmlXPathRoot(ctxt);
xmlXPtrEvalChildSeq(ctxt, NULL);
} else {
xmlChar *name;
name = xmlXPathParseName(ctxt);
if (name == NULL)
XP_ERROR(XPATH_EXPR_ERROR);
if (CUR == '(') {
xmlXPtrEvalFullXPtr(ctxt, name);
/* Short evaluation */
return;
} else {
/* this handle both Bare Names and Child Sequences */
xmlXPtrEvalChildSeq(ctxt, name);
}
}
SKIP_BLANKS;
if (CUR != 0)
XP_ERROR(XPATH_EXPR_ERROR);
}
/************************************************************************
* *
* General routines *
* *
************************************************************************/
static
void xmlXPtrStringRangeFunction(xmlXPathParserContextPtr ctxt, int nargs);
static
void xmlXPtrStartPointFunction(xmlXPathParserContextPtr ctxt, int nargs);
static
void xmlXPtrEndPointFunction(xmlXPathParserContextPtr ctxt, int nargs);
static
void xmlXPtrHereFunction(xmlXPathParserContextPtr ctxt, int nargs);
static
void xmlXPtrOriginFunction(xmlXPathParserContextPtr ctxt, int nargs);
static
void xmlXPtrRangeInsideFunction(xmlXPathParserContextPtr ctxt, int nargs);
static
void xmlXPtrRangeFunction(xmlXPathParserContextPtr ctxt, int nargs);
/**
* xmlXPtrNewContext:
* @doc: the XML document
* @here: the node that directly contains the XPointer being evaluated or NULL
* @origin: the element from which a user or program initiated traversal of
* the link, or NULL.
*
* Create a new XPointer context
*
* Returns the xmlXPathContext just allocated.
*/
xmlXPathContextPtr
xmlXPtrNewContext(xmlDocPtr doc, xmlNodePtr here, xmlNodePtr origin) {
xmlXPathContextPtr ret;
ret = xmlXPathNewContext(doc);
if (ret == NULL)
return(ret);
ret->xptr = 1;
ret->here = here;
ret->origin = origin;
xmlXPathRegisterFunc(ret, (xmlChar *)"range-to",
xmlXPtrRangeToFunction);
xmlXPathRegisterFunc(ret, (xmlChar *)"range",
xmlXPtrRangeFunction);
xmlXPathRegisterFunc(ret, (xmlChar *)"range-inside",
xmlXPtrRangeInsideFunction);
xmlXPathRegisterFunc(ret, (xmlChar *)"string-range",
xmlXPtrStringRangeFunction);
xmlXPathRegisterFunc(ret, (xmlChar *)"start-point",
xmlXPtrStartPointFunction);
xmlXPathRegisterFunc(ret, (xmlChar *)"end-point",
xmlXPtrEndPointFunction);
xmlXPathRegisterFunc(ret, (xmlChar *)"here",
xmlXPtrHereFunction);
xmlXPathRegisterFunc(ret, (xmlChar *)" origin",
xmlXPtrOriginFunction);
return(ret);
}
/**
* xmlXPtrEval:
* @str: the XPointer expression
* @ctx: the XPointer context
*
* Evaluate the XPath Location Path in the given context.
*
* Returns the xmlXPathObjectPtr resulting from the evaluation or NULL.
* the caller has to free the object.
*/
xmlXPathObjectPtr
xmlXPtrEval(const xmlChar *str, xmlXPathContextPtr ctx) {
xmlXPathParserContextPtr ctxt;
xmlXPathObjectPtr res = NULL, tmp;
xmlXPathObjectPtr init = NULL;
int stack = 0;
xmlXPathInit();
if ((ctx == NULL) || (str == NULL))
return(NULL);
ctxt = xmlXPathNewParserContext(str, ctx);
ctxt->xptr = 1;
xmlXPtrEvalXPointer(ctxt);
if ((ctxt->value != NULL) &&
(ctxt->value->type != XPATH_NODESET) &&
(ctxt->value->type != XPATH_LOCATIONSET)) {
xmlXPtrErr(ctxt, XML_XPTR_EVAL_FAILED,
"xmlXPtrEval: evaluation failed to return a node set\n",
NULL);
} else {
res = valuePop(ctxt);
}
do {
tmp = valuePop(ctxt);
if (tmp != NULL) {
if (tmp != init) {
if (tmp->type == XPATH_NODESET) {
/*
* Evaluation may push a root nodeset which is unused
*/
xmlNodeSetPtr set;
set = tmp->nodesetval;
if ((set->nodeNr != 1) ||
(set->nodeTab[0] != (xmlNodePtr) ctx->doc))
stack++;
} else
stack++;
}
xmlXPathFreeObject(tmp);
}
} while (tmp != NULL);
if (stack != 0) {
xmlXPtrErr(ctxt, XML_XPTR_EXTRA_OBJECTS,
"xmlXPtrEval: object(s) left on the eval stack\n",
NULL);
}
if (ctxt->error != XPATH_EXPRESSION_OK) {
xmlXPathFreeObject(res);
res = NULL;
}
xmlXPathFreeParserContext(ctxt);
return(res);
}
/**
* xmlXPtrBuildRangeNodeList:
* @range: a range object
*
* Build a node list tree copy of the range
*
* Returns an xmlNodePtr list or NULL.
* the caller has to free the node tree.
*/
static xmlNodePtr
xmlXPtrBuildRangeNodeList(xmlXPathObjectPtr range) {
/* pointers to generated nodes */
xmlNodePtr list = NULL, last = NULL, parent = NULL, tmp;
/* pointers to traversal nodes */
xmlNodePtr start, cur, end;
int index1, index2;
if (range == NULL)
return(NULL);
if (range->type != XPATH_RANGE)
return(NULL);
start = (xmlNodePtr) range->user;
if (start == NULL)
return(NULL);
end = range->user2;
if (end == NULL)
return(xmlCopyNode(start, 1));
cur = start;
index1 = range->index;
index2 = range->index2;
while (cur != NULL) {
if (cur == end) {
if (cur->type == XML_TEXT_NODE) {
const xmlChar *content = cur->content;
int len;
if (content == NULL) {
tmp = xmlNewTextLen(NULL, 0);
} else {
len = index2;
if ((cur == start) && (index1 > 1)) {
content += (index1 - 1);
len -= (index1 - 1);
index1 = 0;
} else {
len = index2;
}
tmp = xmlNewTextLen(content, len);
}
/* single sub text node selection */
if (list == NULL)
return(tmp);
/* prune and return full set */
if (last != NULL)
xmlAddNextSibling(last, tmp);
else
xmlAddChild(parent, tmp);
return(list);
} else {
tmp = xmlCopyNode(cur, 0);
if (list == NULL)
list = tmp;
else {
if (last != NULL)
xmlAddNextSibling(last, tmp);
else
xmlAddChild(parent, tmp);
}
last = NULL;
parent = tmp;
if (index2 > 1) {
end = xmlXPtrGetNthChild(cur, index2 - 1);
index2 = 0;
}
if ((cur == start) && (index1 > 1)) {
cur = xmlXPtrGetNthChild(cur, index1 - 1);
index1 = 0;
} else {
cur = cur->children;
}
/*
* Now gather the remaining nodes from cur to end
*/
continue; /* while */
}
} else if ((cur == start) &&
(list == NULL) /* looks superfluous but ... */ ) {
if ((cur->type == XML_TEXT_NODE) ||
(cur->type == XML_CDATA_SECTION_NODE)) {
const xmlChar *content = cur->content;
if (content == NULL) {
tmp = xmlNewTextLen(NULL, 0);
} else {
if (index1 > 1) {
content += (index1 - 1);
}
tmp = xmlNewText(content);
}
last = list = tmp;
} else {
if ((cur == start) && (index1 > 1)) {
tmp = xmlCopyNode(cur, 0);
list = tmp;
parent = tmp;
last = NULL;
cur = xmlXPtrGetNthChild(cur, index1 - 1);
index1 = 0;
/*
* Now gather the remaining nodes from cur to end
*/
continue; /* while */
}
tmp = xmlCopyNode(cur, 1);
list = tmp;
parent = NULL;
last = tmp;
}
} else {
tmp = NULL;
switch (cur->type) {
case XML_DTD_NODE:
case XML_ELEMENT_DECL:
case XML_ATTRIBUTE_DECL:
case XML_ENTITY_NODE:
/* Do not copy DTD informations */
break;
case XML_ENTITY_DECL:
TODO /* handle crossing entities -> stack needed */
break;
case XML_XINCLUDE_START:
case XML_XINCLUDE_END:
/* don't consider it part of the tree content */
break;
case XML_ATTRIBUTE_NODE:
/* Humm, should not happen ! */
STRANGE
break;
default:
tmp = xmlCopyNode(cur, 1);
break;
}
if (tmp != NULL) {
if ((list == NULL) || ((last == NULL) && (parent == NULL))) {
STRANGE
return(NULL);
}
if (last != NULL)
xmlAddNextSibling(last, tmp);
else {
xmlAddChild(parent, tmp);
last = tmp;
}
}
}
/*
* Skip to next node in document order
*/
if ((list == NULL) || ((last == NULL) && (parent == NULL))) {
STRANGE
return(NULL);
}
cur = xmlXPtrAdvanceNode(cur, NULL);
}
return(list);
}
/**
* xmlXPtrBuildNodeList:
* @obj: the XPointer result from the evaluation.
*
* Build a node list tree copy of the XPointer result.
* This will drop Attributes and Namespace declarations.
*
* Returns an xmlNodePtr list or NULL.
* the caller has to free the node tree.
*/
xmlNodePtr
xmlXPtrBuildNodeList(xmlXPathObjectPtr obj) {
xmlNodePtr list = NULL, last = NULL;
int i;
if (obj == NULL)
return(NULL);
switch (obj->type) {
case XPATH_NODESET: {
xmlNodeSetPtr set = obj->nodesetval;
if (set == NULL)
return(NULL);
for (i = 0;i < set->nodeNr;i++) {
if (set->nodeTab[i] == NULL)
continue;
switch (set->nodeTab[i]->type) {
case XML_TEXT_NODE:
case XML_CDATA_SECTION_NODE:
case XML_ELEMENT_NODE:
case XML_ENTITY_REF_NODE:
case XML_ENTITY_NODE:
case XML_PI_NODE:
case XML_COMMENT_NODE:
case XML_DOCUMENT_NODE:
case XML_HTML_DOCUMENT_NODE:
#ifdef LIBXML_DOCB_ENABLED
case XML_DOCB_DOCUMENT_NODE:
#endif
case XML_XINCLUDE_START:
case XML_XINCLUDE_END:
break;
case XML_ATTRIBUTE_NODE:
case XML_NAMESPACE_DECL:
case XML_DOCUMENT_TYPE_NODE:
case XML_DOCUMENT_FRAG_NODE:
case XML_NOTATION_NODE:
case XML_DTD_NODE:
case XML_ELEMENT_DECL:
case XML_ATTRIBUTE_DECL:
case XML_ENTITY_DECL:
continue; /* for */
}
if (last == NULL)
list = last = xmlCopyNode(set->nodeTab[i], 1);
else {
xmlAddNextSibling(last, xmlCopyNode(set->nodeTab[i], 1));
if (last->next != NULL)
last = last->next;
}
}
break;
}
case XPATH_LOCATIONSET: {
xmlLocationSetPtr set = (xmlLocationSetPtr) obj->user;
if (set == NULL)
return(NULL);
for (i = 0;i < set->locNr;i++) {
if (last == NULL)
list = last = xmlXPtrBuildNodeList(set->locTab[i]);
else
xmlAddNextSibling(last,
xmlXPtrBuildNodeList(set->locTab[i]));
if (last != NULL) {
while (last->next != NULL)
last = last->next;
}
}
break;
}
case XPATH_RANGE:
return(xmlXPtrBuildRangeNodeList(obj));
case XPATH_POINT:
return(xmlCopyNode(obj->user, 0));
default:
break;
}
return(list);
}
/************************************************************************
* *
* XPointer functions *
* *
************************************************************************/
/**
* xmlXPtrNbLocChildren:
* @node: an xmlNodePtr
*
* Count the number of location children of @node or the length of the
* string value in case of text/PI/Comments nodes
*
* Returns the number of location children
*/
static int
xmlXPtrNbLocChildren(xmlNodePtr node) {
int ret = 0;
if (node == NULL)
return(-1);
switch (node->type) {
case XML_HTML_DOCUMENT_NODE:
case XML_DOCUMENT_NODE:
case XML_ELEMENT_NODE:
node = node->children;
while (node != NULL) {
if (node->type == XML_ELEMENT_NODE)
ret++;
node = node->next;
}
break;
case XML_ATTRIBUTE_NODE:
return(-1);
case XML_PI_NODE:
case XML_COMMENT_NODE:
case XML_TEXT_NODE:
case XML_CDATA_SECTION_NODE:
case XML_ENTITY_REF_NODE:
ret = xmlStrlen(node->content);
break;
default:
return(-1);
}
return(ret);
}
/**
* xmlXPtrHereFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Function implementing here() operation
* as described in 5.4.3
*/
static void
xmlXPtrHereFunction(xmlXPathParserContextPtr ctxt, int nargs) {
CHECK_ARITY(0);
if (ctxt->context->here == NULL)
XP_ERROR(XPTR_SYNTAX_ERROR);
valuePush(ctxt, xmlXPtrNewLocationSetNodes(ctxt->context->here, NULL));
}
/**
* xmlXPtrOriginFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Function implementing origin() operation
* as described in 5.4.3
*/
static void
xmlXPtrOriginFunction(xmlXPathParserContextPtr ctxt, int nargs) {
CHECK_ARITY(0);
if (ctxt->context->origin == NULL)
XP_ERROR(XPTR_SYNTAX_ERROR);
valuePush(ctxt, xmlXPtrNewLocationSetNodes(ctxt->context->origin, NULL));
}
/**
* xmlXPtrStartPointFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Function implementing start-point() operation
* as described in 5.4.3
* ----------------
* location-set start-point(location-set)
*
* For each location x in the argument location-set, start-point adds a
* location of type point to the result location-set. That point represents
* the start point of location x and is determined by the following rules:
*
* - If x is of type point, the start point is x.
* - If x is of type range, the start point is the start point of x.
* - If x is of type root, element, text, comment, or processing instruction,
* - the container node of the start point is x and the index is 0.
* - If x is of type attribute or namespace, the function must signal a
* syntax error.
* ----------------
*
*/
static void
xmlXPtrStartPointFunction(xmlXPathParserContextPtr ctxt, int nargs) {
xmlXPathObjectPtr tmp, obj, point;
xmlLocationSetPtr newset = NULL;
xmlLocationSetPtr oldset = NULL;
CHECK_ARITY(1);
if ((ctxt->value == NULL) ||
((ctxt->value->type != XPATH_LOCATIONSET) &&
(ctxt->value->type != XPATH_NODESET)))
XP_ERROR(XPATH_INVALID_TYPE)
obj = valuePop(ctxt);
if (obj->type == XPATH_NODESET) {
/*
* First convert to a location set
*/
tmp = xmlXPtrNewLocationSetNodeSet(obj->nodesetval);
xmlXPathFreeObject(obj);
obj = tmp;
}
newset = xmlXPtrLocationSetCreate(NULL);
if (newset == NULL) {
xmlXPathFreeObject(obj);
XP_ERROR(XPATH_MEMORY_ERROR);
}
oldset = (xmlLocationSetPtr) obj->user;
if (oldset != NULL) {
int i;
for (i = 0; i < oldset->locNr; i++) {
tmp = oldset->locTab[i];
if (tmp == NULL)
continue;
point = NULL;
switch (tmp->type) {
case XPATH_POINT:
point = xmlXPtrNewPoint(tmp->user, tmp->index);
break;
case XPATH_RANGE: {
xmlNodePtr node = tmp->user;
if (node != NULL) {
if (node->type == XML_ATTRIBUTE_NODE) {
/* TODO: Namespace Nodes ??? */
xmlXPathFreeObject(obj);
xmlXPtrFreeLocationSet(newset);
XP_ERROR(XPTR_SYNTAX_ERROR);
}
point = xmlXPtrNewPoint(node, tmp->index);
}
break;
}
default:
/*** Should we raise an error ?
xmlXPathFreeObject(obj);
xmlXPathFreeObject(newset);
XP_ERROR(XPATH_INVALID_TYPE)
***/
break;
}
if (point != NULL)
xmlXPtrLocationSetAdd(newset, point);
}
}
xmlXPathFreeObject(obj);
valuePush(ctxt, xmlXPtrWrapLocationSet(newset));
}
/**
* xmlXPtrEndPointFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Function implementing end-point() operation
* as described in 5.4.3
* ----------------------------
* location-set end-point(location-set)
*
* For each location x in the argument location-set, end-point adds a
* location of type point to the result location-set. That point represents
* the end point of location x and is determined by the following rules:
*
* - If x is of type point, the resulting point is x.
* - If x is of type range, the resulting point is the end point of x.
* - If x is of type root or element, the container node of the resulting
* point is x and the index is the number of location children of x.
* - If x is of type text, comment, or processing instruction, the container
* node of the resulting point is x and the index is the length of the
* string-value of x.
* - If x is of type attribute or namespace, the function must signal a
* syntax error.
* ----------------------------
*/
static void
xmlXPtrEndPointFunction(xmlXPathParserContextPtr ctxt, int nargs) {
xmlXPathObjectPtr tmp, obj, point;
xmlLocationSetPtr newset = NULL;
xmlLocationSetPtr oldset = NULL;
CHECK_ARITY(1);
if ((ctxt->value == NULL) ||
((ctxt->value->type != XPATH_LOCATIONSET) &&
(ctxt->value->type != XPATH_NODESET)))
XP_ERROR(XPATH_INVALID_TYPE)
obj = valuePop(ctxt);
if (obj->type == XPATH_NODESET) {
/*
* First convert to a location set
*/
tmp = xmlXPtrNewLocationSetNodeSet(obj->nodesetval);
xmlXPathFreeObject(obj);
obj = tmp;
}
newset = xmlXPtrLocationSetCreate(NULL);
oldset = (xmlLocationSetPtr) obj->user;
if (oldset != NULL) {
int i;
for (i = 0; i < oldset->locNr; i++) {
tmp = oldset->locTab[i];
if (tmp == NULL)
continue;
point = NULL;
switch (tmp->type) {
case XPATH_POINT:
point = xmlXPtrNewPoint(tmp->user, tmp->index);
break;
case XPATH_RANGE: {
xmlNodePtr node = tmp->user2;
if (node != NULL) {
if (node->type == XML_ATTRIBUTE_NODE) {
/* TODO: Namespace Nodes ??? */
xmlXPathFreeObject(obj);
xmlXPtrFreeLocationSet(newset);
XP_ERROR(XPTR_SYNTAX_ERROR);
}
point = xmlXPtrNewPoint(node, tmp->index2);
} else if (tmp->user == NULL) {
point = xmlXPtrNewPoint(node,
xmlXPtrNbLocChildren(node));
}
break;
}
default:
/*** Should we raise an error ?
xmlXPathFreeObject(obj);
xmlXPathFreeObject(newset);
XP_ERROR(XPATH_INVALID_TYPE)
***/
break;
}
if (point != NULL)
xmlXPtrLocationSetAdd(newset, point);
}
}
xmlXPathFreeObject(obj);
valuePush(ctxt, xmlXPtrWrapLocationSet(newset));
}
/**
* xmlXPtrCoveringRange:
* @ctxt: the XPointer Parser context
* @loc: the location for which the covering range must be computed
*
* A covering range is a range that wholly encompasses a location
* Section 5.3.3. Covering Ranges for All Location Types
* http://www.w3.org/TR/xptr#N2267
*
* Returns a new location or NULL in case of error
*/
static xmlXPathObjectPtr
xmlXPtrCoveringRange(xmlXPathParserContextPtr ctxt, xmlXPathObjectPtr loc) {
if (loc == NULL)
return(NULL);
if ((ctxt == NULL) || (ctxt->context == NULL) ||
(ctxt->context->doc == NULL))
return(NULL);
switch (loc->type) {
case XPATH_POINT:
return(xmlXPtrNewRange(loc->user, loc->index,
loc->user, loc->index));
case XPATH_RANGE:
if (loc->user2 != NULL) {
return(xmlXPtrNewRange(loc->user, loc->index,
loc->user2, loc->index2));
} else {
xmlNodePtr node = (xmlNodePtr) loc->user;
if (node == (xmlNodePtr) ctxt->context->doc) {
return(xmlXPtrNewRange(node, 0, node,
xmlXPtrGetArity(node)));
} else {
switch (node->type) {
case XML_ATTRIBUTE_NODE:
/* !!! our model is slightly different than XPath */
return(xmlXPtrNewRange(node, 0, node,
xmlXPtrGetArity(node)));
case XML_ELEMENT_NODE:
case XML_TEXT_NODE:
case XML_CDATA_SECTION_NODE:
case XML_ENTITY_REF_NODE:
case XML_PI_NODE:
case XML_COMMENT_NODE:
case XML_DOCUMENT_NODE:
case XML_NOTATION_NODE:
case XML_HTML_DOCUMENT_NODE: {
int indx = xmlXPtrGetIndex(node);
node = node->parent;
return(xmlXPtrNewRange(node, indx - 1,
node, indx + 1));
}
default:
return(NULL);
}
}
}
default:
TODO /* missed one case ??? */
}
return(NULL);
}
/**
* xmlXPtrRangeFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Function implementing the range() function 5.4.3
* location-set range(location-set )
*
* The range function returns ranges covering the locations in
* the argument location-set. For each location x in the argument
* location-set, a range location representing the covering range of
* x is added to the result location-set.
*/
static void
xmlXPtrRangeFunction(xmlXPathParserContextPtr ctxt, int nargs) {
int i;
xmlXPathObjectPtr set;
xmlLocationSetPtr oldset;
xmlLocationSetPtr newset;
CHECK_ARITY(1);
if ((ctxt->value == NULL) ||
((ctxt->value->type != XPATH_LOCATIONSET) &&
(ctxt->value->type != XPATH_NODESET)))
XP_ERROR(XPATH_INVALID_TYPE)
set = valuePop(ctxt);
if (set->type == XPATH_NODESET) {
xmlXPathObjectPtr tmp;
/*
* First convert to a location set
*/
tmp = xmlXPtrNewLocationSetNodeSet(set->nodesetval);
xmlXPathFreeObject(set);
set = tmp;
}
oldset = (xmlLocationSetPtr) set->user;
/*
* The loop is to compute the covering range for each item and add it
*/
newset = xmlXPtrLocationSetCreate(NULL);
for (i = 0;i < oldset->locNr;i++) {
xmlXPtrLocationSetAdd(newset,
xmlXPtrCoveringRange(ctxt, oldset->locTab[i]));
}
/*
* Save the new value and cleanup
*/
valuePush(ctxt, xmlXPtrWrapLocationSet(newset));
xmlXPathFreeObject(set);
}
/**
* xmlXPtrInsideRange:
* @ctxt: the XPointer Parser context
* @loc: the location for which the inside range must be computed
*
* A inside range is a range described in the range-inside() description
*
* Returns a new location or NULL in case of error
*/
static xmlXPathObjectPtr
xmlXPtrInsideRange(xmlXPathParserContextPtr ctxt, xmlXPathObjectPtr loc) {
if (loc == NULL)
return(NULL);
if ((ctxt == NULL) || (ctxt->context == NULL) ||
(ctxt->context->doc == NULL))
return(NULL);
switch (loc->type) {
case XPATH_POINT: {
xmlNodePtr node = (xmlNodePtr) loc->user;
switch (node->type) {
case XML_PI_NODE:
case XML_COMMENT_NODE:
case XML_TEXT_NODE:
case XML_CDATA_SECTION_NODE: {
if (node->content == NULL) {
return(xmlXPtrNewRange(node, 0, node, 0));
} else {
return(xmlXPtrNewRange(node, 0, node,
xmlStrlen(node->content)));
}
}
case XML_ATTRIBUTE_NODE:
case XML_ELEMENT_NODE:
case XML_ENTITY_REF_NODE:
case XML_DOCUMENT_NODE:
case XML_NOTATION_NODE:
case XML_HTML_DOCUMENT_NODE: {
return(xmlXPtrNewRange(node, 0, node,
xmlXPtrGetArity(node)));
}
default:
break;
}
return(NULL);
}
case XPATH_RANGE: {
xmlNodePtr node = (xmlNodePtr) loc->user;
if (loc->user2 != NULL) {
return(xmlXPtrNewRange(node, loc->index,
loc->user2, loc->index2));
} else {
switch (node->type) {
case XML_PI_NODE:
case XML_COMMENT_NODE:
case XML_TEXT_NODE:
case XML_CDATA_SECTION_NODE: {
if (node->content == NULL) {
return(xmlXPtrNewRange(node, 0, node, 0));
} else {
return(xmlXPtrNewRange(node, 0, node,
xmlStrlen(node->content)));
}
}
case XML_ATTRIBUTE_NODE:
case XML_ELEMENT_NODE:
case XML_ENTITY_REF_NODE:
case XML_DOCUMENT_NODE:
case XML_NOTATION_NODE:
case XML_HTML_DOCUMENT_NODE: {
return(xmlXPtrNewRange(node, 0, node,
xmlXPtrGetArity(node)));
}
default:
break;
}
return(NULL);
}
}
default:
TODO /* missed one case ??? */
}
return(NULL);
}
/**
* xmlXPtrRangeInsideFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Function implementing the range-inside() function 5.4.3
* location-set range-inside(location-set )
*
* The range-inside function returns ranges covering the contents of
* the locations in the argument location-set. For each location x in
* the argument location-set, a range location is added to the result
* location-set. If x is a range location, then x is added to the
* result location-set. If x is not a range location, then x is used
* as the container location of the start and end points of the range
* location to be added; the index of the start point of the range is
* zero; if the end point is a character point then its index is the
* length of the string-value of x, and otherwise is the number of
* location children of x.
*
*/
static void
xmlXPtrRangeInsideFunction(xmlXPathParserContextPtr ctxt, int nargs) {
int i;
xmlXPathObjectPtr set;
xmlLocationSetPtr oldset;
xmlLocationSetPtr newset;
CHECK_ARITY(1);
if ((ctxt->value == NULL) ||
((ctxt->value->type != XPATH_LOCATIONSET) &&
(ctxt->value->type != XPATH_NODESET)))
XP_ERROR(XPATH_INVALID_TYPE)
set = valuePop(ctxt);
if (set->type == XPATH_NODESET) {
xmlXPathObjectPtr tmp;
/*
* First convert to a location set
*/
tmp = xmlXPtrNewLocationSetNodeSet(set->nodesetval);
xmlXPathFreeObject(set);
set = tmp;
}
oldset = (xmlLocationSetPtr) set->user;
/*
* The loop is to compute the covering range for each item and add it
*/
newset = xmlXPtrLocationSetCreate(NULL);
for (i = 0;i < oldset->locNr;i++) {
xmlXPtrLocationSetAdd(newset,
xmlXPtrInsideRange(ctxt, oldset->locTab[i]));
}
/*
* Save the new value and cleanup
*/
valuePush(ctxt, xmlXPtrWrapLocationSet(newset));
xmlXPathFreeObject(set);
}
/**
* xmlXPtrRangeToFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Implement the range-to() XPointer function
*/
void
xmlXPtrRangeToFunction(xmlXPathParserContextPtr ctxt, int nargs) {
xmlXPathObjectPtr range;
const xmlChar *cur;
xmlXPathObjectPtr res, obj;
xmlXPathObjectPtr tmp;
xmlLocationSetPtr newset = NULL;
xmlNodeSetPtr oldset;
int i;
if (ctxt == NULL) return;
CHECK_ARITY(1);
/*
* Save the expression pointer since we will have to evaluate
* it multiple times. Initialize the new set.
*/
CHECK_TYPE(XPATH_NODESET);
obj = valuePop(ctxt);
oldset = obj->nodesetval;
ctxt->context->node = NULL;
cur = ctxt->cur;
newset = xmlXPtrLocationSetCreate(NULL);
for (i = 0; i < oldset->nodeNr; i++) {
ctxt->cur = cur;
/*
* Run the evaluation with a node list made of a single item
* in the nodeset.
*/
ctxt->context->node = oldset->nodeTab[i];
tmp = xmlXPathNewNodeSet(ctxt->context->node);
valuePush(ctxt, tmp);
xmlXPathEvalExpr(ctxt);
CHECK_ERROR;
/*
* The result of the evaluation need to be tested to
* decided whether the filter succeeded or not
*/
res = valuePop(ctxt);
range = xmlXPtrNewRangeNodeObject(oldset->nodeTab[i], res);
if (range != NULL) {
xmlXPtrLocationSetAdd(newset, range);
}
/*
* Cleanup
*/
if (res != NULL)
xmlXPathFreeObject(res);
if (ctxt->value == tmp) {
res = valuePop(ctxt);
xmlXPathFreeObject(res);
}
ctxt->context->node = NULL;
}
/*
* The result is used as the new evaluation set.
*/
xmlXPathFreeObject(obj);
ctxt->context->node = NULL;
ctxt->context->contextSize = -1;
ctxt->context->proximityPosition = -1;
valuePush(ctxt, xmlXPtrWrapLocationSet(newset));
}
/**
* xmlXPtrAdvanceNode:
* @cur: the node
* @level: incremented/decremented to show level in tree
*
* Advance to the next element or text node in document order
* TODO: add a stack for entering/exiting entities
*
* Returns -1 in case of failure, 0 otherwise
*/
xmlNodePtr
xmlXPtrAdvanceNode(xmlNodePtr cur, int *level) {
next:
if (cur == NULL)
return(NULL);
if (cur->children != NULL) {
cur = cur->children ;
if (level != NULL)
(*level)++;
goto found;
}
skip: /* This label should only be needed if something is wrong! */
if (cur->next != NULL) {
cur = cur->next;
goto found;
}
do {
cur = cur->parent;
if (level != NULL)
(*level)--;
if (cur == NULL) return(NULL);
if (cur->next != NULL) {
cur = cur->next;
goto found;
}
} while (cur != NULL);
found:
if ((cur->type != XML_ELEMENT_NODE) &&
(cur->type != XML_TEXT_NODE) &&
(cur->type != XML_DOCUMENT_NODE) &&
(cur->type != XML_HTML_DOCUMENT_NODE) &&
(cur->type != XML_CDATA_SECTION_NODE)) {
if (cur->type == XML_ENTITY_REF_NODE) { /* Shouldn't happen */
TODO
goto skip;
}
goto next;
}
return(cur);
}
/**
* xmlXPtrAdvanceChar:
* @node: the node
* @indx: the indx
* @bytes: the number of bytes
*
* Advance a point of the associated number of bytes (not UTF8 chars)
*
* Returns -1 in case of failure, 0 otherwise
*/
static int
xmlXPtrAdvanceChar(xmlNodePtr *node, int *indx, int bytes) {
xmlNodePtr cur;
int pos;
int len;
if ((node == NULL) || (indx == NULL))
return(-1);
cur = *node;
if (cur == NULL)
return(-1);
pos = *indx;
while (bytes >= 0) {
/*
* First position to the beginning of the first text node
* corresponding to this point
*/
while ((cur != NULL) &&
((cur->type == XML_ELEMENT_NODE) ||
(cur->type == XML_DOCUMENT_NODE) ||
(cur->type == XML_HTML_DOCUMENT_NODE))) {
if (pos > 0) {
cur = xmlXPtrGetNthChild(cur, pos);
pos = 0;
} else {
cur = xmlXPtrAdvanceNode(cur, NULL);
pos = 0;
}
}
if (cur == NULL) {
*node = NULL;
*indx = 0;
return(-1);
}
/*
* if there is no move needed return the current value.
*/
if (pos == 0) pos = 1;
if (bytes == 0) {
*node = cur;
*indx = pos;
return(0);
}
/*
* We should have a text (or cdata) node ...
*/
len = 0;
if ((cur->type != XML_ELEMENT_NODE) &&
(cur->content != NULL)) {
len = xmlStrlen(cur->content);
}
if (pos > len) {
/* Strange, the indx in the text node is greater than it's len */
STRANGE
pos = len;
}
if (pos + bytes >= len) {
bytes -= (len - pos);
cur = xmlXPtrAdvanceNode(cur, NULL);
pos = 0;
} else if (pos + bytes < len) {
pos += bytes;
*node = cur;
*indx = pos;
return(0);
}
}
return(-1);
}
/**
* xmlXPtrMatchString:
* @string: the string to search
* @start: the start textnode
* @startindex: the start index
* @end: the end textnode IN/OUT
* @endindex: the end index IN/OUT
*
* Check whether the document contains @string at the position
* (@start, @startindex) and limited by the (@end, @endindex) point
*
* Returns -1 in case of failure, 0 if not found, 1 if found in which case
* (@start, @startindex) will indicate the position of the beginning
* of the range and (@end, @endindex) will indicate the end
* of the range
*/
static int
xmlXPtrMatchString(const xmlChar *string, xmlNodePtr start, int startindex,
xmlNodePtr *end, int *endindex) {
xmlNodePtr cur;
int pos; /* 0 based */
int len; /* in bytes */
int stringlen; /* in bytes */
int match;
if (string == NULL)
return(-1);
if (start == NULL)
return(-1);
if ((end == NULL) || (endindex == NULL))
return(-1);
cur = start;
if (cur == NULL)
return(-1);
pos = startindex - 1;
stringlen = xmlStrlen(string);
while (stringlen > 0) {
if ((cur == *end) && (pos + stringlen > *endindex))
return(0);
if ((cur->type != XML_ELEMENT_NODE) && (cur->content != NULL)) {
len = xmlStrlen(cur->content);
if (len >= pos + stringlen) {
match = (!xmlStrncmp(&cur->content[pos], string, stringlen));
if (match) {
#ifdef DEBUG_RANGES
xmlGenericError(xmlGenericErrorContext,
"found range %d bytes at index %d of ->",
stringlen, pos + 1);
xmlDebugDumpString(stdout, cur->content);
xmlGenericError(xmlGenericErrorContext, "\n");
#endif
*end = cur;
*endindex = pos + stringlen;
return(1);
} else {
return(0);
}
} else {
int sub = len - pos;
match = (!xmlStrncmp(&cur->content[pos], string, sub));
if (match) {
#ifdef DEBUG_RANGES
xmlGenericError(xmlGenericErrorContext,
"found subrange %d bytes at index %d of ->",
sub, pos + 1);
xmlDebugDumpString(stdout, cur->content);
xmlGenericError(xmlGenericErrorContext, "\n");
#endif
string = &string[sub];
stringlen -= sub;
} else {
return(0);
}
}
}
cur = xmlXPtrAdvanceNode(cur, NULL);
if (cur == NULL)
return(0);
pos = 0;
}
return(1);
}
/**
* xmlXPtrSearchString:
* @string: the string to search
* @start: the start textnode IN/OUT
* @startindex: the start index IN/OUT
* @end: the end textnode
* @endindex: the end index
*
* Search the next occurrence of @string within the document content
* until the (@end, @endindex) point is reached
*
* Returns -1 in case of failure, 0 if not found, 1 if found in which case
* (@start, @startindex) will indicate the position of the beginning
* of the range and (@end, @endindex) will indicate the end
* of the range
*/
static int
xmlXPtrSearchString(const xmlChar *string, xmlNodePtr *start, int *startindex,
xmlNodePtr *end, int *endindex) {
xmlNodePtr cur;
const xmlChar *str;
int pos; /* 0 based */
int len; /* in bytes */
xmlChar first;
if (string == NULL)
return(-1);
if ((start == NULL) || (startindex == NULL))
return(-1);
if ((end == NULL) || (endindex == NULL))
return(-1);
cur = *start;
if (cur == NULL)
return(-1);
pos = *startindex - 1;
first = string[0];
while (cur != NULL) {
if ((cur->type != XML_ELEMENT_NODE) && (cur->content != NULL)) {
len = xmlStrlen(cur->content);
while (pos <= len) {
if (first != 0) {
str = xmlStrchr(&cur->content[pos], first);
if (str != NULL) {
pos = (str - (xmlChar *)(cur->content));
#ifdef DEBUG_RANGES
xmlGenericError(xmlGenericErrorContext,
"found '%c' at index %d of ->",
first, pos + 1);
xmlDebugDumpString(stdout, cur->content);
xmlGenericError(xmlGenericErrorContext, "\n");
#endif
if (xmlXPtrMatchString(string, cur, pos + 1,
end, endindex)) {
*start = cur;
*startindex = pos + 1;
return(1);
}
pos++;
} else {
pos = len + 1;
}
} else {
/*
* An empty string is considered to match before each
* character of the string-value and after the final
* character.
*/
#ifdef DEBUG_RANGES
xmlGenericError(xmlGenericErrorContext,
"found '' at index %d of ->",
pos + 1);
xmlDebugDumpString(stdout, cur->content);
xmlGenericError(xmlGenericErrorContext, "\n");
#endif
*start = cur;
*startindex = pos + 1;
*end = cur;
*endindex = pos + 1;
return(1);
}
}
}
if ((cur == *end) && (pos >= *endindex))
return(0);
cur = xmlXPtrAdvanceNode(cur, NULL);
if (cur == NULL)
return(0);
pos = 1;
}
return(0);
}
/**
* xmlXPtrGetLastChar:
* @node: the node
* @index: the index
*
* Computes the point coordinates of the last char of this point
*
* Returns -1 in case of failure, 0 otherwise
*/
static int
xmlXPtrGetLastChar(xmlNodePtr *node, int *indx) {
xmlNodePtr cur;
int pos, len = 0;
if ((node == NULL) || (indx == NULL))
return(-1);
cur = *node;
pos = *indx;
if (cur == NULL)
return(-1);
if ((cur->type == XML_ELEMENT_NODE) ||
(cur->type == XML_DOCUMENT_NODE) ||
(cur->type == XML_HTML_DOCUMENT_NODE)) {
if (pos > 0) {
cur = xmlXPtrGetNthChild(cur, pos);
}
}
while (cur != NULL) {
if (cur->last != NULL)
cur = cur->last;
else if ((cur->type != XML_ELEMENT_NODE) &&
(cur->content != NULL)) {
len = xmlStrlen(cur->content);
break;
} else {
return(-1);
}
}
if (cur == NULL)
return(-1);
*node = cur;
*indx = len;
return(0);
}
/**
* xmlXPtrGetStartPoint:
* @obj: an range
* @node: the resulting node
* @indx: the resulting index
*
* read the object and return the start point coordinates.
*
* Returns -1 in case of failure, 0 otherwise
*/
static int
xmlXPtrGetStartPoint(xmlXPathObjectPtr obj, xmlNodePtr *node, int *indx) {
if ((obj == NULL) || (node == NULL) || (indx == NULL))
return(-1);
switch (obj->type) {
case XPATH_POINT:
*node = obj->user;
if (obj->index <= 0)
*indx = 0;
else
*indx = obj->index;
return(0);
case XPATH_RANGE:
*node = obj->user;
if (obj->index <= 0)
*indx = 0;
else
*indx = obj->index;
return(0);
default:
break;
}
return(-1);
}
/**
* xmlXPtrGetEndPoint:
* @obj: an range
* @node: the resulting node
* @indx: the resulting indx
*
* read the object and return the end point coordinates.
*
* Returns -1 in case of failure, 0 otherwise
*/
static int
xmlXPtrGetEndPoint(xmlXPathObjectPtr obj, xmlNodePtr *node, int *indx) {
if ((obj == NULL) || (node == NULL) || (indx == NULL))
return(-1);
switch (obj->type) {
case XPATH_POINT:
*node = obj->user;
if (obj->index <= 0)
*indx = 0;
else
*indx = obj->index;
return(0);
case XPATH_RANGE:
*node = obj->user;
if (obj->index <= 0)
*indx = 0;
else
*indx = obj->index;
return(0);
default:
break;
}
return(-1);
}
/**
* xmlXPtrStringRangeFunction:
* @ctxt: the XPointer Parser context
* @nargs: the number of args
*
* Function implementing the string-range() function
* range as described in 5.4.2
*
* ------------------------------
* [Definition: For each location in the location-set argument,
* string-range returns a set of string ranges, a set of substrings in a
* string. Specifically, the string-value of the location is searched for
* substrings that match the string argument, and the resulting location-set
* will contain a range location for each non-overlapping match.]
* An empty string is considered to match before each character of the
* string-value and after the final character. Whitespace in a string
* is matched literally, with no normalization except that provided by
* XML for line ends. The third argument gives the position of the first
* character to be in the resulting range, relative to the start of the
* match. The default value is 1, which makes the range start immediately
* before the first character of the matched string. The fourth argument
* gives the number of characters in the range; the default is that the
* range extends to the end of the matched string.
*
* Element boundaries, as well as entire embedded nodes such as processing
* instructions and comments, are ignored as defined in [XPath].
*
* If the string in the second argument is not found in the string-value
* of the location, or if a value in the third or fourth argument indicates
* a string that is beyond the beginning or end of the document, the
* expression fails.
*
* The points of the range-locations in the returned location-set will
* all be character points.
* ------------------------------
*/
static void
xmlXPtrStringRangeFunction(xmlXPathParserContextPtr ctxt, int nargs) {
int i, startindex, endindex = 0, fendindex;
xmlNodePtr start, end = 0, fend;
xmlXPathObjectPtr set;
xmlLocationSetPtr oldset;
xmlLocationSetPtr newset;
xmlXPathObjectPtr string;
xmlXPathObjectPtr position = NULL;
xmlXPathObjectPtr number = NULL;
int found, pos = 0, num = 0;
/*
* Grab the arguments
*/
if ((nargs < 2) || (nargs > 4))
XP_ERROR(XPATH_INVALID_ARITY);
if (nargs >= 4) {
CHECK_TYPE(XPATH_NUMBER);
number = valuePop(ctxt);
if (number != NULL)
num = (int) number->floatval;
}
if (nargs >= 3) {
CHECK_TYPE(XPATH_NUMBER);
position = valuePop(ctxt);
if (position != NULL)
pos = (int) position->floatval;
}
CHECK_TYPE(XPATH_STRING);
string = valuePop(ctxt);
if ((ctxt->value == NULL) ||
((ctxt->value->type != XPATH_LOCATIONSET) &&
(ctxt->value->type != XPATH_NODESET)))
XP_ERROR(XPATH_INVALID_TYPE)
set = valuePop(ctxt);
newset = xmlXPtrLocationSetCreate(NULL);
if (set->nodesetval == NULL) {
goto error;
}
if (set->type == XPATH_NODESET) {
xmlXPathObjectPtr tmp;
/*
* First convert to a location set
*/
tmp = xmlXPtrNewLocationSetNodeSet(set->nodesetval);
xmlXPathFreeObject(set);
set = tmp;
}
oldset = (xmlLocationSetPtr) set->user;
/*
* The loop is to search for each element in the location set
* the list of location set corresponding to that search
*/
for (i = 0;i < oldset->locNr;i++) {
#ifdef DEBUG_RANGES
xmlXPathDebugDumpObject(stdout, oldset->locTab[i], 0);
#endif
xmlXPtrGetStartPoint(oldset->locTab[i], &start, &startindex);
xmlXPtrGetEndPoint(oldset->locTab[i], &end, &endindex);
xmlXPtrAdvanceChar(&start, &startindex, 0);
xmlXPtrGetLastChar(&end, &endindex);
#ifdef DEBUG_RANGES
xmlGenericError(xmlGenericErrorContext,
"from index %d of ->", startindex);
xmlDebugDumpString(stdout, start->content);
xmlGenericError(xmlGenericErrorContext, "\n");
xmlGenericError(xmlGenericErrorContext,
"to index %d of ->", endindex);
xmlDebugDumpString(stdout, end->content);
xmlGenericError(xmlGenericErrorContext, "\n");
#endif
do {
fend = end;
fendindex = endindex;
found = xmlXPtrSearchString(string->stringval, &start, &startindex,
&fend, &fendindex);
if (found == 1) {
if (position == NULL) {
xmlXPtrLocationSetAdd(newset,
xmlXPtrNewRange(start, startindex, fend, fendindex));
} else if (xmlXPtrAdvanceChar(&start, &startindex,
pos - 1) == 0) {
if ((number != NULL) && (num > 0)) {
int rindx;
xmlNodePtr rend;
rend = start;
rindx = startindex - 1;
if (xmlXPtrAdvanceChar(&rend, &rindx,
num) == 0) {
xmlXPtrLocationSetAdd(newset,
xmlXPtrNewRange(start, startindex,
rend, rindx));
}
} else if ((number != NULL) && (num <= 0)) {
xmlXPtrLocationSetAdd(newset,
xmlXPtrNewRange(start, startindex,
start, startindex));
} else {
xmlXPtrLocationSetAdd(newset,
xmlXPtrNewRange(start, startindex,
fend, fendindex));
}
}
start = fend;
startindex = fendindex;
if (string->stringval[0] == 0)
startindex++;
}
} while (found == 1);
}
/*
* Save the new value and cleanup
*/
error:
valuePush(ctxt, xmlXPtrWrapLocationSet(newset));
xmlXPathFreeObject(set);
xmlXPathFreeObject(string);
if (position) xmlXPathFreeObject(position);
if (number) xmlXPathFreeObject(number);
}
/**
* xmlXPtrEvalRangePredicate:
* @ctxt: the XPointer Parser context
*
* [8] Predicate ::= '[' PredicateExpr ']'
* [9] PredicateExpr ::= Expr
*
* Evaluate a predicate as in xmlXPathEvalPredicate() but for
* a Location Set instead of a node set
*/
void
xmlXPtrEvalRangePredicate(xmlXPathParserContextPtr ctxt) {
const xmlChar *cur;
xmlXPathObjectPtr res;
xmlXPathObjectPtr obj, tmp;
xmlLocationSetPtr newset = NULL;
xmlLocationSetPtr oldset;
int i;
if (ctxt == NULL) return;
SKIP_BLANKS;
if (CUR != '[') {
XP_ERROR(XPATH_INVALID_PREDICATE_ERROR);
}
NEXT;
SKIP_BLANKS;
/*
* Extract the old set, and then evaluate the result of the
* expression for all the element in the set. use it to grow
* up a new set.
*/
CHECK_TYPE(XPATH_LOCATIONSET);
obj = valuePop(ctxt);
oldset = obj->user;
ctxt->context->node = NULL;
if ((oldset == NULL) || (oldset->locNr == 0)) {
ctxt->context->contextSize = 0;
ctxt->context->proximityPosition = 0;
xmlXPathEvalExpr(ctxt);
res = valuePop(ctxt);
if (res != NULL)
xmlXPathFreeObject(res);
valuePush(ctxt, obj);
CHECK_ERROR;
} else {
/*
* Save the expression pointer since we will have to evaluate
* it multiple times. Initialize the new set.
*/
cur = ctxt->cur;
newset = xmlXPtrLocationSetCreate(NULL);
for (i = 0; i < oldset->locNr; i++) {
ctxt->cur = cur;
/*
* Run the evaluation with a node list made of a single item
* in the nodeset.
*/
ctxt->context->node = oldset->locTab[i]->user;
tmp = xmlXPathNewNodeSet(ctxt->context->node);
valuePush(ctxt, tmp);
ctxt->context->contextSize = oldset->locNr;
ctxt->context->proximityPosition = i + 1;
xmlXPathEvalExpr(ctxt);
CHECK_ERROR;
/*
* The result of the evaluation need to be tested to
* decided whether the filter succeeded or not
*/
res = valuePop(ctxt);
if (xmlXPathEvaluatePredicateResult(ctxt, res)) {
xmlXPtrLocationSetAdd(newset,
xmlXPathObjectCopy(oldset->locTab[i]));
}
/*
* Cleanup
*/
if (res != NULL)
xmlXPathFreeObject(res);
if (ctxt->value == tmp) {
res = valuePop(ctxt);
xmlXPathFreeObject(res);
}
ctxt->context->node = NULL;
}
/*
* The result is used as the new evaluation set.
*/
xmlXPathFreeObject(obj);
ctxt->context->node = NULL;
ctxt->context->contextSize = -1;
ctxt->context->proximityPosition = -1;
valuePush(ctxt, xmlXPtrWrapLocationSet(newset));
}
if (CUR != ']') {
XP_ERROR(XPATH_INVALID_PREDICATE_ERROR);
}
NEXT;
SKIP_BLANKS;
}
#define bottom_xpointer
#include "elfgcchack.h"
#endif