wine/libs/xml2/xmlschemastypes.c
2023-06-05 14:01:07 +02:00

6439 lines
181 KiB
C

/*
* schemastypes.c : implementation of the XML Schema Datatypes
* definition and validity checking
*
* See Copyright for the status of this software.
*
* Daniel Veillard <veillard@redhat.com>
*/
/* To avoid EBCDIC trouble when parsing on zOS */
#if defined(__MVS__)
#pragma convert("ISO8859-1")
#endif
#define IN_LIBXML
#include "libxml.h"
#ifdef LIBXML_SCHEMAS_ENABLED
#include <string.h>
#include <math.h>
#include <float.h>
#include <libxml/xmlmemory.h>
#include <libxml/parser.h>
#include <libxml/parserInternals.h>
#include <libxml/hash.h>
#include <libxml/valid.h>
#include <libxml/xpath.h>
#include <libxml/uri.h>
#include <libxml/xmlschemas.h>
#include <libxml/schemasInternals.h>
#include <libxml/xmlschemastypes.h>
#include "private/error.h"
#define DEBUG
#ifndef LIBXML_XPATH_ENABLED
extern double xmlXPathNAN;
extern double xmlXPathPINF;
extern double xmlXPathNINF;
#endif
#define TODO \
xmlGenericError(xmlGenericErrorContext, \
"Unimplemented block at %s:%d\n", \
__FILE__, __LINE__);
#define XML_SCHEMAS_NAMESPACE_NAME \
(const xmlChar *)"http://www.w3.org/2001/XMLSchema"
#define IS_WSP_REPLACE_CH(c) ((((c) == 0x9) || ((c) == 0xa)) || \
((c) == 0xd))
#define IS_WSP_SPACE_CH(c) ((c) == 0x20)
#define IS_WSP_BLANK_CH(c) IS_BLANK_CH(c)
/* Date value */
typedef struct _xmlSchemaValDate xmlSchemaValDate;
typedef xmlSchemaValDate *xmlSchemaValDatePtr;
struct _xmlSchemaValDate {
long year;
unsigned int mon :4; /* 1 <= mon <= 12 */
unsigned int day :5; /* 1 <= day <= 31 */
unsigned int hour :5; /* 0 <= hour <= 24 */
unsigned int min :6; /* 0 <= min <= 59 */
double sec;
unsigned int tz_flag :1; /* is tzo explicitly set? */
signed int tzo :12; /* -1440 <= tzo <= 1440;
currently only -840 to +840 are needed */
};
/* Duration value */
typedef struct _xmlSchemaValDuration xmlSchemaValDuration;
typedef xmlSchemaValDuration *xmlSchemaValDurationPtr;
struct _xmlSchemaValDuration {
long mon; /* mon stores years also */
long day;
double sec; /* sec stores min and hour also */
};
typedef struct _xmlSchemaValDecimal xmlSchemaValDecimal;
typedef xmlSchemaValDecimal *xmlSchemaValDecimalPtr;
struct _xmlSchemaValDecimal {
/* would use long long but not portable */
unsigned long lo;
unsigned long mi;
unsigned long hi;
unsigned int extra;
unsigned int sign:1;
unsigned int frac:7;
unsigned int total:8;
};
typedef struct _xmlSchemaValQName xmlSchemaValQName;
typedef xmlSchemaValQName *xmlSchemaValQNamePtr;
struct _xmlSchemaValQName {
xmlChar *name;
xmlChar *uri;
};
typedef struct _xmlSchemaValHex xmlSchemaValHex;
typedef xmlSchemaValHex *xmlSchemaValHexPtr;
struct _xmlSchemaValHex {
xmlChar *str;
unsigned int total;
};
typedef struct _xmlSchemaValBase64 xmlSchemaValBase64;
typedef xmlSchemaValBase64 *xmlSchemaValBase64Ptr;
struct _xmlSchemaValBase64 {
xmlChar *str;
unsigned int total;
};
struct _xmlSchemaVal {
xmlSchemaValType type;
struct _xmlSchemaVal *next;
union {
xmlSchemaValDecimal decimal;
xmlSchemaValDate date;
xmlSchemaValDuration dur;
xmlSchemaValQName qname;
xmlSchemaValHex hex;
xmlSchemaValBase64 base64;
float f;
double d;
int b;
xmlChar *str;
} value;
};
static int xmlSchemaTypesInitialized = 0;
static xmlHashTablePtr xmlSchemaTypesBank = NULL;
/*
* Basic types
*/
static xmlSchemaTypePtr xmlSchemaTypeStringDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeAnyTypeDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeAnySimpleTypeDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeDecimalDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeDatetimeDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeDateDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeTimeDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeGYearDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeGYearMonthDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeGDayDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeGMonthDayDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeGMonthDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeDurationDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeFloatDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeBooleanDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeDoubleDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeHexBinaryDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeBase64BinaryDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeAnyURIDef = NULL;
/*
* Derived types
*/
static xmlSchemaTypePtr xmlSchemaTypePositiveIntegerDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNonPositiveIntegerDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNegativeIntegerDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNonNegativeIntegerDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeIntegerDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeLongDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeIntDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeShortDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeByteDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeUnsignedLongDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeUnsignedIntDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeUnsignedShortDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeUnsignedByteDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNormStringDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeTokenDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeLanguageDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNameDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeQNameDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNCNameDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeIdDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeIdrefDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeIdrefsDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeEntityDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeEntitiesDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNotationDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNmtokenDef = NULL;
static xmlSchemaTypePtr xmlSchemaTypeNmtokensDef = NULL;
/************************************************************************
* *
* Datatype error handlers *
* *
************************************************************************/
/**
* xmlSchemaTypeErrMemory:
* @extra: extra information
*
* Handle an out of memory condition
*/
static void
xmlSchemaTypeErrMemory(xmlNodePtr node, const char *extra)
{
__xmlSimpleError(XML_FROM_DATATYPE, XML_ERR_NO_MEMORY, node, NULL, extra);
}
/************************************************************************
* *
* Base types support *
* *
************************************************************************/
/**
* xmlSchemaNewValue:
* @type: the value type
*
* Allocate a new simple type value
*
* Returns a pointer to the new value or NULL in case of error
*/
static xmlSchemaValPtr
xmlSchemaNewValue(xmlSchemaValType type) {
xmlSchemaValPtr value;
value = (xmlSchemaValPtr) xmlMalloc(sizeof(xmlSchemaVal));
if (value == NULL) {
return(NULL);
}
memset(value, 0, sizeof(xmlSchemaVal));
value->type = type;
return(value);
}
static xmlSchemaFacetPtr
xmlSchemaNewMinLengthFacet(int value)
{
xmlSchemaFacetPtr ret;
ret = xmlSchemaNewFacet();
if (ret == NULL) {
return(NULL);
}
ret->type = XML_SCHEMA_FACET_MINLENGTH;
ret->val = xmlSchemaNewValue(XML_SCHEMAS_NNINTEGER);
if (ret->val == NULL) {
xmlFree(ret);
return(NULL);
}
ret->val->value.decimal.lo = value;
return (ret);
}
/*
* xmlSchemaInitBasicType:
* @name: the type name
* @type: the value type associated
*
* Initialize one primitive built-in type
*/
static xmlSchemaTypePtr
xmlSchemaInitBasicType(const char *name, xmlSchemaValType type,
xmlSchemaTypePtr baseType) {
xmlSchemaTypePtr ret;
ret = (xmlSchemaTypePtr) xmlMalloc(sizeof(xmlSchemaType));
if (ret == NULL) {
xmlSchemaTypeErrMemory(NULL, "could not initialize basic types");
return(NULL);
}
memset(ret, 0, sizeof(xmlSchemaType));
ret->name = (const xmlChar *)name;
ret->targetNamespace = XML_SCHEMAS_NAMESPACE_NAME;
ret->type = XML_SCHEMA_TYPE_BASIC;
ret->baseType = baseType;
ret->contentType = XML_SCHEMA_CONTENT_BASIC;
/*
* Primitive types.
*/
switch (type) {
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_DECIMAL:
case XML_SCHEMAS_DATE:
case XML_SCHEMAS_DATETIME:
case XML_SCHEMAS_TIME:
case XML_SCHEMAS_GYEAR:
case XML_SCHEMAS_GYEARMONTH:
case XML_SCHEMAS_GMONTH:
case XML_SCHEMAS_GMONTHDAY:
case XML_SCHEMAS_GDAY:
case XML_SCHEMAS_DURATION:
case XML_SCHEMAS_FLOAT:
case XML_SCHEMAS_DOUBLE:
case XML_SCHEMAS_BOOLEAN:
case XML_SCHEMAS_ANYURI:
case XML_SCHEMAS_HEXBINARY:
case XML_SCHEMAS_BASE64BINARY:
case XML_SCHEMAS_QNAME:
case XML_SCHEMAS_NOTATION:
ret->flags |= XML_SCHEMAS_TYPE_BUILTIN_PRIMITIVE;
break;
default:
break;
}
/*
* Set variety.
*/
switch (type) {
case XML_SCHEMAS_ANYTYPE:
case XML_SCHEMAS_ANYSIMPLETYPE:
break;
case XML_SCHEMAS_IDREFS:
case XML_SCHEMAS_NMTOKENS:
case XML_SCHEMAS_ENTITIES:
ret->flags |= XML_SCHEMAS_TYPE_VARIETY_LIST;
ret->facets = xmlSchemaNewMinLengthFacet(1);
ret->flags |= XML_SCHEMAS_TYPE_HAS_FACETS;
break;
default:
ret->flags |= XML_SCHEMAS_TYPE_VARIETY_ATOMIC;
break;
}
xmlHashAddEntry2(xmlSchemaTypesBank, ret->name,
XML_SCHEMAS_NAMESPACE_NAME, ret);
ret->builtInType = type;
return(ret);
}
/*
* WARNING: Those type reside normally in xmlschemas.c but are
* redefined here locally in oder of being able to use them for xs:anyType-
* TODO: Remove those definition if we move the types to a header file.
* TODO: Always keep those structs up-to-date with the originals.
*/
#define UNBOUNDED (1 << 30)
typedef struct _xmlSchemaTreeItem xmlSchemaTreeItem;
typedef xmlSchemaTreeItem *xmlSchemaTreeItemPtr;
struct _xmlSchemaTreeItem {
xmlSchemaTypeType type;
xmlSchemaAnnotPtr annot;
xmlSchemaTreeItemPtr next;
xmlSchemaTreeItemPtr children;
};
typedef struct _xmlSchemaParticle xmlSchemaParticle;
typedef xmlSchemaParticle *xmlSchemaParticlePtr;
struct _xmlSchemaParticle {
xmlSchemaTypeType type;
xmlSchemaAnnotPtr annot;
xmlSchemaTreeItemPtr next;
xmlSchemaTreeItemPtr children;
int minOccurs;
int maxOccurs;
xmlNodePtr node;
};
typedef struct _xmlSchemaModelGroup xmlSchemaModelGroup;
typedef xmlSchemaModelGroup *xmlSchemaModelGroupPtr;
struct _xmlSchemaModelGroup {
xmlSchemaTypeType type;
xmlSchemaAnnotPtr annot;
xmlSchemaTreeItemPtr next;
xmlSchemaTreeItemPtr children;
xmlNodePtr node;
};
static xmlSchemaParticlePtr
xmlSchemaAddParticle(void)
{
xmlSchemaParticlePtr ret = NULL;
ret = (xmlSchemaParticlePtr)
xmlMalloc(sizeof(xmlSchemaParticle));
if (ret == NULL) {
xmlSchemaTypeErrMemory(NULL, "allocating particle component");
return (NULL);
}
memset(ret, 0, sizeof(xmlSchemaParticle));
ret->type = XML_SCHEMA_TYPE_PARTICLE;
ret->minOccurs = 1;
ret->maxOccurs = 1;
return (ret);
}
static void
xmlSchemaFreeTypeEntry(void *type, const xmlChar *name ATTRIBUTE_UNUSED) {
xmlSchemaFreeType((xmlSchemaTypePtr) type);
}
/**
* xmlSchemaCleanupTypesInternal:
*
* Cleanup the default XML Schemas type library
*/
static void
xmlSchemaCleanupTypesInternal(void) {
xmlSchemaParticlePtr particle;
/*
* Free xs:anyType.
*/
if (xmlSchemaTypeAnyTypeDef != NULL) {
/* Attribute wildcard. */
xmlSchemaFreeWildcard(xmlSchemaTypeAnyTypeDef->attributeWildcard);
/* Content type. */
particle = (xmlSchemaParticlePtr) xmlSchemaTypeAnyTypeDef->subtypes;
/* Wildcard. */
xmlSchemaFreeWildcard((xmlSchemaWildcardPtr)
particle->children->children->children);
xmlFree((xmlSchemaParticlePtr) particle->children->children);
/* Sequence model group. */
xmlFree((xmlSchemaModelGroupPtr) particle->children);
xmlFree((xmlSchemaParticlePtr) particle);
xmlSchemaTypeAnyTypeDef->subtypes = NULL;
xmlSchemaTypeAnyTypeDef = NULL;
}
xmlHashFree(xmlSchemaTypesBank, xmlSchemaFreeTypeEntry);
xmlSchemaTypesBank = NULL;
/* Note that the xmlSchemaType*Def pointers aren't set to NULL. */
}
/*
* xmlSchemaInitTypes:
*
* Initialize the default XML Schemas type library
*
* Returns 0 on success, -1 on error.
*/
int
xmlSchemaInitTypes(void)
{
if (xmlSchemaTypesInitialized != 0)
return (0);
xmlSchemaTypesBank = xmlHashCreate(40);
if (xmlSchemaTypesBank == NULL) {
xmlSchemaTypeErrMemory(NULL, NULL);
goto error;
}
/*
* 3.4.7 Built-in Complex Type Definition
*/
xmlSchemaTypeAnyTypeDef = xmlSchemaInitBasicType("anyType",
XML_SCHEMAS_ANYTYPE,
NULL);
if (xmlSchemaTypeAnyTypeDef == NULL)
goto error;
xmlSchemaTypeAnyTypeDef->baseType = xmlSchemaTypeAnyTypeDef;
xmlSchemaTypeAnyTypeDef->contentType = XML_SCHEMA_CONTENT_MIXED;
/*
* Init the content type.
*/
xmlSchemaTypeAnyTypeDef->contentType = XML_SCHEMA_CONTENT_MIXED;
{
xmlSchemaParticlePtr particle;
xmlSchemaModelGroupPtr sequence;
xmlSchemaWildcardPtr wild;
/* First particle. */
particle = xmlSchemaAddParticle();
if (particle == NULL)
goto error;
xmlSchemaTypeAnyTypeDef->subtypes = (xmlSchemaTypePtr) particle;
/* Sequence model group. */
sequence = (xmlSchemaModelGroupPtr)
xmlMalloc(sizeof(xmlSchemaModelGroup));
if (sequence == NULL) {
xmlSchemaTypeErrMemory(NULL, "allocating model group component");
goto error;
}
memset(sequence, 0, sizeof(xmlSchemaModelGroup));
sequence->type = XML_SCHEMA_TYPE_SEQUENCE;
particle->children = (xmlSchemaTreeItemPtr) sequence;
/* Second particle. */
particle = xmlSchemaAddParticle();
if (particle == NULL)
goto error;
particle->minOccurs = 0;
particle->maxOccurs = UNBOUNDED;
sequence->children = (xmlSchemaTreeItemPtr) particle;
/* The wildcard */
wild = (xmlSchemaWildcardPtr) xmlMalloc(sizeof(xmlSchemaWildcard));
if (wild == NULL) {
xmlSchemaTypeErrMemory(NULL, "allocating wildcard component");
goto error;
}
memset(wild, 0, sizeof(xmlSchemaWildcard));
wild->type = XML_SCHEMA_TYPE_ANY;
wild->any = 1;
wild->processContents = XML_SCHEMAS_ANY_LAX;
particle->children = (xmlSchemaTreeItemPtr) wild;
/*
* Create the attribute wildcard.
*/
wild = (xmlSchemaWildcardPtr) xmlMalloc(sizeof(xmlSchemaWildcard));
if (wild == NULL) {
xmlSchemaTypeErrMemory(NULL, "could not create an attribute "
"wildcard on anyType");
goto error;
}
memset(wild, 0, sizeof(xmlSchemaWildcard));
wild->any = 1;
wild->processContents = XML_SCHEMAS_ANY_LAX;
xmlSchemaTypeAnyTypeDef->attributeWildcard = wild;
}
xmlSchemaTypeAnySimpleTypeDef = xmlSchemaInitBasicType("anySimpleType",
XML_SCHEMAS_ANYSIMPLETYPE,
xmlSchemaTypeAnyTypeDef);
if (xmlSchemaTypeAnySimpleTypeDef == NULL)
goto error;
/*
* primitive datatypes
*/
xmlSchemaTypeStringDef = xmlSchemaInitBasicType("string",
XML_SCHEMAS_STRING,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeStringDef == NULL)
goto error;
xmlSchemaTypeDecimalDef = xmlSchemaInitBasicType("decimal",
XML_SCHEMAS_DECIMAL,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeDecimalDef == NULL)
goto error;
xmlSchemaTypeDateDef = xmlSchemaInitBasicType("date",
XML_SCHEMAS_DATE,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeDateDef == NULL)
goto error;
xmlSchemaTypeDatetimeDef = xmlSchemaInitBasicType("dateTime",
XML_SCHEMAS_DATETIME,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeDatetimeDef == NULL)
goto error;
xmlSchemaTypeTimeDef = xmlSchemaInitBasicType("time",
XML_SCHEMAS_TIME,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeTimeDef == NULL)
goto error;
xmlSchemaTypeGYearDef = xmlSchemaInitBasicType("gYear",
XML_SCHEMAS_GYEAR,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeGYearDef == NULL)
goto error;
xmlSchemaTypeGYearMonthDef = xmlSchemaInitBasicType("gYearMonth",
XML_SCHEMAS_GYEARMONTH,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeGYearMonthDef == NULL)
goto error;
xmlSchemaTypeGMonthDef = xmlSchemaInitBasicType("gMonth",
XML_SCHEMAS_GMONTH,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeGMonthDef == NULL)
goto error;
xmlSchemaTypeGMonthDayDef = xmlSchemaInitBasicType("gMonthDay",
XML_SCHEMAS_GMONTHDAY,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeGMonthDayDef == NULL)
goto error;
xmlSchemaTypeGDayDef = xmlSchemaInitBasicType("gDay",
XML_SCHEMAS_GDAY,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeGDayDef == NULL)
goto error;
xmlSchemaTypeDurationDef = xmlSchemaInitBasicType("duration",
XML_SCHEMAS_DURATION,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeDurationDef == NULL)
goto error;
xmlSchemaTypeFloatDef = xmlSchemaInitBasicType("float",
XML_SCHEMAS_FLOAT,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeFloatDef == NULL)
goto error;
xmlSchemaTypeDoubleDef = xmlSchemaInitBasicType("double",
XML_SCHEMAS_DOUBLE,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeDoubleDef == NULL)
goto error;
xmlSchemaTypeBooleanDef = xmlSchemaInitBasicType("boolean",
XML_SCHEMAS_BOOLEAN,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeBooleanDef == NULL)
goto error;
xmlSchemaTypeAnyURIDef = xmlSchemaInitBasicType("anyURI",
XML_SCHEMAS_ANYURI,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeAnyURIDef == NULL)
goto error;
xmlSchemaTypeHexBinaryDef = xmlSchemaInitBasicType("hexBinary",
XML_SCHEMAS_HEXBINARY,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeHexBinaryDef == NULL)
goto error;
xmlSchemaTypeBase64BinaryDef
= xmlSchemaInitBasicType("base64Binary", XML_SCHEMAS_BASE64BINARY,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeBase64BinaryDef == NULL)
goto error;
xmlSchemaTypeNotationDef = xmlSchemaInitBasicType("NOTATION",
XML_SCHEMAS_NOTATION,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeNotationDef == NULL)
goto error;
xmlSchemaTypeQNameDef = xmlSchemaInitBasicType("QName",
XML_SCHEMAS_QNAME,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeQNameDef == NULL)
goto error;
/*
* derived datatypes
*/
xmlSchemaTypeIntegerDef = xmlSchemaInitBasicType("integer",
XML_SCHEMAS_INTEGER,
xmlSchemaTypeDecimalDef);
if (xmlSchemaTypeIntegerDef == NULL)
goto error;
xmlSchemaTypeNonPositiveIntegerDef =
xmlSchemaInitBasicType("nonPositiveInteger",
XML_SCHEMAS_NPINTEGER,
xmlSchemaTypeIntegerDef);
if (xmlSchemaTypeNonPositiveIntegerDef == NULL)
goto error;
xmlSchemaTypeNegativeIntegerDef =
xmlSchemaInitBasicType("negativeInteger", XML_SCHEMAS_NINTEGER,
xmlSchemaTypeNonPositiveIntegerDef);
if (xmlSchemaTypeNegativeIntegerDef == NULL)
goto error;
xmlSchemaTypeLongDef =
xmlSchemaInitBasicType("long", XML_SCHEMAS_LONG,
xmlSchemaTypeIntegerDef);
if (xmlSchemaTypeLongDef == NULL)
goto error;
xmlSchemaTypeIntDef = xmlSchemaInitBasicType("int", XML_SCHEMAS_INT,
xmlSchemaTypeLongDef);
if (xmlSchemaTypeIntDef == NULL)
goto error;
xmlSchemaTypeShortDef = xmlSchemaInitBasicType("short",
XML_SCHEMAS_SHORT,
xmlSchemaTypeIntDef);
if (xmlSchemaTypeShortDef == NULL)
goto error;
xmlSchemaTypeByteDef = xmlSchemaInitBasicType("byte",
XML_SCHEMAS_BYTE,
xmlSchemaTypeShortDef);
if (xmlSchemaTypeByteDef == NULL)
goto error;
xmlSchemaTypeNonNegativeIntegerDef =
xmlSchemaInitBasicType("nonNegativeInteger",
XML_SCHEMAS_NNINTEGER,
xmlSchemaTypeIntegerDef);
if (xmlSchemaTypeNonNegativeIntegerDef == NULL)
goto error;
xmlSchemaTypeUnsignedLongDef =
xmlSchemaInitBasicType("unsignedLong", XML_SCHEMAS_ULONG,
xmlSchemaTypeNonNegativeIntegerDef);
if (xmlSchemaTypeUnsignedLongDef == NULL)
goto error;
xmlSchemaTypeUnsignedIntDef =
xmlSchemaInitBasicType("unsignedInt", XML_SCHEMAS_UINT,
xmlSchemaTypeUnsignedLongDef);
if (xmlSchemaTypeUnsignedIntDef == NULL)
goto error;
xmlSchemaTypeUnsignedShortDef =
xmlSchemaInitBasicType("unsignedShort", XML_SCHEMAS_USHORT,
xmlSchemaTypeUnsignedIntDef);
if (xmlSchemaTypeUnsignedShortDef == NULL)
goto error;
xmlSchemaTypeUnsignedByteDef =
xmlSchemaInitBasicType("unsignedByte", XML_SCHEMAS_UBYTE,
xmlSchemaTypeUnsignedShortDef);
if (xmlSchemaTypeUnsignedByteDef == NULL)
goto error;
xmlSchemaTypePositiveIntegerDef =
xmlSchemaInitBasicType("positiveInteger", XML_SCHEMAS_PINTEGER,
xmlSchemaTypeNonNegativeIntegerDef);
if (xmlSchemaTypePositiveIntegerDef == NULL)
goto error;
xmlSchemaTypeNormStringDef = xmlSchemaInitBasicType("normalizedString",
XML_SCHEMAS_NORMSTRING,
xmlSchemaTypeStringDef);
if (xmlSchemaTypeNormStringDef == NULL)
goto error;
xmlSchemaTypeTokenDef = xmlSchemaInitBasicType("token",
XML_SCHEMAS_TOKEN,
xmlSchemaTypeNormStringDef);
if (xmlSchemaTypeTokenDef == NULL)
goto error;
xmlSchemaTypeLanguageDef = xmlSchemaInitBasicType("language",
XML_SCHEMAS_LANGUAGE,
xmlSchemaTypeTokenDef);
if (xmlSchemaTypeLanguageDef == NULL)
goto error;
xmlSchemaTypeNameDef = xmlSchemaInitBasicType("Name",
XML_SCHEMAS_NAME,
xmlSchemaTypeTokenDef);
if (xmlSchemaTypeNameDef == NULL)
goto error;
xmlSchemaTypeNmtokenDef = xmlSchemaInitBasicType("NMTOKEN",
XML_SCHEMAS_NMTOKEN,
xmlSchemaTypeTokenDef);
if (xmlSchemaTypeNmtokenDef == NULL)
goto error;
xmlSchemaTypeNCNameDef = xmlSchemaInitBasicType("NCName",
XML_SCHEMAS_NCNAME,
xmlSchemaTypeNameDef);
if (xmlSchemaTypeNCNameDef == NULL)
goto error;
xmlSchemaTypeIdDef = xmlSchemaInitBasicType("ID", XML_SCHEMAS_ID,
xmlSchemaTypeNCNameDef);
if (xmlSchemaTypeIdDef == NULL)
goto error;
xmlSchemaTypeIdrefDef = xmlSchemaInitBasicType("IDREF",
XML_SCHEMAS_IDREF,
xmlSchemaTypeNCNameDef);
if (xmlSchemaTypeIdrefDef == NULL)
goto error;
xmlSchemaTypeEntityDef = xmlSchemaInitBasicType("ENTITY",
XML_SCHEMAS_ENTITY,
xmlSchemaTypeNCNameDef);
if (xmlSchemaTypeEntityDef == NULL)
goto error;
/*
* Derived list types.
*/
/* ENTITIES */
xmlSchemaTypeEntitiesDef = xmlSchemaInitBasicType("ENTITIES",
XML_SCHEMAS_ENTITIES,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeEntitiesDef == NULL)
goto error;
xmlSchemaTypeEntitiesDef->subtypes = xmlSchemaTypeEntityDef;
/* IDREFS */
xmlSchemaTypeIdrefsDef = xmlSchemaInitBasicType("IDREFS",
XML_SCHEMAS_IDREFS,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeIdrefsDef == NULL)
goto error;
xmlSchemaTypeIdrefsDef->subtypes = xmlSchemaTypeIdrefDef;
/* NMTOKENS */
xmlSchemaTypeNmtokensDef = xmlSchemaInitBasicType("NMTOKENS",
XML_SCHEMAS_NMTOKENS,
xmlSchemaTypeAnySimpleTypeDef);
if (xmlSchemaTypeNmtokensDef == NULL)
goto error;
xmlSchemaTypeNmtokensDef->subtypes = xmlSchemaTypeNmtokenDef;
xmlSchemaTypesInitialized = 1;
return (0);
error:
xmlSchemaCleanupTypesInternal();
return (-1);
}
/**
* xmlSchemaCleanupTypes:
*
* DEPRECATED: This function will be made private. Call xmlCleanupParser
* to free global state but see the warnings there. xmlCleanupParser
* should be only called once at program exit. In most cases, you don't
* have to call cleanup functions at all.
*
* Cleanup the default XML Schemas type library
*/
void
xmlSchemaCleanupTypes(void) {
if (xmlSchemaTypesInitialized != 0) {
xmlSchemaCleanupTypesInternal();
xmlSchemaTypesInitialized = 0;
}
}
/**
* xmlSchemaIsBuiltInTypeFacet:
* @type: the built-in type
* @facetType: the facet type
*
* Evaluates if a specific facet can be
* used in conjunction with a type.
*
* Returns 1 if the facet can be used with the given built-in type,
* 0 otherwise and -1 in case the type is not a built-in type.
*/
int
xmlSchemaIsBuiltInTypeFacet(xmlSchemaTypePtr type, int facetType)
{
if (type == NULL)
return (-1);
if (type->type != XML_SCHEMA_TYPE_BASIC)
return (-1);
switch (type->builtInType) {
case XML_SCHEMAS_BOOLEAN:
if ((facetType == XML_SCHEMA_FACET_PATTERN) ||
(facetType == XML_SCHEMA_FACET_WHITESPACE))
return (1);
else
return (0);
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_NOTATION:
case XML_SCHEMAS_QNAME:
case XML_SCHEMAS_ANYURI:
case XML_SCHEMAS_BASE64BINARY:
case XML_SCHEMAS_HEXBINARY:
if ((facetType == XML_SCHEMA_FACET_LENGTH) ||
(facetType == XML_SCHEMA_FACET_MINLENGTH) ||
(facetType == XML_SCHEMA_FACET_MAXLENGTH) ||
(facetType == XML_SCHEMA_FACET_PATTERN) ||
(facetType == XML_SCHEMA_FACET_ENUMERATION) ||
(facetType == XML_SCHEMA_FACET_WHITESPACE))
return (1);
else
return (0);
case XML_SCHEMAS_DECIMAL:
if ((facetType == XML_SCHEMA_FACET_TOTALDIGITS) ||
(facetType == XML_SCHEMA_FACET_FRACTIONDIGITS) ||
(facetType == XML_SCHEMA_FACET_PATTERN) ||
(facetType == XML_SCHEMA_FACET_WHITESPACE) ||
(facetType == XML_SCHEMA_FACET_ENUMERATION) ||
(facetType == XML_SCHEMA_FACET_MAXINCLUSIVE) ||
(facetType == XML_SCHEMA_FACET_MAXEXCLUSIVE) ||
(facetType == XML_SCHEMA_FACET_MININCLUSIVE) ||
(facetType == XML_SCHEMA_FACET_MINEXCLUSIVE))
return (1);
else
return (0);
case XML_SCHEMAS_TIME:
case XML_SCHEMAS_GDAY:
case XML_SCHEMAS_GMONTH:
case XML_SCHEMAS_GMONTHDAY:
case XML_SCHEMAS_GYEAR:
case XML_SCHEMAS_GYEARMONTH:
case XML_SCHEMAS_DATE:
case XML_SCHEMAS_DATETIME:
case XML_SCHEMAS_DURATION:
case XML_SCHEMAS_FLOAT:
case XML_SCHEMAS_DOUBLE:
if ((facetType == XML_SCHEMA_FACET_PATTERN) ||
(facetType == XML_SCHEMA_FACET_ENUMERATION) ||
(facetType == XML_SCHEMA_FACET_WHITESPACE) ||
(facetType == XML_SCHEMA_FACET_MAXINCLUSIVE) ||
(facetType == XML_SCHEMA_FACET_MAXEXCLUSIVE) ||
(facetType == XML_SCHEMA_FACET_MININCLUSIVE) ||
(facetType == XML_SCHEMA_FACET_MINEXCLUSIVE))
return (1);
else
return (0);
default:
break;
}
return (0);
}
/**
* xmlSchemaGetBuiltInType:
* @type: the type of the built in type
*
* Gives you the type struct for a built-in
* type by its type id.
*
* Returns the type if found, NULL otherwise.
*/
xmlSchemaTypePtr
xmlSchemaGetBuiltInType(xmlSchemaValType type)
{
if ((xmlSchemaTypesInitialized == 0) &&
(xmlSchemaInitTypes() < 0))
return (NULL);
switch (type) {
case XML_SCHEMAS_ANYSIMPLETYPE:
return (xmlSchemaTypeAnySimpleTypeDef);
case XML_SCHEMAS_STRING:
return (xmlSchemaTypeStringDef);
case XML_SCHEMAS_NORMSTRING:
return (xmlSchemaTypeNormStringDef);
case XML_SCHEMAS_DECIMAL:
return (xmlSchemaTypeDecimalDef);
case XML_SCHEMAS_TIME:
return (xmlSchemaTypeTimeDef);
case XML_SCHEMAS_GDAY:
return (xmlSchemaTypeGDayDef);
case XML_SCHEMAS_GMONTH:
return (xmlSchemaTypeGMonthDef);
case XML_SCHEMAS_GMONTHDAY:
return (xmlSchemaTypeGMonthDayDef);
case XML_SCHEMAS_GYEAR:
return (xmlSchemaTypeGYearDef);
case XML_SCHEMAS_GYEARMONTH:
return (xmlSchemaTypeGYearMonthDef);
case XML_SCHEMAS_DATE:
return (xmlSchemaTypeDateDef);
case XML_SCHEMAS_DATETIME:
return (xmlSchemaTypeDatetimeDef);
case XML_SCHEMAS_DURATION:
return (xmlSchemaTypeDurationDef);
case XML_SCHEMAS_FLOAT:
return (xmlSchemaTypeFloatDef);
case XML_SCHEMAS_DOUBLE:
return (xmlSchemaTypeDoubleDef);
case XML_SCHEMAS_BOOLEAN:
return (xmlSchemaTypeBooleanDef);
case XML_SCHEMAS_TOKEN:
return (xmlSchemaTypeTokenDef);
case XML_SCHEMAS_LANGUAGE:
return (xmlSchemaTypeLanguageDef);
case XML_SCHEMAS_NMTOKEN:
return (xmlSchemaTypeNmtokenDef);
case XML_SCHEMAS_NMTOKENS:
return (xmlSchemaTypeNmtokensDef);
case XML_SCHEMAS_NAME:
return (xmlSchemaTypeNameDef);
case XML_SCHEMAS_QNAME:
return (xmlSchemaTypeQNameDef);
case XML_SCHEMAS_NCNAME:
return (xmlSchemaTypeNCNameDef);
case XML_SCHEMAS_ID:
return (xmlSchemaTypeIdDef);
case XML_SCHEMAS_IDREF:
return (xmlSchemaTypeIdrefDef);
case XML_SCHEMAS_IDREFS:
return (xmlSchemaTypeIdrefsDef);
case XML_SCHEMAS_ENTITY:
return (xmlSchemaTypeEntityDef);
case XML_SCHEMAS_ENTITIES:
return (xmlSchemaTypeEntitiesDef);
case XML_SCHEMAS_NOTATION:
return (xmlSchemaTypeNotationDef);
case XML_SCHEMAS_ANYURI:
return (xmlSchemaTypeAnyURIDef);
case XML_SCHEMAS_INTEGER:
return (xmlSchemaTypeIntegerDef);
case XML_SCHEMAS_NPINTEGER:
return (xmlSchemaTypeNonPositiveIntegerDef);
case XML_SCHEMAS_NINTEGER:
return (xmlSchemaTypeNegativeIntegerDef);
case XML_SCHEMAS_NNINTEGER:
return (xmlSchemaTypeNonNegativeIntegerDef);
case XML_SCHEMAS_PINTEGER:
return (xmlSchemaTypePositiveIntegerDef);
case XML_SCHEMAS_INT:
return (xmlSchemaTypeIntDef);
case XML_SCHEMAS_UINT:
return (xmlSchemaTypeUnsignedIntDef);
case XML_SCHEMAS_LONG:
return (xmlSchemaTypeLongDef);
case XML_SCHEMAS_ULONG:
return (xmlSchemaTypeUnsignedLongDef);
case XML_SCHEMAS_SHORT:
return (xmlSchemaTypeShortDef);
case XML_SCHEMAS_USHORT:
return (xmlSchemaTypeUnsignedShortDef);
case XML_SCHEMAS_BYTE:
return (xmlSchemaTypeByteDef);
case XML_SCHEMAS_UBYTE:
return (xmlSchemaTypeUnsignedByteDef);
case XML_SCHEMAS_HEXBINARY:
return (xmlSchemaTypeHexBinaryDef);
case XML_SCHEMAS_BASE64BINARY:
return (xmlSchemaTypeBase64BinaryDef);
case XML_SCHEMAS_ANYTYPE:
return (xmlSchemaTypeAnyTypeDef);
default:
return (NULL);
}
}
/**
* xmlSchemaValueAppend:
* @prev: the value
* @cur: the value to be appended
*
* Appends a next sibling to a list of computed values.
*
* Returns 0 if succeeded and -1 on API errors.
*/
int
xmlSchemaValueAppend(xmlSchemaValPtr prev, xmlSchemaValPtr cur) {
if ((prev == NULL) || (cur == NULL))
return (-1);
prev->next = cur;
return (0);
}
/**
* xmlSchemaValueGetNext:
* @cur: the value
*
* Accessor for the next sibling of a list of computed values.
*
* Returns the next value or NULL if there was none, or on
* API errors.
*/
xmlSchemaValPtr
xmlSchemaValueGetNext(xmlSchemaValPtr cur) {
if (cur == NULL)
return (NULL);
return (cur->next);
}
/**
* xmlSchemaValueGetAsString:
* @val: the value
*
* Accessor for the string value of a computed value.
*
* Returns the string value or NULL if there was none, or on
* API errors.
*/
const xmlChar *
xmlSchemaValueGetAsString(xmlSchemaValPtr val)
{
if (val == NULL)
return (NULL);
switch (val->type) {
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_NORMSTRING:
case XML_SCHEMAS_ANYSIMPLETYPE:
case XML_SCHEMAS_TOKEN:
case XML_SCHEMAS_LANGUAGE:
case XML_SCHEMAS_NMTOKEN:
case XML_SCHEMAS_NAME:
case XML_SCHEMAS_NCNAME:
case XML_SCHEMAS_ID:
case XML_SCHEMAS_IDREF:
case XML_SCHEMAS_ENTITY:
case XML_SCHEMAS_ANYURI:
return (BAD_CAST val->value.str);
default:
break;
}
return (NULL);
}
/**
* xmlSchemaValueGetAsBoolean:
* @val: the value
*
* Accessor for the boolean value of a computed value.
*
* Returns 1 if true and 0 if false, or in case of an error. Hmm.
*/
int
xmlSchemaValueGetAsBoolean(xmlSchemaValPtr val)
{
if ((val == NULL) || (val->type != XML_SCHEMAS_BOOLEAN))
return (0);
return (val->value.b);
}
/**
* xmlSchemaNewStringValue:
* @type: the value type
* @value: the value
*
* Allocate a new simple type value. The type can be
* of XML_SCHEMAS_STRING.
* WARNING: This one is intended to be expanded for other
* string based types. We need this for anySimpleType as well.
* The given value is consumed and freed with the struct.
*
* Returns a pointer to the new value or NULL in case of error
*/
xmlSchemaValPtr
xmlSchemaNewStringValue(xmlSchemaValType type,
const xmlChar *value)
{
xmlSchemaValPtr val;
if (type != XML_SCHEMAS_STRING)
return(NULL);
val = (xmlSchemaValPtr) xmlMalloc(sizeof(xmlSchemaVal));
if (val == NULL) {
return(NULL);
}
memset(val, 0, sizeof(xmlSchemaVal));
val->type = type;
val->value.str = (xmlChar *) value;
return(val);
}
/**
* xmlSchemaNewNOTATIONValue:
* @name: the notation name
* @ns: the notation namespace name or NULL
*
* Allocate a new NOTATION value.
* The given values are consumed and freed with the struct.
*
* Returns a pointer to the new value or NULL in case of error
*/
xmlSchemaValPtr
xmlSchemaNewNOTATIONValue(const xmlChar *name,
const xmlChar *ns)
{
xmlSchemaValPtr val;
val = xmlSchemaNewValue(XML_SCHEMAS_NOTATION);
if (val == NULL)
return (NULL);
val->value.qname.name = (xmlChar *)name;
if (ns != NULL)
val->value.qname.uri = (xmlChar *)ns;
return(val);
}
/**
* xmlSchemaNewQNameValue:
* @namespaceName: the namespace name
* @localName: the local name
*
* Allocate a new QName value.
* The given values are consumed and freed with the struct.
*
* Returns a pointer to the new value or NULL in case of an error.
*/
xmlSchemaValPtr
xmlSchemaNewQNameValue(const xmlChar *namespaceName,
const xmlChar *localName)
{
xmlSchemaValPtr val;
val = xmlSchemaNewValue(XML_SCHEMAS_QNAME);
if (val == NULL)
return (NULL);
val->value.qname.name = (xmlChar *) localName;
val->value.qname.uri = (xmlChar *) namespaceName;
return(val);
}
/**
* xmlSchemaFreeValue:
* @value: the value to free
*
* Cleanup the default XML Schemas type library
*/
void
xmlSchemaFreeValue(xmlSchemaValPtr value) {
xmlSchemaValPtr prev;
while (value != NULL) {
switch (value->type) {
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_NORMSTRING:
case XML_SCHEMAS_TOKEN:
case XML_SCHEMAS_LANGUAGE:
case XML_SCHEMAS_NMTOKEN:
case XML_SCHEMAS_NMTOKENS:
case XML_SCHEMAS_NAME:
case XML_SCHEMAS_NCNAME:
case XML_SCHEMAS_ID:
case XML_SCHEMAS_IDREF:
case XML_SCHEMAS_IDREFS:
case XML_SCHEMAS_ENTITY:
case XML_SCHEMAS_ENTITIES:
case XML_SCHEMAS_ANYURI:
case XML_SCHEMAS_ANYSIMPLETYPE:
if (value->value.str != NULL)
xmlFree(value->value.str);
break;
case XML_SCHEMAS_NOTATION:
case XML_SCHEMAS_QNAME:
if (value->value.qname.uri != NULL)
xmlFree(value->value.qname.uri);
if (value->value.qname.name != NULL)
xmlFree(value->value.qname.name);
break;
case XML_SCHEMAS_HEXBINARY:
if (value->value.hex.str != NULL)
xmlFree(value->value.hex.str);
break;
case XML_SCHEMAS_BASE64BINARY:
if (value->value.base64.str != NULL)
xmlFree(value->value.base64.str);
break;
default:
break;
}
prev = value;
value = value->next;
xmlFree(prev);
}
}
/**
* xmlSchemaGetPredefinedType:
* @name: the type name
* @ns: the URI of the namespace usually "http://www.w3.org/2001/XMLSchema"
*
* Lookup a type in the default XML Schemas type library
*
* Returns the type if found, NULL otherwise
*/
xmlSchemaTypePtr
xmlSchemaGetPredefinedType(const xmlChar *name, const xmlChar *ns) {
if ((xmlSchemaTypesInitialized == 0) &&
(xmlSchemaInitTypes() < 0))
return (NULL);
if (name == NULL)
return(NULL);
return((xmlSchemaTypePtr) xmlHashLookup2(xmlSchemaTypesBank, name, ns));
}
/**
* xmlSchemaGetBuiltInListSimpleTypeItemType:
* @type: the built-in simple type.
*
* Lookup function
*
* Returns the item type of @type as defined by the built-in datatype
* hierarchy of XML Schema Part 2: Datatypes, or NULL in case of an error.
*/
xmlSchemaTypePtr
xmlSchemaGetBuiltInListSimpleTypeItemType(xmlSchemaTypePtr type)
{
if ((type == NULL) || (type->type != XML_SCHEMA_TYPE_BASIC))
return (NULL);
switch (type->builtInType) {
case XML_SCHEMAS_NMTOKENS:
return (xmlSchemaTypeNmtokenDef );
case XML_SCHEMAS_IDREFS:
return (xmlSchemaTypeIdrefDef);
case XML_SCHEMAS_ENTITIES:
return (xmlSchemaTypeEntityDef);
default:
return (NULL);
}
}
/****************************************************************
* *
* Convenience macros and functions *
* *
****************************************************************/
#define IS_TZO_CHAR(c) \
((c == 0) || (c == 'Z') || (c == '+') || (c == '-'))
#define VALID_YEAR(yr) (yr != 0)
#define VALID_MONTH(mon) ((mon >= 1) && (mon <= 12))
/* VALID_DAY should only be used when month is unknown */
#define VALID_DAY(day) ((day >= 1) && (day <= 31))
#define VALID_HOUR(hr) ((hr >= 0) && (hr <= 23))
#define VALID_MIN(min) ((min >= 0) && (min <= 59))
#define VALID_SEC(sec) ((sec >= 0) && (sec < 60))
#define VALID_TZO(tzo) ((tzo >= -840) && (tzo <= 840))
#define IS_LEAP(y) \
(((y % 4 == 0) && (y % 100 != 0)) || (y % 400 == 0))
static const unsigned int daysInMonth[12] =
{ 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 };
static const unsigned int daysInMonthLeap[12] =
{ 31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31 };
#define MAX_DAYINMONTH(yr,mon) \
(IS_LEAP(yr) ? daysInMonthLeap[mon - 1] : daysInMonth[mon - 1])
#define VALID_MDAY(dt) \
(IS_LEAP(dt->year) ? \
(dt->day <= daysInMonthLeap[dt->mon - 1]) : \
(dt->day <= daysInMonth[dt->mon - 1]))
#define VALID_DATE(dt) \
(VALID_YEAR(dt->year) && VALID_MONTH(dt->mon) && VALID_MDAY(dt))
#define VALID_END_OF_DAY(dt) \
((dt)->hour == 24 && (dt)->min == 0 && (dt)->sec == 0)
#define VALID_TIME(dt) \
(((VALID_HOUR(dt->hour) && VALID_MIN(dt->min) && \
VALID_SEC(dt->sec)) || VALID_END_OF_DAY(dt)) && \
VALID_TZO(dt->tzo))
#define VALID_DATETIME(dt) \
(VALID_DATE(dt) && VALID_TIME(dt))
#define SECS_PER_MIN 60
#define MINS_PER_HOUR 60
#define HOURS_PER_DAY 24
#define SECS_PER_HOUR (MINS_PER_HOUR * SECS_PER_MIN)
#define SECS_PER_DAY (HOURS_PER_DAY * SECS_PER_HOUR)
#define MINS_PER_DAY (HOURS_PER_DAY * MINS_PER_HOUR)
static const long dayInYearByMonth[12] =
{ 0, 31, 59, 90, 120, 151, 181, 212, 243, 273, 304, 334 };
static const long dayInLeapYearByMonth[12] =
{ 0, 31, 60, 91, 121, 152, 182, 213, 244, 274, 305, 335 };
#define DAY_IN_YEAR(day, month, year) \
((IS_LEAP(year) ? \
dayInLeapYearByMonth[month - 1] : \
dayInYearByMonth[month - 1]) + day)
#ifdef DEBUG
#define DEBUG_DATE(dt) \
xmlGenericError(xmlGenericErrorContext, \
"type=%o %04ld-%02u-%02uT%02u:%02u:%03f", \
dt->type,dt->value.date.year,dt->value.date.mon, \
dt->value.date.day,dt->value.date.hour,dt->value.date.min, \
dt->value.date.sec); \
if (dt->value.date.tz_flag) \
if (dt->value.date.tzo != 0) \
xmlGenericError(xmlGenericErrorContext, \
"%+05d\n",dt->value.date.tzo); \
else \
xmlGenericError(xmlGenericErrorContext, "Z\n"); \
else \
xmlGenericError(xmlGenericErrorContext,"\n")
#else
#define DEBUG_DATE(dt)
#endif
/**
* _xmlSchemaParseGYear:
* @dt: pointer to a date structure
* @str: pointer to the string to analyze
*
* Parses a xs:gYear without time zone and fills in the appropriate
* field of the @dt structure. @str is updated to point just after the
* xs:gYear. It is supposed that @dt->year is big enough to contain
* the year.
*
* Returns 0 or the error code
*/
static int
_xmlSchemaParseGYear (xmlSchemaValDatePtr dt, const xmlChar **str) {
const xmlChar *cur = *str, *firstChar;
int isneg = 0, digcnt = 0;
if (((*cur < '0') || (*cur > '9')) &&
(*cur != '-') && (*cur != '+'))
return -1;
if (*cur == '-') {
isneg = 1;
cur++;
}
firstChar = cur;
while ((*cur >= '0') && (*cur <= '9')) {
int digit = *cur - '0';
if (dt->year > LONG_MAX / 10)
return 2;
dt->year *= 10;
if (dt->year > LONG_MAX - digit)
return 2;
dt->year += digit;
cur++;
digcnt++;
}
/* year must be at least 4 digits (CCYY); over 4
* digits cannot have a leading zero. */
if ((digcnt < 4) || ((digcnt > 4) && (*firstChar == '0')))
return 1;
if (isneg)
dt->year = - dt->year;
if (!VALID_YEAR(dt->year))
return 2;
*str = cur;
return 0;
}
/**
* PARSE_2_DIGITS:
* @num: the integer to fill in
* @cur: an #xmlChar *
* @invalid: an integer
*
* Parses a 2-digits integer and updates @num with the value. @cur is
* updated to point just after the integer.
* In case of error, @invalid is set to %TRUE, values of @num and
* @cur are undefined.
*/
#define PARSE_2_DIGITS(num, cur, invalid) \
if ((cur[0] < '0') || (cur[0] > '9') || \
(cur[1] < '0') || (cur[1] > '9')) \
invalid = 1; \
else \
num = (cur[0] - '0') * 10 + (cur[1] - '0'); \
cur += 2;
/**
* PARSE_FLOAT:
* @num: the double to fill in
* @cur: an #xmlChar *
* @invalid: an integer
*
* Parses a float and updates @num with the value. @cur is
* updated to point just after the float. The float must have a
* 2-digits integer part and may or may not have a decimal part.
* In case of error, @invalid is set to %TRUE, values of @num and
* @cur are undefined.
*/
#define PARSE_FLOAT(num, cur, invalid) \
PARSE_2_DIGITS(num, cur, invalid); \
if (!invalid && (*cur == '.')) { \
double mult = 1; \
cur++; \
if ((*cur < '0') || (*cur > '9')) \
invalid = 1; \
while ((*cur >= '0') && (*cur <= '9')) { \
mult /= 10; \
num += (*cur - '0') * mult; \
cur++; \
} \
}
/**
* _xmlSchemaParseGMonth:
* @dt: pointer to a date structure
* @str: pointer to the string to analyze
*
* Parses a xs:gMonth without time zone and fills in the appropriate
* field of the @dt structure. @str is updated to point just after the
* xs:gMonth.
*
* Returns 0 or the error code
*/
static int
_xmlSchemaParseGMonth (xmlSchemaValDatePtr dt, const xmlChar **str) {
const xmlChar *cur = *str;
int ret = 0;
unsigned int value = 0;
PARSE_2_DIGITS(value, cur, ret);
if (ret != 0)
return ret;
if (!VALID_MONTH(value))
return 2;
dt->mon = value;
*str = cur;
return 0;
}
/**
* _xmlSchemaParseGDay:
* @dt: pointer to a date structure
* @str: pointer to the string to analyze
*
* Parses a xs:gDay without time zone and fills in the appropriate
* field of the @dt structure. @str is updated to point just after the
* xs:gDay.
*
* Returns 0 or the error code
*/
static int
_xmlSchemaParseGDay (xmlSchemaValDatePtr dt, const xmlChar **str) {
const xmlChar *cur = *str;
int ret = 0;
unsigned int value = 0;
PARSE_2_DIGITS(value, cur, ret);
if (ret != 0)
return ret;
if (!VALID_DAY(value))
return 2;
dt->day = value;
*str = cur;
return 0;
}
/**
* _xmlSchemaParseTime:
* @dt: pointer to a date structure
* @str: pointer to the string to analyze
*
* Parses a xs:time without time zone and fills in the appropriate
* fields of the @dt structure. @str is updated to point just after the
* xs:time.
* In case of error, values of @dt fields are undefined.
*
* Returns 0 or the error code
*/
static int
_xmlSchemaParseTime (xmlSchemaValDatePtr dt, const xmlChar **str) {
const xmlChar *cur = *str;
int ret = 0;
int value = 0;
PARSE_2_DIGITS(value, cur, ret);
if (ret != 0)
return ret;
if (*cur != ':')
return 1;
if (!VALID_HOUR(value) && value != 24 /* Allow end-of-day hour */)
return 2;
cur++;
/* the ':' insures this string is xs:time */
dt->hour = value;
PARSE_2_DIGITS(value, cur, ret);
if (ret != 0)
return ret;
if (!VALID_MIN(value))
return 2;
dt->min = value;
if (*cur != ':')
return 1;
cur++;
PARSE_FLOAT(dt->sec, cur, ret);
if (ret != 0)
return ret;
if (!VALID_TIME(dt))
return 2;
*str = cur;
return 0;
}
/**
* _xmlSchemaParseTimeZone:
* @dt: pointer to a date structure
* @str: pointer to the string to analyze
*
* Parses a time zone without time zone and fills in the appropriate
* field of the @dt structure. @str is updated to point just after the
* time zone.
*
* Returns 0 or the error code
*/
static int
_xmlSchemaParseTimeZone (xmlSchemaValDatePtr dt, const xmlChar **str) {
const xmlChar *cur;
int ret = 0;
if (str == NULL)
return -1;
cur = *str;
switch (*cur) {
case 0:
dt->tz_flag = 0;
dt->tzo = 0;
break;
case 'Z':
dt->tz_flag = 1;
dt->tzo = 0;
cur++;
break;
case '+':
case '-': {
int isneg = 0, tmp = 0;
isneg = (*cur == '-');
cur++;
PARSE_2_DIGITS(tmp, cur, ret);
if (ret != 0)
return ret;
if (!VALID_HOUR(tmp))
return 2;
if (*cur != ':')
return 1;
cur++;
dt->tzo = tmp * 60;
PARSE_2_DIGITS(tmp, cur, ret);
if (ret != 0)
return ret;
if (!VALID_MIN(tmp))
return 2;
dt->tzo += tmp;
if (isneg)
dt->tzo = - dt->tzo;
if (!VALID_TZO(dt->tzo))
return 2;
dt->tz_flag = 1;
break;
}
default:
return 1;
}
*str = cur;
return 0;
}
/**
* _xmlSchemaBase64Decode:
* @ch: a character
*
* Converts a base64 encoded character to its base 64 value.
*
* Returns 0-63 (value), 64 (pad), or -1 (not recognized)
*/
static int
_xmlSchemaBase64Decode (const xmlChar ch) {
if (('A' <= ch) && (ch <= 'Z')) return ch - 'A';
if (('a' <= ch) && (ch <= 'z')) return ch - 'a' + 26;
if (('0' <= ch) && (ch <= '9')) return ch - '0' + 52;
if ('+' == ch) return 62;
if ('/' == ch) return 63;
if ('=' == ch) return 64;
return -1;
}
/****************************************************************
* *
* XML Schema Dates/Times Datatypes Handling *
* *
****************************************************************/
/**
* PARSE_DIGITS:
* @num: the integer to fill in
* @cur: an #xmlChar *
* @num_type: an integer flag
*
* Parses a digits integer and updates @num with the value. @cur is
* updated to point just after the integer.
* In case of error, @num_type is set to -1, values of @num and
* @cur are undefined.
*/
#define PARSE_DIGITS(num, cur, num_type) \
if ((*cur < '0') || (*cur > '9')) \
num_type = -1; \
else \
while ((*cur >= '0') && (*cur <= '9')) { \
num = num * 10 + (*cur - '0'); \
cur++; \
}
/**
* PARSE_NUM:
* @num: the double to fill in
* @cur: an #xmlChar *
* @num_type: an integer flag
*
* Parses a float or integer and updates @num with the value. @cur is
* updated to point just after the number. If the number is a float,
* then it must have an integer part and a decimal part; @num_type will
* be set to 1. If there is no decimal part, @num_type is set to zero.
* In case of error, @num_type is set to -1, values of @num and
* @cur are undefined.
*/
#define PARSE_NUM(num, cur, num_type) \
num = 0; \
PARSE_DIGITS(num, cur, num_type); \
if (!num_type && (*cur == '.')) { \
double mult = 1; \
cur++; \
if ((*cur < '0') || (*cur > '9')) \
num_type = -1; \
else \
num_type = 1; \
while ((*cur >= '0') && (*cur <= '9')) { \
mult /= 10; \
num += (*cur - '0') * mult; \
cur++; \
} \
}
/**
* xmlSchemaValidateDates:
* @type: the expected type or XML_SCHEMAS_UNKNOWN
* @dateTime: string to analyze
* @val: the return computed value
*
* Check that @dateTime conforms to the lexical space of one of the date types.
* if true a value is computed and returned in @val.
*
* Returns 0 if this validates, a positive error code number otherwise
* and -1 in case of internal or API error.
*/
static int
xmlSchemaValidateDates (xmlSchemaValType type,
const xmlChar *dateTime, xmlSchemaValPtr *val,
int collapse) {
xmlSchemaValPtr dt;
int ret;
const xmlChar *cur = dateTime;
#define RETURN_TYPE_IF_VALID(t) \
if (IS_TZO_CHAR(*cur)) { \
ret = _xmlSchemaParseTimeZone(&(dt->value.date), &cur); \
if (ret == 0) { \
if (*cur != 0) \
goto error; \
dt->type = t; \
goto done; \
} \
}
if (dateTime == NULL)
return -1;
if (collapse)
while IS_WSP_BLANK_CH(*cur) cur++;
if ((*cur != '-') && (*cur < '0') && (*cur > '9'))
return 1;
dt = xmlSchemaNewValue(XML_SCHEMAS_UNKNOWN);
if (dt == NULL)
return -1;
if ((cur[0] == '-') && (cur[1] == '-')) {
/*
* It's an incomplete date (xs:gMonthDay, xs:gMonth or
* xs:gDay)
*/
cur += 2;
/* is it an xs:gDay? */
if (*cur == '-') {
if (type == XML_SCHEMAS_GMONTH)
goto error;
++cur;
ret = _xmlSchemaParseGDay(&(dt->value.date), &cur);
if (ret != 0)
goto error;
RETURN_TYPE_IF_VALID(XML_SCHEMAS_GDAY);
goto error;
}
/*
* it should be an xs:gMonthDay or xs:gMonth
*/
ret = _xmlSchemaParseGMonth(&(dt->value.date), &cur);
if (ret != 0)
goto error;
/*
* a '-' char could indicate this type is xs:gMonthDay or
* a negative time zone offset. Check for xs:gMonthDay first.
* Also the first three char's of a negative tzo (-MM:SS) can
* appear to be a valid day; so even if the day portion
* of the xs:gMonthDay verifies, we must insure it was not
* a tzo.
*/
if (*cur == '-') {
const xmlChar *rewnd = cur;
cur++;
ret = _xmlSchemaParseGDay(&(dt->value.date), &cur);
if ((ret == 0) && ((*cur == 0) || (*cur != ':'))) {
/*
* we can use the VALID_MDAY macro to validate the month
* and day because the leap year test will flag year zero
* as a leap year (even though zero is an invalid year).
* FUTURE TODO: Zero will become valid in XML Schema 1.1
* probably.
*/
if (VALID_MDAY((&(dt->value.date)))) {
RETURN_TYPE_IF_VALID(XML_SCHEMAS_GMONTHDAY);
goto error;
}
}
/*
* not xs:gMonthDay so rewind and check if just xs:gMonth
* with an optional time zone.
*/
cur = rewnd;
}
RETURN_TYPE_IF_VALID(XML_SCHEMAS_GMONTH);
goto error;
}
/*
* It's a right-truncated date or an xs:time.
* Try to parse an xs:time then fallback on right-truncated dates.
*/
if ((*cur >= '0') && (*cur <= '9')) {
ret = _xmlSchemaParseTime(&(dt->value.date), &cur);
if (ret == 0) {
/* it's an xs:time */
RETURN_TYPE_IF_VALID(XML_SCHEMAS_TIME);
}
}
/* fallback on date parsing */
cur = dateTime;
ret = _xmlSchemaParseGYear(&(dt->value.date), &cur);
if (ret != 0)
goto error;
/* is it an xs:gYear? */
RETURN_TYPE_IF_VALID(XML_SCHEMAS_GYEAR);
if (*cur != '-')
goto error;
cur++;
ret = _xmlSchemaParseGMonth(&(dt->value.date), &cur);
if (ret != 0)
goto error;
/* is it an xs:gYearMonth? */
RETURN_TYPE_IF_VALID(XML_SCHEMAS_GYEARMONTH);
if (*cur != '-')
goto error;
cur++;
ret = _xmlSchemaParseGDay(&(dt->value.date), &cur);
if ((ret != 0) || !VALID_DATE((&(dt->value.date))))
goto error;
/* is it an xs:date? */
RETURN_TYPE_IF_VALID(XML_SCHEMAS_DATE);
if (*cur != 'T')
goto error;
cur++;
/* it should be an xs:dateTime */
ret = _xmlSchemaParseTime(&(dt->value.date), &cur);
if (ret != 0)
goto error;
ret = _xmlSchemaParseTimeZone(&(dt->value.date), &cur);
if (collapse)
while IS_WSP_BLANK_CH(*cur) cur++;
if ((ret != 0) || (*cur != 0) || (!(VALID_DATETIME((&(dt->value.date))))))
goto error;
dt->type = XML_SCHEMAS_DATETIME;
done:
#if 1
if ((type != XML_SCHEMAS_UNKNOWN) && (type != dt->type))
goto error;
#else
/*
* insure the parsed type is equal to or less significant (right
* truncated) than the desired type.
*/
if ((type != XML_SCHEMAS_UNKNOWN) && (type != dt->type)) {
/* time only matches time */
if ((type == XML_SCHEMAS_TIME) && (dt->type == XML_SCHEMAS_TIME))
goto error;
if ((type == XML_SCHEMAS_DATETIME) &&
((dt->type != XML_SCHEMAS_DATE) ||
(dt->type != XML_SCHEMAS_GYEARMONTH) ||
(dt->type != XML_SCHEMAS_GYEAR)))
goto error;
if ((type == XML_SCHEMAS_DATE) &&
((dt->type != XML_SCHEMAS_GYEAR) ||
(dt->type != XML_SCHEMAS_GYEARMONTH)))
goto error;
if ((type == XML_SCHEMAS_GYEARMONTH) && (dt->type != XML_SCHEMAS_GYEAR))
goto error;
if ((type == XML_SCHEMAS_GMONTHDAY) && (dt->type != XML_SCHEMAS_GMONTH))
goto error;
}
#endif
if (val != NULL)
*val = dt;
else
xmlSchemaFreeValue(dt);
return 0;
error:
if (dt != NULL)
xmlSchemaFreeValue(dt);
return 1;
}
/**
* xmlSchemaValidateDuration:
* @type: the predefined type
* @duration: string to analyze
* @val: the return computed value
*
* Check that @duration conforms to the lexical space of the duration type.
* if true a value is computed and returned in @val.
*
* Returns 0 if this validates, a positive error code number otherwise
* and -1 in case of internal or API error.
*/
static int
xmlSchemaValidateDuration (xmlSchemaTypePtr type ATTRIBUTE_UNUSED,
const xmlChar *duration, xmlSchemaValPtr *val,
int collapse) {
const xmlChar *cur = duration;
xmlSchemaValPtr dur;
int isneg = 0;
unsigned int seq = 0;
long days, secs = 0;
double sec_frac = 0.0;
if (duration == NULL)
return -1;
if (collapse)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur == '-') {
isneg = 1;
cur++;
}
/* duration must start with 'P' (after sign) */
if (*cur++ != 'P')
return 1;
if (*cur == 0)
return 1;
dur = xmlSchemaNewValue(XML_SCHEMAS_DURATION);
if (dur == NULL)
return -1;
while (*cur != 0) {
long num = 0;
size_t has_digits = 0;
int has_frac = 0;
const xmlChar desig[] = {'Y', 'M', 'D', 'H', 'M', 'S'};
/* input string should be empty or invalid date/time item */
if (seq >= sizeof(desig))
goto error;
/* T designator must be present for time items */
if (*cur == 'T') {
if (seq > 3)
goto error;
cur++;
seq = 3;
} else if (seq == 3)
goto error;
/* Parse integral part. */
while (*cur >= '0' && *cur <= '9') {
long digit = *cur - '0';
if (num > LONG_MAX / 10)
goto error;
num *= 10;
if (num > LONG_MAX - digit)
goto error;
num += digit;
has_digits = 1;
cur++;
}
if (*cur == '.') {
/* Parse fractional part. */
double mult = 1.0;
cur++;
has_frac = 1;
while (*cur >= '0' && *cur <= '9') {
mult /= 10.0;
sec_frac += (*cur - '0') * mult;
has_digits = 1;
cur++;
}
}
while (*cur != desig[seq]) {
seq++;
/* No T designator or invalid char. */
if (seq == 3 || seq == sizeof(desig))
goto error;
}
cur++;
if (!has_digits || (has_frac && (seq != 5)))
goto error;
switch (seq) {
case 0:
/* Year */
if (num > LONG_MAX / 12)
goto error;
dur->value.dur.mon = num * 12;
break;
case 1:
/* Month */
if (dur->value.dur.mon > LONG_MAX - num)
goto error;
dur->value.dur.mon += num;
break;
case 2:
/* Day */
dur->value.dur.day = num;
break;
case 3:
/* Hour */
days = num / HOURS_PER_DAY;
if (dur->value.dur.day > LONG_MAX - days)
goto error;
dur->value.dur.day += days;
secs = (num % HOURS_PER_DAY) * SECS_PER_HOUR;
break;
case 4:
/* Minute */
days = num / MINS_PER_DAY;
if (dur->value.dur.day > LONG_MAX - days)
goto error;
dur->value.dur.day += days;
secs += (num % MINS_PER_DAY) * SECS_PER_MIN;
break;
case 5:
/* Second */
days = num / SECS_PER_DAY;
if (dur->value.dur.day > LONG_MAX - days)
goto error;
dur->value.dur.day += days;
secs += num % SECS_PER_DAY;
break;
}
seq++;
}
days = secs / SECS_PER_DAY;
if (dur->value.dur.day > LONG_MAX - days)
goto error;
dur->value.dur.day += days;
dur->value.dur.sec = (secs % SECS_PER_DAY) + sec_frac;
if (isneg) {
dur->value.dur.mon = -dur->value.dur.mon;
dur->value.dur.day = -dur->value.dur.day;
dur->value.dur.sec = -dur->value.dur.sec;
}
if (val != NULL)
*val = dur;
else
xmlSchemaFreeValue(dur);
return 0;
error:
if (dur != NULL)
xmlSchemaFreeValue(dur);
return 1;
}
/**
* xmlSchemaStrip:
* @value: a value
*
* Removes the leading and ending spaces of a string
*
* Returns the new string or NULL if no change was required.
*/
static xmlChar *
xmlSchemaStrip(const xmlChar *value) {
const xmlChar *start = value, *end, *f;
if (value == NULL) return(NULL);
while ((*start != 0) && (IS_BLANK_CH(*start))) start++;
end = start;
while (*end != 0) end++;
f = end;
end--;
while ((end > start) && (IS_BLANK_CH(*end))) end--;
end++;
if ((start == value) && (f == end)) return(NULL);
return(xmlStrndup(start, end - start));
}
/**
* xmlSchemaWhiteSpaceReplace:
* @value: a value
*
* Replaces 0xd, 0x9 and 0xa with a space.
*
* Returns the new string or NULL if no change was required.
*/
xmlChar *
xmlSchemaWhiteSpaceReplace(const xmlChar *value) {
const xmlChar *cur = value;
xmlChar *ret = NULL, *mcur;
if (value == NULL)
return(NULL);
while ((*cur != 0) &&
(((*cur) != 0xd) && ((*cur) != 0x9) && ((*cur) != 0xa))) {
cur++;
}
if (*cur == 0)
return (NULL);
ret = xmlStrdup(value);
/* TODO FIXME: I guess gcc will bark at this. */
mcur = (xmlChar *) (ret + (cur - value));
do {
if ( ((*mcur) == 0xd) || ((*mcur) == 0x9) || ((*mcur) == 0xa) )
*mcur = ' ';
mcur++;
} while (*mcur != 0);
return(ret);
}
/**
* xmlSchemaCollapseString:
* @value: a value
*
* Removes and normalize white spaces in the string
*
* Returns the new string or NULL if no change was required.
*/
xmlChar *
xmlSchemaCollapseString(const xmlChar *value) {
const xmlChar *start = value, *end, *f;
xmlChar *g;
int col = 0;
if (value == NULL) return(NULL);
while ((*start != 0) && (IS_BLANK_CH(*start))) start++;
end = start;
while (*end != 0) {
if ((*end == ' ') && (IS_BLANK_CH(end[1]))) {
col = end - start;
break;
} else if ((*end == 0xa) || (*end == 0x9) || (*end == 0xd)) {
col = end - start;
break;
}
end++;
}
if (col == 0) {
f = end;
end--;
while ((end > start) && (IS_BLANK_CH(*end))) end--;
end++;
if ((start == value) && (f == end)) return(NULL);
return(xmlStrndup(start, end - start));
}
start = xmlStrdup(start);
if (start == NULL) return(NULL);
g = (xmlChar *) (start + col);
end = g;
while (*end != 0) {
if (IS_BLANK_CH(*end)) {
end++;
while (IS_BLANK_CH(*end)) end++;
if (*end != 0)
*g++ = ' ';
} else
*g++ = *end++;
}
*g = 0;
return((xmlChar *) start);
}
/**
* xmlSchemaValAtomicListNode:
* @type: the predefined atomic type for a token in the list
* @value: the list value to check
* @ret: the return computed value
* @node: the node containing the value
*
* Check that a value conforms to the lexical space of the predefined
* list type. if true a value is computed and returned in @ret.
*
* Returns the number of items if this validates, a negative error code
* number otherwise
*/
static int
xmlSchemaValAtomicListNode(xmlSchemaTypePtr type, const xmlChar *value,
xmlSchemaValPtr *ret, xmlNodePtr node) {
xmlChar *val, *cur, *endval;
int nb_values = 0;
int tmp = 0;
if (value == NULL) {
return(-1);
}
val = xmlStrdup(value);
if (val == NULL) {
return(-1);
}
if (ret != NULL) {
*ret = NULL;
}
cur = val;
/*
* Split the list
*/
while (IS_BLANK_CH(*cur)) *cur++ = 0;
while (*cur != 0) {
if (IS_BLANK_CH(*cur)) {
*cur = 0;
cur++;
while (IS_BLANK_CH(*cur)) *cur++ = 0;
} else {
nb_values++;
cur++;
while ((*cur != 0) && (!IS_BLANK_CH(*cur))) cur++;
}
}
if (nb_values == 0) {
xmlFree(val);
return(nb_values);
}
endval = cur;
cur = val;
while ((*cur == 0) && (cur != endval)) cur++;
while (cur != endval) {
tmp = xmlSchemaValPredefTypeNode(type, cur, NULL, node);
if (tmp != 0)
break;
while (*cur != 0) cur++;
while ((*cur == 0) && (cur != endval)) cur++;
}
/* TODO what return value ? c.f. bug #158628
if (ret != NULL) {
TODO
} */
xmlFree(val);
if (tmp == 0)
return(nb_values);
return(-1);
}
/**
* xmlSchemaParseUInt:
* @str: pointer to the string R/W
* @llo: pointer to the low result
* @lmi: pointer to the mid result
* @lhi: pointer to the high result
*
* Parse an unsigned long into 3 fields.
*
* Returns the number of significant digits in the number or
* -1 if overflow of the capacity and -2 if it's not a number.
*/
static int
xmlSchemaParseUInt(const xmlChar **str, unsigned long *llo,
unsigned long *lmi, unsigned long *lhi) {
unsigned long lo = 0, mi = 0, hi = 0;
const xmlChar *tmp, *cur = *str;
int ret = 0, i = 0;
if (!((*cur >= '0') && (*cur <= '9')))
return(-2);
while (*cur == '0') { /* ignore leading zeroes */
cur++;
}
tmp = cur;
while ((*tmp != 0) && (*tmp >= '0') && (*tmp <= '9')) {
i++;tmp++;ret++;
}
if (i > 24) {
*str = tmp;
return(-1);
}
while (i > 16) {
hi = hi * 10 + (*cur++ - '0');
i--;
}
while (i > 8) {
mi = mi * 10 + (*cur++ - '0');
i--;
}
while (i > 0) {
lo = lo * 10 + (*cur++ - '0');
i--;
}
*str = cur;
*llo = lo;
*lmi = mi;
*lhi = hi;
return(ret);
}
/*
* xmlSchemaCheckLanguageType
* @value: the value to check
*
* Check that a value conforms to the lexical space of the language datatype.
* Must conform to [a-zA-Z]{1,8}(-[a-zA-Z0-9]{1,8})*
*
* Returns 1 if this validates, 0 otherwise.
*/
static int
xmlSchemaCheckLanguageType(const xmlChar* value) {
int first = 1, len = 0;
const xmlChar* cur = value;
if (value == NULL)
return (0);
while (cur[0] != 0) {
if (!( ((cur[0] >= 'a') && (cur[0] <= 'z')) || ((cur[0] >= 'A') && (cur[0] <= 'Z'))
|| (cur[0] == '-')
|| ((first == 0) && (xmlIsDigit_ch(cur[0]))) ))
return (0);
if (cur[0] == '-') {
if ((len < 1) || (len > 8))
return (0);
len = 0;
first = 0;
}
else
len++;
cur++;
}
if ((len < 1) || (len > 8))
return (0);
return (1);
}
/**
* xmlSchemaValAtomicType:
* @type: the predefined type
* @value: the value to check
* @val: the return computed value
* @node: the node containing the value
* flags: flags to control the validation
*
* Check that a value conforms to the lexical space of the atomic type.
* if true a value is computed and returned in @val.
* This checks the value space for list types as well (IDREFS, NMTOKENS).
*
* Returns 0 if this validates, a positive error code number otherwise
* and -1 in case of internal or API error.
*/
static int
xmlSchemaValAtomicType(xmlSchemaTypePtr type, const xmlChar * value,
xmlSchemaValPtr * val, xmlNodePtr node, int flags,
xmlSchemaWhitespaceValueType ws,
int normOnTheFly, int applyNorm, int createStringValue)
{
xmlSchemaValPtr v;
xmlChar *norm = NULL;
int ret = 0;
if ((xmlSchemaTypesInitialized == 0) &&
(xmlSchemaInitTypes() < 0))
return (-1);
if (type == NULL)
return (-1);
/*
* validating a non existent text node is similar to validating
* an empty one.
*/
if (value == NULL)
value = BAD_CAST "";
if (val != NULL)
*val = NULL;
if ((flags == 0) && (value != NULL)) {
if ((type->builtInType != XML_SCHEMAS_STRING) &&
(type->builtInType != XML_SCHEMAS_ANYTYPE) &&
(type->builtInType != XML_SCHEMAS_ANYSIMPLETYPE)) {
if (type->builtInType == XML_SCHEMAS_NORMSTRING)
norm = xmlSchemaWhiteSpaceReplace(value);
else
norm = xmlSchemaCollapseString(value);
if (norm != NULL)
value = norm;
}
}
switch (type->builtInType) {
case XML_SCHEMAS_UNKNOWN:
goto error;
case XML_SCHEMAS_ANYTYPE:
case XML_SCHEMAS_ANYSIMPLETYPE:
if ((createStringValue) && (val != NULL)) {
v = xmlSchemaNewValue(XML_SCHEMAS_ANYSIMPLETYPE);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
goto return0;
case XML_SCHEMAS_STRING:
if (! normOnTheFly) {
const xmlChar *cur = value;
if (ws == XML_SCHEMA_WHITESPACE_REPLACE) {
while (*cur != 0) {
if ((*cur == 0xd) || (*cur == 0xa) || (*cur == 0x9)) {
goto return1;
} else {
cur++;
}
}
} else if (ws == XML_SCHEMA_WHITESPACE_COLLAPSE) {
while (*cur != 0) {
if ((*cur == 0xd) || (*cur == 0xa) || (*cur == 0x9)) {
goto return1;
} else if IS_WSP_SPACE_CH(*cur) {
cur++;
if IS_WSP_SPACE_CH(*cur)
goto return1;
} else {
cur++;
}
}
}
}
if (createStringValue && (val != NULL)) {
if (applyNorm) {
if (ws == XML_SCHEMA_WHITESPACE_COLLAPSE)
norm = xmlSchemaCollapseString(value);
else if (ws == XML_SCHEMA_WHITESPACE_REPLACE)
norm = xmlSchemaWhiteSpaceReplace(value);
if (norm != NULL)
value = norm;
}
v = xmlSchemaNewValue(XML_SCHEMAS_STRING);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
goto return0;
case XML_SCHEMAS_NORMSTRING:{
if (normOnTheFly) {
if (applyNorm) {
if (ws == XML_SCHEMA_WHITESPACE_COLLAPSE)
norm = xmlSchemaCollapseString(value);
else
norm = xmlSchemaWhiteSpaceReplace(value);
if (norm != NULL)
value = norm;
}
} else {
const xmlChar *cur = value;
while (*cur != 0) {
if ((*cur == 0xd) || (*cur == 0xa) || (*cur == 0x9)) {
goto return1;
} else {
cur++;
}
}
}
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_NORMSTRING);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
goto return0;
}
case XML_SCHEMAS_DECIMAL:{
const xmlChar *cur = value;
unsigned int len, neg, integ, hasLeadingZeroes;
xmlChar cval[25];
xmlChar *cptr = cval;
if ((cur == NULL) || (*cur == 0))
goto return1;
/*
* xs:decimal has a whitespace-facet value of 'collapse'.
*/
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
/*
* First we handle an optional sign.
*/
neg = 0;
if (*cur == '-') {
neg = 1;
cur++;
} else if (*cur == '+')
cur++;
/*
* Disallow: "", "-", "- "
*/
if (*cur == 0)
goto return1;
/*
* Next we "pre-parse" the number, in preparation for calling
* the common routine xmlSchemaParseUInt. We get rid of any
* leading zeroes (because we have reserved only 25 chars),
* and note the position of a decimal point.
*/
len = 0;
integ = ~0u;
hasLeadingZeroes = 0;
/*
* Skip leading zeroes.
*/
while (*cur == '0') {
cur++;
hasLeadingZeroes = 1;
}
if (*cur != 0) {
do {
if ((*cur >= '0') && (*cur <= '9')) {
*cptr++ = *cur++;
len++;
} else if (*cur == '.') {
cur++;
integ = len;
do {
if ((*cur >= '0') && (*cur <= '9')) {
*cptr++ = *cur++;
len++;
} else
break;
} while (len < 24);
/*
* Disallow "." but allow "00."
*/
if ((len == 0) && (!hasLeadingZeroes))
goto return1;
break;
} else
break;
} while (len < 24);
}
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur != 0)
goto return1; /* error if any extraneous chars */
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_DECIMAL);
if (v != NULL) {
/*
* Now evaluate the significant digits of the number
*/
if (len != 0) {
if (integ != ~0u) {
/*
* Get rid of trailing zeroes in the
* fractional part.
*/
while ((len != integ) && (*(cptr-1) == '0')) {
cptr--;
len--;
}
}
/*
* Terminate the (preparsed) string.
*/
if (len != 0) {
*cptr = 0;
cptr = cval;
xmlSchemaParseUInt((const xmlChar **)&cptr,
&v->value.decimal.lo,
&v->value.decimal.mi,
&v->value.decimal.hi);
}
}
/*
* Set the total digits to 1 if a zero value.
*/
v->value.decimal.sign = neg;
if (len == 0) {
/* Speedup for zero values. */
v->value.decimal.total = 1;
} else {
v->value.decimal.total = len;
if (integ == ~0u)
v->value.decimal.frac = 0;
else
v->value.decimal.frac = len - integ;
}
*val = v;
}
}
goto return0;
}
case XML_SCHEMAS_TIME:
case XML_SCHEMAS_GDAY:
case XML_SCHEMAS_GMONTH:
case XML_SCHEMAS_GMONTHDAY:
case XML_SCHEMAS_GYEAR:
case XML_SCHEMAS_GYEARMONTH:
case XML_SCHEMAS_DATE:
case XML_SCHEMAS_DATETIME:
ret = xmlSchemaValidateDates(type->builtInType, value, val,
normOnTheFly);
break;
case XML_SCHEMAS_DURATION:
ret = xmlSchemaValidateDuration(type, value, val,
normOnTheFly);
break;
case XML_SCHEMAS_FLOAT:
case XML_SCHEMAS_DOUBLE: {
const xmlChar *cur = value;
int neg = 0;
int digits_before = 0;
int digits_after = 0;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if ((cur[0] == 'N') && (cur[1] == 'a') && (cur[2] == 'N')) {
cur += 3;
if (*cur != 0)
goto return1;
if (val != NULL) {
if (type == xmlSchemaTypeFloatDef) {
v = xmlSchemaNewValue(XML_SCHEMAS_FLOAT);
if (v != NULL) {
v->value.f = (float) xmlXPathNAN;
} else {
xmlSchemaFreeValue(v);
goto error;
}
} else {
v = xmlSchemaNewValue(XML_SCHEMAS_DOUBLE);
if (v != NULL) {
v->value.d = xmlXPathNAN;
} else {
xmlSchemaFreeValue(v);
goto error;
}
}
*val = v;
}
goto return0;
}
if (*cur == '-') {
neg = 1;
cur++;
}
if ((cur[0] == 'I') && (cur[1] == 'N') && (cur[2] == 'F')) {
cur += 3;
if (*cur != 0)
goto return1;
if (val != NULL) {
if (type == xmlSchemaTypeFloatDef) {
v = xmlSchemaNewValue(XML_SCHEMAS_FLOAT);
if (v != NULL) {
if (neg)
v->value.f = (float) xmlXPathNINF;
else
v->value.f = (float) xmlXPathPINF;
} else {
xmlSchemaFreeValue(v);
goto error;
}
} else {
v = xmlSchemaNewValue(XML_SCHEMAS_DOUBLE);
if (v != NULL) {
if (neg)
v->value.d = xmlXPathNINF;
else
v->value.d = xmlXPathPINF;
} else {
xmlSchemaFreeValue(v);
goto error;
}
}
*val = v;
}
goto return0;
}
if ((neg == 0) && (*cur == '+'))
cur++;
if ((cur[0] == 0) || (cur[0] == '+') || (cur[0] == '-'))
goto return1;
while ((*cur >= '0') && (*cur <= '9')) {
cur++;
digits_before++;
}
if (*cur == '.') {
cur++;
while ((*cur >= '0') && (*cur <= '9')) {
cur++;
digits_after++;
}
}
if ((digits_before == 0) && (digits_after == 0))
goto return1;
if ((*cur == 'e') || (*cur == 'E')) {
cur++;
if ((*cur == '-') || (*cur == '+'))
cur++;
while ((*cur >= '0') && (*cur <= '9'))
cur++;
}
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur != 0)
goto return1;
if (val != NULL) {
if (type == xmlSchemaTypeFloatDef) {
v = xmlSchemaNewValue(XML_SCHEMAS_FLOAT);
if (v != NULL) {
/*
* TODO: sscanf seems not to give the correct
* value for extremely high/low values.
* E.g. "1E-149" results in zero.
*/
if (sscanf((const char *) value, "%f",
&(v->value.f)) == 1) {
*val = v;
} else {
xmlSchemaFreeValue(v);
goto return1;
}
} else {
goto error;
}
} else {
v = xmlSchemaNewValue(XML_SCHEMAS_DOUBLE);
if (v != NULL) {
/*
* TODO: sscanf seems not to give the correct
* value for extremely high/low values.
*/
if (sscanf((const char *) value, "%lf",
&(v->value.d)) == 1) {
*val = v;
} else {
xmlSchemaFreeValue(v);
goto return1;
}
} else {
goto error;
}
}
}
goto return0;
}
case XML_SCHEMAS_BOOLEAN:{
const xmlChar *cur = value;
if (normOnTheFly) {
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur == '0') {
ret = 0;
cur++;
} else if (*cur == '1') {
ret = 1;
cur++;
} else if (*cur == 't') {
cur++;
if ((*cur++ == 'r') && (*cur++ == 'u') &&
(*cur++ == 'e')) {
ret = 1;
} else
goto return1;
} else if (*cur == 'f') {
cur++;
if ((*cur++ == 'a') && (*cur++ == 'l') &&
(*cur++ == 's') && (*cur++ == 'e')) {
ret = 0;
} else
goto return1;
} else
goto return1;
if (*cur != 0) {
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur != 0)
goto return1;
}
} else {
if ((cur[0] == '0') && (cur[1] == 0))
ret = 0;
else if ((cur[0] == '1') && (cur[1] == 0))
ret = 1;
else if ((cur[0] == 't') && (cur[1] == 'r')
&& (cur[2] == 'u') && (cur[3] == 'e')
&& (cur[4] == 0))
ret = 1;
else if ((cur[0] == 'f') && (cur[1] == 'a')
&& (cur[2] == 'l') && (cur[3] == 's')
&& (cur[4] == 'e') && (cur[5] == 0))
ret = 0;
else
goto return1;
}
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_BOOLEAN);
if (v != NULL) {
v->value.b = ret;
*val = v;
} else {
goto error;
}
}
goto return0;
}
case XML_SCHEMAS_TOKEN:{
const xmlChar *cur = value;
if (! normOnTheFly) {
while (*cur != 0) {
if ((*cur == 0xd) || (*cur == 0xa) || (*cur == 0x9)) {
goto return1;
} else if (*cur == ' ') {
cur++;
if (*cur == 0)
goto return1;
if (*cur == ' ')
goto return1;
} else {
cur++;
}
}
}
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_TOKEN);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
goto return0;
}
case XML_SCHEMAS_LANGUAGE:
if ((norm == NULL) && (normOnTheFly)) {
norm = xmlSchemaCollapseString(value);
if (norm != NULL)
value = norm;
}
if (xmlSchemaCheckLanguageType(value) == 1) {
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_LANGUAGE);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
goto return0;
}
goto return1;
case XML_SCHEMAS_NMTOKEN:
if (xmlValidateNMToken(value, 1) == 0) {
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_NMTOKEN);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
goto return0;
}
goto return1;
case XML_SCHEMAS_NMTOKENS:
ret = xmlSchemaValAtomicListNode(xmlSchemaTypeNmtokenDef,
value, val, node);
if (ret > 0)
ret = 0;
else
ret = 1;
goto done;
case XML_SCHEMAS_NAME:
ret = xmlValidateName(value, 1);
if ((ret == 0) && (val != NULL) && (value != NULL)) {
v = xmlSchemaNewValue(XML_SCHEMAS_NAME);
if (v != NULL) {
const xmlChar *start = value, *end;
while (IS_BLANK_CH(*start)) start++;
end = start;
while ((*end != 0) && (!IS_BLANK_CH(*end))) end++;
v->value.str = xmlStrndup(start, end - start);
*val = v;
} else {
goto error;
}
}
goto done;
case XML_SCHEMAS_QNAME:{
const xmlChar *uri = NULL;
xmlChar *local = NULL;
ret = xmlValidateQName(value, 1);
if (ret != 0)
goto done;
if (node != NULL) {
xmlChar *prefix;
xmlNsPtr ns;
local = xmlSplitQName2(value, &prefix);
ns = xmlSearchNs(node->doc, node, prefix);
if ((ns == NULL) && (prefix != NULL)) {
xmlFree(prefix);
if (local != NULL)
xmlFree(local);
goto return1;
}
if (ns != NULL)
uri = ns->href;
if (prefix != NULL)
xmlFree(prefix);
}
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_QNAME);
if (v == NULL) {
if (local != NULL)
xmlFree(local);
goto error;
}
if (local != NULL)
v->value.qname.name = local;
else
v->value.qname.name = xmlStrdup(value);
if (uri != NULL)
v->value.qname.uri = xmlStrdup(uri);
*val = v;
} else
if (local != NULL)
xmlFree(local);
goto done;
}
case XML_SCHEMAS_NCNAME:
ret = xmlValidateNCName(value, 1);
if ((ret == 0) && (val != NULL)) {
v = xmlSchemaNewValue(XML_SCHEMAS_NCNAME);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
goto done;
case XML_SCHEMAS_ID:
ret = xmlValidateNCName(value, 1);
if ((ret == 0) && (val != NULL)) {
v = xmlSchemaNewValue(XML_SCHEMAS_ID);
if (v != NULL) {
v->value.str = xmlStrdup(value);
*val = v;
} else {
goto error;
}
}
if ((ret == 0) && (node != NULL) &&
(node->type == XML_ATTRIBUTE_NODE)) {
xmlAttrPtr attr = (xmlAttrPtr) node;
/*
* NOTE: the IDness might have already be declared in the DTD
*/
if (attr->atype != XML_ATTRIBUTE_ID) {
xmlIDPtr res;
xmlChar *strip;
strip = xmlSchemaStrip(value);
if (strip != NULL) {
res = xmlAddID(NULL, node->doc, strip, attr);
xmlFree(strip);
} else
res = xmlAddID(NULL, node->doc, value, attr);
if (res == NULL) {
ret = 2;
} else {
attr->atype = XML_ATTRIBUTE_ID;
}
}
}
goto done;
case XML_SCHEMAS_IDREF:
ret = xmlValidateNCName(value, 1);
if ((ret == 0) && (val != NULL)) {
v = xmlSchemaNewValue(XML_SCHEMAS_IDREF);
if (v == NULL)
goto error;
v->value.str = xmlStrdup(value);
*val = v;
}
if ((ret == 0) && (node != NULL) &&
(node->type == XML_ATTRIBUTE_NODE)) {
xmlAttrPtr attr = (xmlAttrPtr) node;
xmlChar *strip;
strip = xmlSchemaStrip(value);
if (strip != NULL) {
xmlAddRef(NULL, node->doc, strip, attr);
xmlFree(strip);
} else
xmlAddRef(NULL, node->doc, value, attr);
attr->atype = XML_ATTRIBUTE_IDREF;
}
goto done;
case XML_SCHEMAS_IDREFS:
ret = xmlSchemaValAtomicListNode(xmlSchemaTypeIdrefDef,
value, val, node);
if (ret < 0)
ret = 2;
else
ret = 0;
if ((ret == 0) && (node != NULL) &&
(node->type == XML_ATTRIBUTE_NODE)) {
xmlAttrPtr attr = (xmlAttrPtr) node;
attr->atype = XML_ATTRIBUTE_IDREFS;
}
goto done;
case XML_SCHEMAS_ENTITY:{
xmlChar *strip;
ret = xmlValidateNCName(value, 1);
if ((node == NULL) || (node->doc == NULL))
ret = 3;
if (ret == 0) {
xmlEntityPtr ent;
strip = xmlSchemaStrip(value);
if (strip != NULL) {
ent = xmlGetDocEntity(node->doc, strip);
xmlFree(strip);
} else {
ent = xmlGetDocEntity(node->doc, value);
}
if ((ent == NULL) ||
(ent->etype !=
XML_EXTERNAL_GENERAL_UNPARSED_ENTITY))
ret = 4;
}
if ((ret == 0) && (val != NULL)) {
TODO;
}
if ((ret == 0) && (node != NULL) &&
(node->type == XML_ATTRIBUTE_NODE)) {
xmlAttrPtr attr = (xmlAttrPtr) node;
attr->atype = XML_ATTRIBUTE_ENTITY;
}
goto done;
}
case XML_SCHEMAS_ENTITIES:
if ((node == NULL) || (node->doc == NULL))
goto return3;
ret = xmlSchemaValAtomicListNode(xmlSchemaTypeEntityDef,
value, val, node);
if (ret <= 0)
ret = 1;
else
ret = 0;
if ((ret == 0) && (node != NULL) &&
(node->type == XML_ATTRIBUTE_NODE)) {
xmlAttrPtr attr = (xmlAttrPtr) node;
attr->atype = XML_ATTRIBUTE_ENTITIES;
}
goto done;
case XML_SCHEMAS_NOTATION:{
xmlChar *uri = NULL;
xmlChar *local = NULL;
ret = xmlValidateQName(value, 1);
if ((ret == 0) && (node != NULL)) {
xmlChar *prefix;
local = xmlSplitQName2(value, &prefix);
if (prefix != NULL) {
xmlNsPtr ns;
ns = xmlSearchNs(node->doc, node, prefix);
if (ns == NULL)
ret = 1;
else if (val != NULL)
uri = xmlStrdup(ns->href);
}
if ((local != NULL) && ((val == NULL) || (ret != 0)))
xmlFree(local);
if (prefix != NULL)
xmlFree(prefix);
}
if ((node == NULL) || (node->doc == NULL))
ret = 3;
if (ret == 0) {
ret = xmlValidateNotationUse(NULL, node->doc, value);
if (ret == 1)
ret = 0;
else
ret = 1;
}
if ((ret == 0) && (val != NULL)) {
v = xmlSchemaNewValue(XML_SCHEMAS_NOTATION);
if (v != NULL) {
if (local != NULL)
v->value.qname.name = local;
else
v->value.qname.name = xmlStrdup(value);
if (uri != NULL)
v->value.qname.uri = uri;
*val = v;
} else {
if (local != NULL)
xmlFree(local);
if (uri != NULL)
xmlFree(uri);
goto error;
}
}
goto done;
}
case XML_SCHEMAS_ANYURI:{
if (*value != 0) {
xmlURIPtr uri;
xmlChar *tmpval, *cur;
if ((norm == NULL) && (normOnTheFly)) {
norm = xmlSchemaCollapseString(value);
if (norm != NULL)
value = norm;
}
tmpval = xmlStrdup(value);
if (tmpval == NULL)
goto error;
for (cur = tmpval; *cur; ++cur) {
if (*cur < 32 || *cur >= 127 || *cur == ' ' ||
*cur == '<' || *cur == '>' || *cur == '"' ||
*cur == '{' || *cur == '}' || *cur == '|' ||
*cur == '\\' || *cur == '^' || *cur == '`' ||
*cur == '\'')
*cur = '_';
}
uri = xmlParseURI((const char *) tmpval);
xmlFree(tmpval);
if (uri == NULL)
goto return1;
xmlFreeURI(uri);
}
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_ANYURI);
if (v == NULL)
goto error;
v->value.str = xmlStrdup(value);
*val = v;
}
goto return0;
}
case XML_SCHEMAS_HEXBINARY:{
const xmlChar *cur = value, *start;
xmlChar *base;
int total, i = 0;
if (cur == NULL)
goto return1;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
start = cur;
while (((*cur >= '0') && (*cur <= '9')) ||
((*cur >= 'A') && (*cur <= 'F')) ||
((*cur >= 'a') && (*cur <= 'f'))) {
i++;
cur++;
}
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur != 0)
goto return1;
if ((i % 2) != 0)
goto return1;
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_HEXBINARY);
if (v == NULL)
goto error;
/*
* Copy only the normalized piece.
* CRITICAL TODO: Check this.
*/
cur = xmlStrndup(start, i);
if (cur == NULL) {
xmlSchemaTypeErrMemory(node, "allocating hexbin data");
xmlFree(v);
goto return1;
}
total = i / 2; /* number of octets */
base = (xmlChar *) cur;
while (i-- > 0) {
if (*base >= 'a')
*base = *base - ('a' - 'A');
base++;
}
v->value.hex.str = (xmlChar *) cur;
v->value.hex.total = total;
*val = v;
}
goto return0;
}
case XML_SCHEMAS_BASE64BINARY:{
/* ISSUE:
*
* Ignore all stray characters? (yes, currently)
* Worry about long lines? (no, currently)
*
* rfc2045.txt:
*
* "The encoded output stream must be represented in lines of
* no more than 76 characters each. All line breaks or other
* characters not found in Table 1 must be ignored by decoding
* software. In base64 data, characters other than those in
* Table 1, line breaks, and other white space probably
* indicate a transmission error, about which a warning
* message or even a message rejection might be appropriate
* under some circumstances." */
const xmlChar *cur = value;
xmlChar *base;
int total, i = 0, pad = 0;
if (cur == NULL)
goto return1;
for (; *cur; ++cur) {
int decc;
decc = _xmlSchemaBase64Decode(*cur);
if (decc < 0) ;
else if (decc < 64)
i++;
else
break;
}
for (; *cur; ++cur) {
int decc;
decc = _xmlSchemaBase64Decode(*cur);
if (decc < 0) ;
else if (decc < 64)
goto return1;
if (decc == 64)
pad++;
}
/* rfc2045.txt: "Special processing is performed if fewer than
* 24 bits are available at the end of the data being encoded.
* A full encoding quantum is always completed at the end of a
* body. When fewer than 24 input bits are available in an
* input group, zero bits are added (on the right) to form an
* integral number of 6-bit groups. Padding at the end of the
* data is performed using the "=" character. Since all
* base64 input is an integral number of octets, only the
* following cases can arise: (1) the final quantum of
* encoding input is an integral multiple of 24 bits; here,
* the final unit of encoded output will be an integral
* multiple of indent: Standard input:701: Warning:old style
* assignment ambiguity in "=*". Assuming "= *" 4 characters
* with no "=" padding, (2) the final
* quantum of encoding input is exactly 8 bits; here, the
* final unit of encoded output will be two characters
* followed by two "=" padding characters, or (3) the final
* quantum of encoding input is exactly 16 bits; here, the
* final unit of encoded output will be three characters
* followed by one "=" padding character." */
total = 3 * (i / 4);
if (pad == 0) {
if (i % 4 != 0)
goto return1;
} else if (pad == 1) {
int decc;
if (i % 4 != 3)
goto return1;
for (decc = _xmlSchemaBase64Decode(*cur);
(decc < 0) || (decc > 63);
decc = _xmlSchemaBase64Decode(*cur))
--cur;
/* 16bits in 24bits means 2 pad bits: nnnnnn nnmmmm mmmm00*/
/* 00111100 -> 0x3c */
if (decc & ~0x3c)
goto return1;
total += 2;
} else if (pad == 2) {
int decc;
if (i % 4 != 2)
goto return1;
for (decc = _xmlSchemaBase64Decode(*cur);
(decc < 0) || (decc > 63);
decc = _xmlSchemaBase64Decode(*cur))
--cur;
/* 8bits in 12bits means 4 pad bits: nnnnnn nn0000 */
/* 00110000 -> 0x30 */
if (decc & ~0x30)
goto return1;
total += 1;
} else
goto return1;
if (val != NULL) {
v = xmlSchemaNewValue(XML_SCHEMAS_BASE64BINARY);
if (v == NULL)
goto error;
base =
(xmlChar *) xmlMallocAtomic(i + pad + 1);
if (base == NULL) {
xmlSchemaTypeErrMemory(node, "allocating base64 data");
xmlFree(v);
goto return1;
}
v->value.base64.str = base;
for (cur = value; *cur; ++cur)
if (_xmlSchemaBase64Decode(*cur) >= 0) {
*base = *cur;
++base;
}
*base = 0;
v->value.base64.total = total;
*val = v;
}
goto return0;
}
case XML_SCHEMAS_INTEGER:
case XML_SCHEMAS_PINTEGER:
case XML_SCHEMAS_NPINTEGER:
case XML_SCHEMAS_NINTEGER:
case XML_SCHEMAS_NNINTEGER:{
const xmlChar *cur = value;
unsigned long lo, mi, hi;
int sign = 0;
if (cur == NULL)
goto return1;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur == '-') {
sign = 1;
cur++;
} else if (*cur == '+')
cur++;
ret = xmlSchemaParseUInt(&cur, &lo, &mi, &hi);
if (ret < 0)
goto return1;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur != 0)
goto return1;
if (type->builtInType == XML_SCHEMAS_NPINTEGER) {
if ((sign == 0) &&
((hi != 0) || (mi != 0) || (lo != 0)))
goto return1;
} else if (type->builtInType == XML_SCHEMAS_PINTEGER) {
if (sign == 1)
goto return1;
if ((hi == 0) && (mi == 0) && (lo == 0))
goto return1;
} else if (type->builtInType == XML_SCHEMAS_NINTEGER) {
if (sign == 0)
goto return1;
if ((hi == 0) && (mi == 0) && (lo == 0))
goto return1;
} else if (type->builtInType == XML_SCHEMAS_NNINTEGER) {
if ((sign == 1) &&
((hi != 0) || (mi != 0) || (lo != 0)))
goto return1;
}
if (val != NULL) {
v = xmlSchemaNewValue(type->builtInType);
if (v != NULL) {
if (ret == 0)
ret++;
v->value.decimal.lo = lo;
v->value.decimal.mi = mi;
v->value.decimal.hi = hi;
v->value.decimal.sign = sign;
v->value.decimal.frac = 0;
v->value.decimal.total = ret;
*val = v;
}
}
goto return0;
}
case XML_SCHEMAS_LONG:
case XML_SCHEMAS_BYTE:
case XML_SCHEMAS_SHORT:
case XML_SCHEMAS_INT:{
const xmlChar *cur = value;
unsigned long lo, mi, hi;
int sign = 0;
if (cur == NULL)
goto return1;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur == '-') {
sign = 1;
cur++;
} else if (*cur == '+')
cur++;
ret = xmlSchemaParseUInt(&cur, &lo, &mi, &hi);
if (ret < 0)
goto return1;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur != 0)
goto return1;
if (type->builtInType == XML_SCHEMAS_LONG) {
if (hi >= 922) {
if (hi > 922)
goto return1;
if (mi >= 33720368) {
if (mi > 33720368)
goto return1;
if ((sign == 0) && (lo > 54775807))
goto return1;
if ((sign == 1) && (lo > 54775808))
goto return1;
}
}
} else if (type->builtInType == XML_SCHEMAS_INT) {
if (hi != 0)
goto return1;
if (mi >= 21) {
if (mi > 21)
goto return1;
if ((sign == 0) && (lo > 47483647))
goto return1;
if ((sign == 1) && (lo > 47483648))
goto return1;
}
} else if (type->builtInType == XML_SCHEMAS_SHORT) {
if ((mi != 0) || (hi != 0))
goto return1;
if ((sign == 1) && (lo > 32768))
goto return1;
if ((sign == 0) && (lo > 32767))
goto return1;
} else if (type->builtInType == XML_SCHEMAS_BYTE) {
if ((mi != 0) || (hi != 0))
goto return1;
if ((sign == 1) && (lo > 128))
goto return1;
if ((sign == 0) && (lo > 127))
goto return1;
}
if (val != NULL) {
v = xmlSchemaNewValue(type->builtInType);
if (v != NULL) {
v->value.decimal.lo = lo;
v->value.decimal.mi = mi;
v->value.decimal.hi = hi;
v->value.decimal.sign = sign;
v->value.decimal.frac = 0;
v->value.decimal.total = ret;
*val = v;
}
}
goto return0;
}
case XML_SCHEMAS_UINT:
case XML_SCHEMAS_ULONG:
case XML_SCHEMAS_USHORT:
case XML_SCHEMAS_UBYTE:{
const xmlChar *cur = value;
unsigned long lo, mi, hi;
if (cur == NULL)
goto return1;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
ret = xmlSchemaParseUInt(&cur, &lo, &mi, &hi);
if (ret < 0)
goto return1;
if (normOnTheFly)
while IS_WSP_BLANK_CH(*cur) cur++;
if (*cur != 0)
goto return1;
if (type->builtInType == XML_SCHEMAS_ULONG) {
if (hi >= 1844) {
if (hi > 1844)
goto return1;
if (mi >= 67440737) {
if (mi > 67440737)
goto return1;
if (lo > 9551615)
goto return1;
}
}
} else if (type->builtInType == XML_SCHEMAS_UINT) {
if (hi != 0)
goto return1;
if (mi >= 42) {
if (mi > 42)
goto return1;
if (lo > 94967295)
goto return1;
}
} else if (type->builtInType == XML_SCHEMAS_USHORT) {
if ((mi != 0) || (hi != 0))
goto return1;
if (lo > 65535)
goto return1;
} else if (type->builtInType == XML_SCHEMAS_UBYTE) {
if ((mi != 0) || (hi != 0))
goto return1;
if (lo > 255)
goto return1;
}
if (val != NULL) {
v = xmlSchemaNewValue(type->builtInType);
if (v != NULL) {
v->value.decimal.lo = lo;
v->value.decimal.mi = mi;
v->value.decimal.hi = hi;
v->value.decimal.sign = 0;
v->value.decimal.frac = 0;
v->value.decimal.total = ret;
*val = v;
}
}
goto return0;
}
}
done:
if (norm != NULL)
xmlFree(norm);
return (ret);
return3:
if (norm != NULL)
xmlFree(norm);
return (3);
return1:
if (norm != NULL)
xmlFree(norm);
return (1);
return0:
if (norm != NULL)
xmlFree(norm);
return (0);
error:
if (norm != NULL)
xmlFree(norm);
return (-1);
}
/**
* xmlSchemaValPredefTypeNode:
* @type: the predefined type
* @value: the value to check
* @val: the return computed value
* @node: the node containing the value
*
* Check that a value conforms to the lexical space of the predefined type.
* if true a value is computed and returned in @val.
*
* Returns 0 if this validates, a positive error code number otherwise
* and -1 in case of internal or API error.
*/
int
xmlSchemaValPredefTypeNode(xmlSchemaTypePtr type, const xmlChar *value,
xmlSchemaValPtr *val, xmlNodePtr node) {
return(xmlSchemaValAtomicType(type, value, val, node, 0,
XML_SCHEMA_WHITESPACE_UNKNOWN, 1, 1, 0));
}
/**
* xmlSchemaValPredefTypeNodeNoNorm:
* @type: the predefined type
* @value: the value to check
* @val: the return computed value
* @node: the node containing the value
*
* Check that a value conforms to the lexical space of the predefined type.
* if true a value is computed and returned in @val.
* This one does apply any normalization to the value.
*
* Returns 0 if this validates, a positive error code number otherwise
* and -1 in case of internal or API error.
*/
int
xmlSchemaValPredefTypeNodeNoNorm(xmlSchemaTypePtr type, const xmlChar *value,
xmlSchemaValPtr *val, xmlNodePtr node) {
return(xmlSchemaValAtomicType(type, value, val, node, 1,
XML_SCHEMA_WHITESPACE_UNKNOWN, 1, 0, 1));
}
/**
* xmlSchemaValidatePredefinedType:
* @type: the predefined type
* @value: the value to check
* @val: the return computed value
*
* Check that a value conforms to the lexical space of the predefined type.
* if true a value is computed and returned in @val.
*
* Returns 0 if this validates, a positive error code number otherwise
* and -1 in case of internal or API error.
*/
int
xmlSchemaValidatePredefinedType(xmlSchemaTypePtr type, const xmlChar *value,
xmlSchemaValPtr *val) {
return(xmlSchemaValPredefTypeNode(type, value, val, NULL));
}
/**
* xmlSchemaCompareDecimals:
* @x: a first decimal value
* @y: a second decimal value
*
* Compare 2 decimals
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y and -2 in case of error
*/
static int
xmlSchemaCompareDecimals(xmlSchemaValPtr x, xmlSchemaValPtr y)
{
xmlSchemaValPtr swp;
int order = 1, integx, integy, dlen;
unsigned long hi, mi, lo;
/*
* First test: If x is -ve and not zero
*/
if ((x->value.decimal.sign) &&
((x->value.decimal.lo != 0) ||
(x->value.decimal.mi != 0) ||
(x->value.decimal.hi != 0))) {
/*
* Then if y is -ve and not zero reverse the compare
*/
if ((y->value.decimal.sign) &&
((y->value.decimal.lo != 0) ||
(y->value.decimal.mi != 0) ||
(y->value.decimal.hi != 0)))
order = -1;
/*
* Otherwise (y >= 0) we have the answer
*/
else
return (-1);
/*
* If x is not -ve and y is -ve we have the answer
*/
} else if ((y->value.decimal.sign) &&
((y->value.decimal.lo != 0) ||
(y->value.decimal.mi != 0) ||
(y->value.decimal.hi != 0))) {
return (1);
}
/*
* If it's not simply determined by a difference in sign,
* then we need to compare the actual values of the two nums.
* To do this, we start by looking at the integral parts.
* If the number of integral digits differ, then we have our
* answer.
*/
integx = x->value.decimal.total - x->value.decimal.frac;
integy = y->value.decimal.total - y->value.decimal.frac;
/*
* NOTE: We changed the "total" for values like "0.1"
* (or "-0.1" or ".1") to be 1, which was 2 previously.
* Therefore the special case, when such values are
* compared with 0, needs to be handled separately;
* otherwise a zero would be recognized incorrectly as
* greater than those values. This has the nice side effect
* that we gain an overall optimized comparison with zeroes.
* Note that a "0" has a "total" of 1 already.
*/
if (integx == 1) {
if (x->value.decimal.lo == 0) {
if (integy != 1)
return -order;
else if (y->value.decimal.lo != 0)
return -order;
else
return(0);
}
}
if (integy == 1) {
if (y->value.decimal.lo == 0) {
if (integx != 1)
return order;
else if (x->value.decimal.lo != 0)
return order;
else
return(0);
}
}
if (integx > integy)
return order;
else if (integy > integx)
return -order;
/*
* If the number of integral digits is the same for both numbers,
* then things get a little more complicated. We need to "normalize"
* the numbers in order to properly compare them. To do this, we
* look at the total length of each number (length => number of
* significant digits), and divide the "shorter" by 10 (decreasing
* the length) until they are of equal length.
*/
dlen = x->value.decimal.total - y->value.decimal.total;
if (dlen < 0) { /* y has more digits than x */
swp = x;
hi = y->value.decimal.hi;
mi = y->value.decimal.mi;
lo = y->value.decimal.lo;
dlen = -dlen;
order = -order;
} else { /* x has more digits than y */
swp = y;
hi = x->value.decimal.hi;
mi = x->value.decimal.mi;
lo = x->value.decimal.lo;
}
while (dlen > 8) { /* in effect, right shift by 10**8 */
lo = mi;
mi = hi;
hi = 0;
dlen -= 8;
}
while (dlen > 0) {
unsigned long rem1, rem2;
rem1 = (hi % 10) * 100000000L;
hi = hi / 10;
rem2 = (mi % 10) * 100000000L;
mi = (mi + rem1) / 10;
lo = (lo + rem2) / 10;
dlen--;
}
if (hi > swp->value.decimal.hi) {
return order;
} else if (hi == swp->value.decimal.hi) {
if (mi > swp->value.decimal.mi) {
return order;
} else if (mi == swp->value.decimal.mi) {
if (lo > swp->value.decimal.lo) {
return order;
} else if (lo == swp->value.decimal.lo) {
if (x->value.decimal.total == y->value.decimal.total) {
return 0;
} else {
return order;
}
}
}
}
return -order;
}
/**
* xmlSchemaCompareDurations:
* @x: a first duration value
* @y: a second duration value
*
* Compare 2 durations
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, 2 if x <> y, and -2 in
* case of error
*/
static int
xmlSchemaCompareDurations(xmlSchemaValPtr x, xmlSchemaValPtr y)
{
long carry, mon, day;
double sec;
int invert = 1;
long xmon, xday, myear, minday, maxday;
static const long dayRange [2][12] = {
{ 0, 28, 59, 89, 120, 150, 181, 212, 242, 273, 303, 334, },
{ 0, 31, 62, 92, 123, 153, 184, 215, 245, 276, 306, 337} };
if ((x == NULL) || (y == NULL))
return -2;
/* months */
mon = x->value.dur.mon - y->value.dur.mon;
/* seconds */
sec = x->value.dur.sec - y->value.dur.sec;
carry = (long)(sec / SECS_PER_DAY);
sec -= ((double)carry) * SECS_PER_DAY;
/* days */
day = x->value.dur.day - y->value.dur.day + carry;
/* easy test */
if (mon == 0) {
if (day == 0)
if (sec == 0.0)
return 0;
else if (sec < 0.0)
return -1;
else
return 1;
else if (day < 0)
return -1;
else
return 1;
}
if (mon > 0) {
if ((day >= 0) && (sec >= 0.0))
return 1;
else {
xmon = mon;
xday = -day;
}
} else if ((day <= 0) && (sec <= 0.0)) {
return -1;
} else {
invert = -1;
xmon = -mon;
xday = day;
}
myear = xmon / 12;
if (myear == 0) {
minday = 0;
maxday = 0;
} else {
if (myear > LONG_MAX / 366)
return -2;
/* FIXME: This doesn't take leap year exceptions every 100/400 years
into account. */
maxday = 365 * myear + (myear + 3) / 4;
/* FIXME: Needs to be calculated separately */
minday = maxday - 1;
}
xmon = xmon % 12;
minday += dayRange[0][xmon];
maxday += dayRange[1][xmon];
if ((maxday == minday) && (maxday == xday))
return(0); /* can this really happen ? */
if (maxday < xday)
return(-invert);
if (minday > xday)
return(invert);
/* indeterminate */
return 2;
}
/*
* macros for adding date/times and durations
*/
#define FQUOTIENT(a,b) (floor(((double)a/(double)b)))
#define MODULO(a,b) (a - FQUOTIENT(a,b) * b)
#define FQUOTIENT_RANGE(a,low,high) (FQUOTIENT((a-low),(high-low)))
#define MODULO_RANGE(a,low,high) ((MODULO((a-low),(high-low)))+low)
/**
* xmlSchemaDupVal:
* @v: the #xmlSchemaValPtr value to duplicate
*
* Makes a copy of @v. The calling program is responsible for freeing
* the returned value.
*
* returns a pointer to a duplicated #xmlSchemaValPtr or NULL if error.
*/
static xmlSchemaValPtr
xmlSchemaDupVal (xmlSchemaValPtr v)
{
xmlSchemaValPtr ret = xmlSchemaNewValue(v->type);
if (ret == NULL)
return NULL;
memcpy(ret, v, sizeof(xmlSchemaVal));
ret->next = NULL;
return ret;
}
/**
* xmlSchemaCopyValue:
* @val: the precomputed value to be copied
*
* Copies the precomputed value. This duplicates any string within.
*
* Returns the copy or NULL if a copy for a data-type is not implemented.
*/
xmlSchemaValPtr
xmlSchemaCopyValue(xmlSchemaValPtr val)
{
xmlSchemaValPtr ret = NULL, prev = NULL, cur;
/*
* Copy the string values.
*/
while (val != NULL) {
switch (val->type) {
case XML_SCHEMAS_ANYTYPE:
case XML_SCHEMAS_IDREFS:
case XML_SCHEMAS_ENTITIES:
case XML_SCHEMAS_NMTOKENS:
xmlSchemaFreeValue(ret);
return (NULL);
case XML_SCHEMAS_ANYSIMPLETYPE:
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_NORMSTRING:
case XML_SCHEMAS_TOKEN:
case XML_SCHEMAS_LANGUAGE:
case XML_SCHEMAS_NAME:
case XML_SCHEMAS_NCNAME:
case XML_SCHEMAS_ID:
case XML_SCHEMAS_IDREF:
case XML_SCHEMAS_ENTITY:
case XML_SCHEMAS_NMTOKEN:
case XML_SCHEMAS_ANYURI:
cur = xmlSchemaDupVal(val);
if (val->value.str != NULL)
cur->value.str = xmlStrdup(BAD_CAST val->value.str);
break;
case XML_SCHEMAS_QNAME:
case XML_SCHEMAS_NOTATION:
cur = xmlSchemaDupVal(val);
if (val->value.qname.name != NULL)
cur->value.qname.name =
xmlStrdup(BAD_CAST val->value.qname.name);
if (val->value.qname.uri != NULL)
cur->value.qname.uri =
xmlStrdup(BAD_CAST val->value.qname.uri);
break;
case XML_SCHEMAS_HEXBINARY:
cur = xmlSchemaDupVal(val);
if (val->value.hex.str != NULL)
cur->value.hex.str = xmlStrdup(BAD_CAST val->value.hex.str);
break;
case XML_SCHEMAS_BASE64BINARY:
cur = xmlSchemaDupVal(val);
if (val->value.base64.str != NULL)
cur->value.base64.str =
xmlStrdup(BAD_CAST val->value.base64.str);
break;
default:
cur = xmlSchemaDupVal(val);
break;
}
if (ret == NULL)
ret = cur;
else
prev->next = cur;
prev = cur;
val = val->next;
}
return (ret);
}
/**
* _xmlSchemaDateAdd:
* @dt: an #xmlSchemaValPtr
* @dur: an #xmlSchemaValPtr of type #XS_DURATION
*
* Compute a new date/time from @dt and @dur. This function assumes @dt
* is either #XML_SCHEMAS_DATETIME, #XML_SCHEMAS_DATE, #XML_SCHEMAS_GYEARMONTH,
* or #XML_SCHEMAS_GYEAR. The returned #xmlSchemaVal is the same type as
* @dt. The calling program is responsible for freeing the returned value.
*
* Returns a pointer to a new #xmlSchemaVal or NULL if error.
*/
static xmlSchemaValPtr
_xmlSchemaDateAdd (xmlSchemaValPtr dt, xmlSchemaValPtr dur)
{
xmlSchemaValPtr ret, tmp;
long carry, tempdays, temp;
xmlSchemaValDatePtr r, d;
xmlSchemaValDurationPtr u;
if ((dt == NULL) || (dur == NULL))
return NULL;
ret = xmlSchemaNewValue(dt->type);
if (ret == NULL)
return NULL;
/* make a copy so we don't alter the original value */
tmp = xmlSchemaDupVal(dt);
if (tmp == NULL) {
xmlSchemaFreeValue(ret);
return NULL;
}
r = &(ret->value.date);
d = &(tmp->value.date);
u = &(dur->value.dur);
/* normalization */
if (d->mon == 0)
d->mon = 1;
/* normalize for time zone offset */
u->sec -= (d->tzo * 60);
d->tzo = 0;
/* normalization */
if (d->day == 0)
d->day = 1;
/* month */
carry = d->mon + u->mon;
r->mon = (unsigned int) MODULO_RANGE(carry, 1, 13);
carry = (long) FQUOTIENT_RANGE(carry, 1, 13);
/* year (may be modified later) */
r->year = d->year + carry;
if (r->year == 0) {
if (d->year > 0)
r->year--;
else
r->year++;
}
/* time zone */
r->tzo = d->tzo;
r->tz_flag = d->tz_flag;
/* seconds */
r->sec = d->sec + u->sec;
carry = (long) FQUOTIENT((long)r->sec, 60);
if (r->sec != 0.0) {
r->sec = MODULO(r->sec, 60.0);
}
/* minute */
carry += d->min;
r->min = (unsigned int) MODULO(carry, 60);
carry = (long) FQUOTIENT(carry, 60);
/* hours */
carry += d->hour;
r->hour = (unsigned int) MODULO(carry, 24);
carry = (long)FQUOTIENT(carry, 24);
/*
* days
* Note we use tempdays because the temporary values may need more
* than 5 bits
*/
if ((VALID_YEAR(r->year)) && (VALID_MONTH(r->mon)) &&
(d->day > MAX_DAYINMONTH(r->year, r->mon)))
tempdays = MAX_DAYINMONTH(r->year, r->mon);
else if (d->day < 1)
tempdays = 1;
else
tempdays = d->day;
tempdays += u->day + carry;
while (1) {
if (tempdays < 1) {
long tmon = (long) MODULO_RANGE((int)r->mon-1, 1, 13);
long tyr = r->year + (long)FQUOTIENT_RANGE((int)r->mon-1, 1, 13);
if (tyr == 0)
tyr--;
/*
* Coverity detected an overrun in daysInMonth
* of size 12 at position 12 with index variable "((r)->mon - 1)"
*/
if (tmon < 1)
tmon = 1;
if (tmon > 12)
tmon = 12;
tempdays += MAX_DAYINMONTH(tyr, tmon);
carry = -1;
} else if (VALID_YEAR(r->year) && VALID_MONTH(r->mon) &&
tempdays > (long) MAX_DAYINMONTH(r->year, r->mon)) {
tempdays = tempdays - MAX_DAYINMONTH(r->year, r->mon);
carry = 1;
} else
break;
temp = r->mon + carry;
r->mon = (unsigned int) MODULO_RANGE(temp, 1, 13);
r->year = r->year + (long) FQUOTIENT_RANGE(temp, 1, 13);
if (r->year == 0) {
if (temp < 1)
r->year--;
else
r->year++;
}
}
r->day = tempdays;
/*
* adjust the date/time type to the date values
*/
if (ret->type != XML_SCHEMAS_DATETIME) {
if ((r->hour) || (r->min) || (r->sec))
ret->type = XML_SCHEMAS_DATETIME;
else if (ret->type != XML_SCHEMAS_DATE) {
if ((r->mon != 1) && (r->day != 1))
ret->type = XML_SCHEMAS_DATE;
else if ((ret->type != XML_SCHEMAS_GYEARMONTH) && (r->mon != 1))
ret->type = XML_SCHEMAS_GYEARMONTH;
}
}
xmlSchemaFreeValue(tmp);
return ret;
}
/**
* xmlSchemaDateNormalize:
* @dt: an #xmlSchemaValPtr of a date/time type value.
* @offset: number of seconds to adjust @dt by.
*
* Normalize @dt to GMT time. The @offset parameter is subtracted from
* the return value is a time-zone offset is present on @dt.
*
* Returns a normalized copy of @dt or NULL if error.
*/
static xmlSchemaValPtr
xmlSchemaDateNormalize (xmlSchemaValPtr dt, double offset)
{
xmlSchemaValPtr dur, ret;
if (dt == NULL)
return NULL;
if (((dt->type != XML_SCHEMAS_TIME) &&
(dt->type != XML_SCHEMAS_DATETIME) &&
(dt->type != XML_SCHEMAS_DATE)) || (dt->value.date.tzo == 0))
return xmlSchemaDupVal(dt);
dur = xmlSchemaNewValue(XML_SCHEMAS_DURATION);
if (dur == NULL)
return NULL;
dur->value.date.sec -= offset;
ret = _xmlSchemaDateAdd(dt, dur);
if (ret == NULL)
return NULL;
xmlSchemaFreeValue(dur);
/* ret->value.date.tzo = 0; */
return ret;
}
/**
* _xmlSchemaDateCastYMToDays:
* @dt: an #xmlSchemaValPtr
*
* Convert mon and year of @dt to total number of days. Take the
* number of years since (or before) 1 AD and add the number of leap
* years. This is a function because negative
* years must be handled a little differently and there is no zero year.
*
* Returns number of days.
*/
static long
_xmlSchemaDateCastYMToDays (const xmlSchemaValPtr dt)
{
long ret;
int mon;
mon = dt->value.date.mon;
if (mon <= 0) mon = 1; /* normalization */
if (dt->value.date.year <= 0)
ret = (dt->value.date.year * 365) +
(((dt->value.date.year+1)/4)-((dt->value.date.year+1)/100)+
((dt->value.date.year+1)/400)) +
DAY_IN_YEAR(0, mon, dt->value.date.year);
else
ret = ((dt->value.date.year-1) * 365) +
(((dt->value.date.year-1)/4)-((dt->value.date.year-1)/100)+
((dt->value.date.year-1)/400)) +
DAY_IN_YEAR(0, mon, dt->value.date.year);
return ret;
}
/**
* TIME_TO_NUMBER:
* @dt: an #xmlSchemaValPtr
*
* Calculates the number of seconds in the time portion of @dt.
*
* Returns seconds.
*/
#define TIME_TO_NUMBER(dt) \
((double)((dt->value.date.hour * SECS_PER_HOUR) + \
(dt->value.date.min * SECS_PER_MIN) + \
(dt->value.date.tzo * SECS_PER_MIN)) + \
dt->value.date.sec)
/**
* xmlSchemaCompareDates:
* @x: a first date/time value
* @y: a second date/time value
*
* Compare 2 date/times
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, 2 if x <> y, and -2 in
* case of error
*/
static int
xmlSchemaCompareDates (xmlSchemaValPtr x, xmlSchemaValPtr y)
{
unsigned char xmask, ymask, xor_mask, and_mask;
xmlSchemaValPtr p1, p2, q1, q2;
long p1d, p2d, q1d, q2d;
if ((x == NULL) || (y == NULL))
return -2;
if ((x->value.date.year > LONG_MAX / 366) ||
(x->value.date.year < LONG_MIN / 366) ||
(y->value.date.year > LONG_MAX / 366) ||
(y->value.date.year < LONG_MIN / 366)) {
/* Possible overflow when converting to days. */
return -2;
}
if (x->value.date.tz_flag) {
if (!y->value.date.tz_flag) {
p1 = xmlSchemaDateNormalize(x, 0);
if (p1 == NULL)
return -2;
p1d = _xmlSchemaDateCastYMToDays(p1) + p1->value.date.day;
/* normalize y + 14:00 */
q1 = xmlSchemaDateNormalize(y, (14 * SECS_PER_HOUR));
if (q1 == NULL) {
xmlSchemaFreeValue(p1);
return -2;
}
q1d = _xmlSchemaDateCastYMToDays(q1) + q1->value.date.day;
if (p1d < q1d) {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
return -1;
} else if (p1d == q1d) {
double sec;
sec = TIME_TO_NUMBER(p1) - TIME_TO_NUMBER(q1);
if (sec < 0.0) {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
return -1;
} else {
int ret = 0;
/* normalize y - 14:00 */
q2 = xmlSchemaDateNormalize(y, -(14 * SECS_PER_HOUR));
if (q2 == NULL) {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
return -2;
}
q2d = _xmlSchemaDateCastYMToDays(q2) + q2->value.date.day;
if (p1d > q2d)
ret = 1;
else if (p1d == q2d) {
sec = TIME_TO_NUMBER(p1) - TIME_TO_NUMBER(q2);
if (sec > 0.0)
ret = 1;
else
ret = 2; /* indeterminate */
}
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
xmlSchemaFreeValue(q2);
if (ret != 0)
return(ret);
}
} else {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
}
}
} else if (y->value.date.tz_flag) {
q1 = xmlSchemaDateNormalize(y, 0);
if (q1 == NULL)
return -2;
q1d = _xmlSchemaDateCastYMToDays(q1) + q1->value.date.day;
/* normalize x - 14:00 */
p1 = xmlSchemaDateNormalize(x, -(14 * SECS_PER_HOUR));
if (p1 == NULL) {
xmlSchemaFreeValue(q1);
return -2;
}
p1d = _xmlSchemaDateCastYMToDays(p1) + p1->value.date.day;
if (p1d < q1d) {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
return -1;
} else if (p1d == q1d) {
double sec;
sec = TIME_TO_NUMBER(p1) - TIME_TO_NUMBER(q1);
if (sec < 0.0) {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
return -1;
} else {
int ret = 0;
/* normalize x + 14:00 */
p2 = xmlSchemaDateNormalize(x, (14 * SECS_PER_HOUR));
if (p2 == NULL) {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
return -2;
}
p2d = _xmlSchemaDateCastYMToDays(p2) + p2->value.date.day;
if (p2d > q1d) {
ret = 1;
} else if (p2d == q1d) {
sec = TIME_TO_NUMBER(p2) - TIME_TO_NUMBER(q1);
if (sec > 0.0)
ret = 1;
else
ret = 2; /* indeterminate */
}
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
xmlSchemaFreeValue(p2);
if (ret != 0)
return(ret);
}
} else {
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
}
}
/*
* if the same type then calculate the difference
*/
if (x->type == y->type) {
int ret = 0;
q1 = xmlSchemaDateNormalize(y, 0);
if (q1 == NULL)
return -2;
q1d = _xmlSchemaDateCastYMToDays(q1) + q1->value.date.day;
p1 = xmlSchemaDateNormalize(x, 0);
if (p1 == NULL) {
xmlSchemaFreeValue(q1);
return -2;
}
p1d = _xmlSchemaDateCastYMToDays(p1) + p1->value.date.day;
if (p1d < q1d) {
ret = -1;
} else if (p1d > q1d) {
ret = 1;
} else {
double sec;
sec = TIME_TO_NUMBER(p1) - TIME_TO_NUMBER(q1);
if (sec < 0.0)
ret = -1;
else if (sec > 0.0)
ret = 1;
}
xmlSchemaFreeValue(p1);
xmlSchemaFreeValue(q1);
return(ret);
}
switch (x->type) {
case XML_SCHEMAS_DATETIME:
xmask = 0xf;
break;
case XML_SCHEMAS_DATE:
xmask = 0x7;
break;
case XML_SCHEMAS_GYEAR:
xmask = 0x1;
break;
case XML_SCHEMAS_GMONTH:
xmask = 0x2;
break;
case XML_SCHEMAS_GDAY:
xmask = 0x3;
break;
case XML_SCHEMAS_GYEARMONTH:
xmask = 0x3;
break;
case XML_SCHEMAS_GMONTHDAY:
xmask = 0x6;
break;
case XML_SCHEMAS_TIME:
xmask = 0x8;
break;
default:
xmask = 0;
break;
}
switch (y->type) {
case XML_SCHEMAS_DATETIME:
ymask = 0xf;
break;
case XML_SCHEMAS_DATE:
ymask = 0x7;
break;
case XML_SCHEMAS_GYEAR:
ymask = 0x1;
break;
case XML_SCHEMAS_GMONTH:
ymask = 0x2;
break;
case XML_SCHEMAS_GDAY:
ymask = 0x3;
break;
case XML_SCHEMAS_GYEARMONTH:
ymask = 0x3;
break;
case XML_SCHEMAS_GMONTHDAY:
ymask = 0x6;
break;
case XML_SCHEMAS_TIME:
ymask = 0x8;
break;
default:
ymask = 0;
break;
}
xor_mask = xmask ^ ymask; /* mark type differences */
and_mask = xmask & ymask; /* mark field specification */
/* year */
if (xor_mask & 1)
return 2; /* indeterminate */
else if (and_mask & 1) {
if (x->value.date.year < y->value.date.year)
return -1;
else if (x->value.date.year > y->value.date.year)
return 1;
}
/* month */
if (xor_mask & 2)
return 2; /* indeterminate */
else if (and_mask & 2) {
if (x->value.date.mon < y->value.date.mon)
return -1;
else if (x->value.date.mon > y->value.date.mon)
return 1;
}
/* day */
if (xor_mask & 4)
return 2; /* indeterminate */
else if (and_mask & 4) {
if (x->value.date.day < y->value.date.day)
return -1;
else if (x->value.date.day > y->value.date.day)
return 1;
}
/* time */
if (xor_mask & 8)
return 2; /* indeterminate */
else if (and_mask & 8) {
if (x->value.date.hour < y->value.date.hour)
return -1;
else if (x->value.date.hour > y->value.date.hour)
return 1;
else if (x->value.date.min < y->value.date.min)
return -1;
else if (x->value.date.min > y->value.date.min)
return 1;
else if (x->value.date.sec < y->value.date.sec)
return -1;
else if (x->value.date.sec > y->value.date.sec)
return 1;
}
return 0;
}
/**
* xmlSchemaComparePreserveReplaceStrings:
* @x: a first string value
* @y: a second string value
* @invert: inverts the result if x < y or x > y.
*
* Compare 2 string for their normalized values.
* @x is a string with whitespace of "preserve", @y is
* a string with a whitespace of "replace". I.e. @x could
* be an "xsd:string" and @y an "xsd:normalizedString".
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, and -2 in
* case of error
*/
static int
xmlSchemaComparePreserveReplaceStrings(const xmlChar *x,
const xmlChar *y,
int invert)
{
int tmp;
while ((*x != 0) && (*y != 0)) {
if (IS_WSP_REPLACE_CH(*y)) {
if (! IS_WSP_SPACE_CH(*x)) {
if ((*x - 0x20) < 0) {
if (invert)
return(1);
else
return(-1);
} else {
if (invert)
return(-1);
else
return(1);
}
}
} else {
tmp = *x - *y;
if (tmp < 0) {
if (invert)
return(1);
else
return(-1);
}
if (tmp > 0) {
if (invert)
return(-1);
else
return(1);
}
}
x++;
y++;
}
if (*x != 0) {
if (invert)
return(-1);
else
return(1);
}
if (*y != 0) {
if (invert)
return(1);
else
return(-1);
}
return(0);
}
/**
* xmlSchemaComparePreserveCollapseStrings:
* @x: a first string value
* @y: a second string value
*
* Compare 2 string for their normalized values.
* @x is a string with whitespace of "preserve", @y is
* a string with a whitespace of "collapse". I.e. @x could
* be an "xsd:string" and @y an "xsd:normalizedString".
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, and -2 in
* case of error
*/
static int
xmlSchemaComparePreserveCollapseStrings(const xmlChar *x,
const xmlChar *y,
int invert)
{
int tmp;
/*
* Skip leading blank chars of the collapsed string.
*/
while IS_WSP_BLANK_CH(*y)
y++;
while ((*x != 0) && (*y != 0)) {
if IS_WSP_BLANK_CH(*y) {
if (! IS_WSP_SPACE_CH(*x)) {
/*
* The yv character would have been replaced to 0x20.
*/
if ((*x - 0x20) < 0) {
if (invert)
return(1);
else
return(-1);
} else {
if (invert)
return(-1);
else
return(1);
}
}
x++;
y++;
/*
* Skip contiguous blank chars of the collapsed string.
*/
while IS_WSP_BLANK_CH(*y)
y++;
} else {
tmp = *x++ - *y++;
if (tmp < 0) {
if (invert)
return(1);
else
return(-1);
}
if (tmp > 0) {
if (invert)
return(-1);
else
return(1);
}
}
}
if (*x != 0) {
if (invert)
return(-1);
else
return(1);
}
if (*y != 0) {
/*
* Skip trailing blank chars of the collapsed string.
*/
while IS_WSP_BLANK_CH(*y)
y++;
if (*y != 0) {
if (invert)
return(1);
else
return(-1);
}
}
return(0);
}
/**
* xmlSchemaComparePreserveCollapseStrings:
* @x: a first string value
* @y: a second string value
*
* Compare 2 string for their normalized values.
* @x is a string with whitespace of "preserve", @y is
* a string with a whitespace of "collapse". I.e. @x could
* be an "xsd:string" and @y an "xsd:normalizedString".
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, and -2 in
* case of error
*/
static int
xmlSchemaCompareReplaceCollapseStrings(const xmlChar *x,
const xmlChar *y,
int invert)
{
int tmp;
/*
* Skip leading blank chars of the collapsed string.
*/
while IS_WSP_BLANK_CH(*y)
y++;
while ((*x != 0) && (*y != 0)) {
if IS_WSP_BLANK_CH(*y) {
if (! IS_WSP_BLANK_CH(*x)) {
/*
* The yv character would have been replaced to 0x20.
*/
if ((*x - 0x20) < 0) {
if (invert)
return(1);
else
return(-1);
} else {
if (invert)
return(-1);
else
return(1);
}
}
x++;
y++;
/*
* Skip contiguous blank chars of the collapsed string.
*/
while IS_WSP_BLANK_CH(*y)
y++;
} else {
if IS_WSP_BLANK_CH(*x) {
/*
* The xv character would have been replaced to 0x20.
*/
if ((0x20 - *y) < 0) {
if (invert)
return(1);
else
return(-1);
} else {
if (invert)
return(-1);
else
return(1);
}
}
tmp = *x++ - *y++;
if (tmp < 0)
return(-1);
if (tmp > 0)
return(1);
}
}
if (*x != 0) {
if (invert)
return(-1);
else
return(1);
}
if (*y != 0) {
/*
* Skip trailing blank chars of the collapsed string.
*/
while IS_WSP_BLANK_CH(*y)
y++;
if (*y != 0) {
if (invert)
return(1);
else
return(-1);
}
}
return(0);
}
/**
* xmlSchemaCompareReplacedStrings:
* @x: a first string value
* @y: a second string value
*
* Compare 2 string for their normalized values.
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, and -2 in
* case of error
*/
static int
xmlSchemaCompareReplacedStrings(const xmlChar *x,
const xmlChar *y)
{
int tmp;
while ((*x != 0) && (*y != 0)) {
if IS_WSP_BLANK_CH(*y) {
if (! IS_WSP_BLANK_CH(*x)) {
if ((*x - 0x20) < 0)
return(-1);
else
return(1);
}
} else {
if IS_WSP_BLANK_CH(*x) {
if ((0x20 - *y) < 0)
return(-1);
else
return(1);
}
tmp = *x - *y;
if (tmp < 0)
return(-1);
if (tmp > 0)
return(1);
}
x++;
y++;
}
if (*x != 0)
return(1);
if (*y != 0)
return(-1);
return(0);
}
/**
* xmlSchemaCompareNormStrings:
* @x: a first string value
* @y: a second string value
*
* Compare 2 string for their normalized values.
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, and -2 in
* case of error
*/
static int
xmlSchemaCompareNormStrings(const xmlChar *x,
const xmlChar *y) {
int tmp;
while (IS_BLANK_CH(*x)) x++;
while (IS_BLANK_CH(*y)) y++;
while ((*x != 0) && (*y != 0)) {
if (IS_BLANK_CH(*x)) {
if (!IS_BLANK_CH(*y)) {
tmp = *x - *y;
return(tmp);
}
while (IS_BLANK_CH(*x)) x++;
while (IS_BLANK_CH(*y)) y++;
} else {
tmp = *x++ - *y++;
if (tmp < 0)
return(-1);
if (tmp > 0)
return(1);
}
}
if (*x != 0) {
while (IS_BLANK_CH(*x)) x++;
if (*x != 0)
return(1);
}
if (*y != 0) {
while (IS_BLANK_CH(*y)) y++;
if (*y != 0)
return(-1);
}
return(0);
}
/**
* xmlSchemaCompareFloats:
* @x: a first float or double value
* @y: a second float or double value
*
* Compare 2 values
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, 2 if x <> y, and -2 in
* case of error
*/
static int
xmlSchemaCompareFloats(xmlSchemaValPtr x, xmlSchemaValPtr y) {
double d1, d2;
if ((x == NULL) || (y == NULL))
return(-2);
/*
* Cast everything to doubles.
*/
if (x->type == XML_SCHEMAS_DOUBLE)
d1 = x->value.d;
else if (x->type == XML_SCHEMAS_FLOAT)
d1 = x->value.f;
else
return(-2);
if (y->type == XML_SCHEMAS_DOUBLE)
d2 = y->value.d;
else if (y->type == XML_SCHEMAS_FLOAT)
d2 = y->value.f;
else
return(-2);
/*
* Check for special cases.
*/
if (xmlXPathIsNaN(d1)) {
if (xmlXPathIsNaN(d2))
return(0);
return(1);
}
if (xmlXPathIsNaN(d2))
return(-1);
if (d1 == xmlXPathPINF) {
if (d2 == xmlXPathPINF)
return(0);
return(1);
}
if (d2 == xmlXPathPINF)
return(-1);
if (d1 == xmlXPathNINF) {
if (d2 == xmlXPathNINF)
return(0);
return(-1);
}
if (d2 == xmlXPathNINF)
return(1);
/*
* basic tests, the last one we should have equality, but
* portability is more important than speed and handling
* NaN or Inf in a portable way is always a challenge, so ...
*/
if (d1 < d2)
return(-1);
if (d1 > d2)
return(1);
if (d1 == d2)
return(0);
return(2);
}
/**
* xmlSchemaCompareValues:
* @x: a first value
* @xvalue: the first value as a string (optional)
* @xwtsp: the whitespace type
* @y: a second value
* @xvalue: the second value as a string (optional)
* @ywtsp: the whitespace type
*
* Compare 2 values
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, 2 if x <> y, 3 if not
* comparable and -2 in case of error
*/
static int
xmlSchemaCompareValuesInternal(xmlSchemaValType xtype,
xmlSchemaValPtr x,
const xmlChar *xvalue,
xmlSchemaWhitespaceValueType xws,
xmlSchemaValType ytype,
xmlSchemaValPtr y,
const xmlChar *yvalue,
xmlSchemaWhitespaceValueType yws)
{
switch (xtype) {
case XML_SCHEMAS_UNKNOWN:
case XML_SCHEMAS_ANYTYPE:
return(-2);
case XML_SCHEMAS_INTEGER:
case XML_SCHEMAS_NPINTEGER:
case XML_SCHEMAS_NINTEGER:
case XML_SCHEMAS_NNINTEGER:
case XML_SCHEMAS_PINTEGER:
case XML_SCHEMAS_INT:
case XML_SCHEMAS_UINT:
case XML_SCHEMAS_LONG:
case XML_SCHEMAS_ULONG:
case XML_SCHEMAS_SHORT:
case XML_SCHEMAS_USHORT:
case XML_SCHEMAS_BYTE:
case XML_SCHEMAS_UBYTE:
case XML_SCHEMAS_DECIMAL:
if ((x == NULL) || (y == NULL))
return(-2);
if (ytype == xtype)
return(xmlSchemaCompareDecimals(x, y));
if ((ytype == XML_SCHEMAS_DECIMAL) ||
(ytype == XML_SCHEMAS_INTEGER) ||
(ytype == XML_SCHEMAS_NPINTEGER) ||
(ytype == XML_SCHEMAS_NINTEGER) ||
(ytype == XML_SCHEMAS_NNINTEGER) ||
(ytype == XML_SCHEMAS_PINTEGER) ||
(ytype == XML_SCHEMAS_INT) ||
(ytype == XML_SCHEMAS_UINT) ||
(ytype == XML_SCHEMAS_LONG) ||
(ytype == XML_SCHEMAS_ULONG) ||
(ytype == XML_SCHEMAS_SHORT) ||
(ytype == XML_SCHEMAS_USHORT) ||
(ytype == XML_SCHEMAS_BYTE) ||
(ytype == XML_SCHEMAS_UBYTE))
return(xmlSchemaCompareDecimals(x, y));
return(-2);
case XML_SCHEMAS_DURATION:
if ((x == NULL) || (y == NULL))
return(-2);
if (ytype == XML_SCHEMAS_DURATION)
return(xmlSchemaCompareDurations(x, y));
return(-2);
case XML_SCHEMAS_TIME:
case XML_SCHEMAS_GDAY:
case XML_SCHEMAS_GMONTH:
case XML_SCHEMAS_GMONTHDAY:
case XML_SCHEMAS_GYEAR:
case XML_SCHEMAS_GYEARMONTH:
case XML_SCHEMAS_DATE:
case XML_SCHEMAS_DATETIME:
if ((x == NULL) || (y == NULL))
return(-2);
if ((ytype == XML_SCHEMAS_DATETIME) ||
(ytype == XML_SCHEMAS_TIME) ||
(ytype == XML_SCHEMAS_GDAY) ||
(ytype == XML_SCHEMAS_GMONTH) ||
(ytype == XML_SCHEMAS_GMONTHDAY) ||
(ytype == XML_SCHEMAS_GYEAR) ||
(ytype == XML_SCHEMAS_DATE) ||
(ytype == XML_SCHEMAS_GYEARMONTH))
return (xmlSchemaCompareDates(x, y));
return (-2);
/*
* Note that we will support comparison of string types against
* anySimpleType as well.
*/
case XML_SCHEMAS_ANYSIMPLETYPE:
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_NORMSTRING:
case XML_SCHEMAS_TOKEN:
case XML_SCHEMAS_LANGUAGE:
case XML_SCHEMAS_NMTOKEN:
case XML_SCHEMAS_NAME:
case XML_SCHEMAS_NCNAME:
case XML_SCHEMAS_ID:
case XML_SCHEMAS_IDREF:
case XML_SCHEMAS_ENTITY:
case XML_SCHEMAS_ANYURI:
{
const xmlChar *xv, *yv;
if (x == NULL)
xv = xvalue;
else
xv = x->value.str;
if (y == NULL)
yv = yvalue;
else
yv = y->value.str;
/*
* TODO: Compare those against QName.
*/
if (ytype == XML_SCHEMAS_QNAME) {
TODO
if (y == NULL)
return(-2);
return (-2);
}
if ((ytype == XML_SCHEMAS_ANYSIMPLETYPE) ||
(ytype == XML_SCHEMAS_STRING) ||
(ytype == XML_SCHEMAS_NORMSTRING) ||
(ytype == XML_SCHEMAS_TOKEN) ||
(ytype == XML_SCHEMAS_LANGUAGE) ||
(ytype == XML_SCHEMAS_NMTOKEN) ||
(ytype == XML_SCHEMAS_NAME) ||
(ytype == XML_SCHEMAS_NCNAME) ||
(ytype == XML_SCHEMAS_ID) ||
(ytype == XML_SCHEMAS_IDREF) ||
(ytype == XML_SCHEMAS_ENTITY) ||
(ytype == XML_SCHEMAS_ANYURI)) {
if (xws == XML_SCHEMA_WHITESPACE_PRESERVE) {
if (yws == XML_SCHEMA_WHITESPACE_PRESERVE) {
/* TODO: What about x < y or x > y. */
if (xmlStrEqual(xv, yv))
return (0);
else
return (2);
} else if (yws == XML_SCHEMA_WHITESPACE_REPLACE)
return (xmlSchemaComparePreserveReplaceStrings(xv, yv, 0));
else if (yws == XML_SCHEMA_WHITESPACE_COLLAPSE)
return (xmlSchemaComparePreserveCollapseStrings(xv, yv, 0));
} else if (xws == XML_SCHEMA_WHITESPACE_REPLACE) {
if (yws == XML_SCHEMA_WHITESPACE_PRESERVE)
return (xmlSchemaComparePreserveReplaceStrings(yv, xv, 1));
if (yws == XML_SCHEMA_WHITESPACE_REPLACE)
return (xmlSchemaCompareReplacedStrings(xv, yv));
if (yws == XML_SCHEMA_WHITESPACE_COLLAPSE)
return (xmlSchemaCompareReplaceCollapseStrings(xv, yv, 0));
} else if (xws == XML_SCHEMA_WHITESPACE_COLLAPSE) {
if (yws == XML_SCHEMA_WHITESPACE_PRESERVE)
return (xmlSchemaComparePreserveCollapseStrings(yv, xv, 1));
if (yws == XML_SCHEMA_WHITESPACE_REPLACE)
return (xmlSchemaCompareReplaceCollapseStrings(yv, xv, 1));
if (yws == XML_SCHEMA_WHITESPACE_COLLAPSE)
return (xmlSchemaCompareNormStrings(xv, yv));
} else
return (-2);
}
return (-2);
}
case XML_SCHEMAS_QNAME:
case XML_SCHEMAS_NOTATION:
if ((x == NULL) || (y == NULL))
return(-2);
if ((ytype == XML_SCHEMAS_QNAME) ||
(ytype == XML_SCHEMAS_NOTATION)) {
if ((xmlStrEqual(x->value.qname.name, y->value.qname.name)) &&
(xmlStrEqual(x->value.qname.uri, y->value.qname.uri)))
return(0);
return(2);
}
return (-2);
case XML_SCHEMAS_FLOAT:
case XML_SCHEMAS_DOUBLE:
if ((x == NULL) || (y == NULL))
return(-2);
if ((ytype == XML_SCHEMAS_FLOAT) ||
(ytype == XML_SCHEMAS_DOUBLE))
return (xmlSchemaCompareFloats(x, y));
return (-2);
case XML_SCHEMAS_BOOLEAN:
if ((x == NULL) || (y == NULL))
return(-2);
if (ytype == XML_SCHEMAS_BOOLEAN) {
if (x->value.b == y->value.b)
return(0);
if (x->value.b == 0)
return(-1);
return(1);
}
return (-2);
case XML_SCHEMAS_HEXBINARY:
if ((x == NULL) || (y == NULL))
return(-2);
if (ytype == XML_SCHEMAS_HEXBINARY) {
if (x->value.hex.total == y->value.hex.total) {
int ret = xmlStrcmp(x->value.hex.str, y->value.hex.str);
if (ret > 0)
return(1);
else if (ret == 0)
return(0);
}
else if (x->value.hex.total > y->value.hex.total)
return(1);
return(-1);
}
return (-2);
case XML_SCHEMAS_BASE64BINARY:
if ((x == NULL) || (y == NULL))
return(-2);
if (ytype == XML_SCHEMAS_BASE64BINARY) {
if (x->value.base64.total == y->value.base64.total) {
int ret = xmlStrcmp(x->value.base64.str,
y->value.base64.str);
if (ret > 0)
return(1);
else if (ret == 0)
return(0);
else
return(-1);
}
else if (x->value.base64.total > y->value.base64.total)
return(1);
else
return(-1);
}
return (-2);
case XML_SCHEMAS_IDREFS:
case XML_SCHEMAS_ENTITIES:
case XML_SCHEMAS_NMTOKENS:
TODO
break;
}
return -2;
}
/**
* xmlSchemaCompareValues:
* @x: a first value
* @y: a second value
*
* Compare 2 values
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, 2 if x <> y, and -2 in
* case of error
*/
int
xmlSchemaCompareValues(xmlSchemaValPtr x, xmlSchemaValPtr y) {
xmlSchemaWhitespaceValueType xws, yws;
if ((x == NULL) || (y == NULL))
return(-2);
if (x->type == XML_SCHEMAS_STRING)
xws = XML_SCHEMA_WHITESPACE_PRESERVE;
else if (x->type == XML_SCHEMAS_NORMSTRING)
xws = XML_SCHEMA_WHITESPACE_REPLACE;
else
xws = XML_SCHEMA_WHITESPACE_COLLAPSE;
if (y->type == XML_SCHEMAS_STRING)
yws = XML_SCHEMA_WHITESPACE_PRESERVE;
else if (y->type == XML_SCHEMAS_NORMSTRING)
yws = XML_SCHEMA_WHITESPACE_REPLACE;
else
yws = XML_SCHEMA_WHITESPACE_COLLAPSE;
return(xmlSchemaCompareValuesInternal(x->type, x, NULL, xws, y->type,
y, NULL, yws));
}
/**
* xmlSchemaCompareValuesWhtsp:
* @x: a first value
* @xws: the whitespace value of x
* @y: a second value
* @yws: the whitespace value of y
*
* Compare 2 values
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, 2 if x <> y, and -2 in
* case of error
*/
int
xmlSchemaCompareValuesWhtsp(xmlSchemaValPtr x,
xmlSchemaWhitespaceValueType xws,
xmlSchemaValPtr y,
xmlSchemaWhitespaceValueType yws)
{
if ((x == NULL) || (y == NULL))
return(-2);
return(xmlSchemaCompareValuesInternal(x->type, x, NULL, xws, y->type,
y, NULL, yws));
}
/**
* xmlSchemaCompareValuesWhtspExt:
* @x: a first value
* @xws: the whitespace value of x
* @y: a second value
* @yws: the whitespace value of y
*
* Compare 2 values
*
* Returns -1 if x < y, 0 if x == y, 1 if x > y, 2 if x <> y, and -2 in
* case of error
*/
static int
xmlSchemaCompareValuesWhtspExt(xmlSchemaValType xtype,
xmlSchemaValPtr x,
const xmlChar *xvalue,
xmlSchemaWhitespaceValueType xws,
xmlSchemaValType ytype,
xmlSchemaValPtr y,
const xmlChar *yvalue,
xmlSchemaWhitespaceValueType yws)
{
return(xmlSchemaCompareValuesInternal(xtype, x, xvalue, xws, ytype, y,
yvalue, yws));
}
/**
* xmlSchemaNormLen:
* @value: a string
*
* Computes the UTF8 length of the normalized value of the string
*
* Returns the length or -1 in case of error.
*/
static int
xmlSchemaNormLen(const xmlChar *value) {
const xmlChar *utf;
int ret = 0;
if (value == NULL)
return(-1);
utf = value;
while (IS_BLANK_CH(*utf)) utf++;
while (*utf != 0) {
if (utf[0] & 0x80) {
if ((utf[1] & 0xc0) != 0x80)
return(-1);
if ((utf[0] & 0xe0) == 0xe0) {
if ((utf[2] & 0xc0) != 0x80)
return(-1);
if ((utf[0] & 0xf0) == 0xf0) {
if ((utf[0] & 0xf8) != 0xf0 || (utf[3] & 0xc0) != 0x80)
return(-1);
utf += 4;
} else {
utf += 3;
}
} else {
utf += 2;
}
} else if (IS_BLANK_CH(*utf)) {
while (IS_BLANK_CH(*utf)) utf++;
if (*utf == 0)
break;
} else {
utf++;
}
ret++;
}
return(ret);
}
/**
* xmlSchemaGetFacetValueAsULong:
* @facet: an schemas type facet
*
* Extract the value of a facet
*
* Returns the value as a long
*/
unsigned long
xmlSchemaGetFacetValueAsULong(xmlSchemaFacetPtr facet)
{
/*
* TODO: Check if this is a decimal.
*/
if (facet == NULL || facet->val == NULL)
return 0;
return ((unsigned long) facet->val->value.decimal.lo);
}
/**
* xmlSchemaValidateListSimpleTypeFacet:
* @facet: the facet to check
* @value: the lexical repr of the value to validate
* @actualLen: the number of list items
* @expectedLen: the resulting expected number of list items
*
* Checks the value of a list simple type against a facet.
*
* Returns 0 if the value is valid, a positive error code
* number otherwise and -1 in case of an internal error.
*/
int
xmlSchemaValidateListSimpleTypeFacet(xmlSchemaFacetPtr facet,
const xmlChar *value,
unsigned long actualLen,
unsigned long *expectedLen)
{
if (facet == NULL)
return(-1);
/*
* TODO: Check if this will work with large numbers.
* (compare value.decimal.mi and value.decimal.hi as well?).
*/
if (facet->type == XML_SCHEMA_FACET_LENGTH) {
if (actualLen != facet->val->value.decimal.lo) {
if (expectedLen != NULL)
*expectedLen = facet->val->value.decimal.lo;
return (XML_SCHEMAV_CVC_LENGTH_VALID);
}
} else if (facet->type == XML_SCHEMA_FACET_MINLENGTH) {
if (actualLen < facet->val->value.decimal.lo) {
if (expectedLen != NULL)
*expectedLen = facet->val->value.decimal.lo;
return (XML_SCHEMAV_CVC_MINLENGTH_VALID);
}
} else if (facet->type == XML_SCHEMA_FACET_MAXLENGTH) {
if (actualLen > facet->val->value.decimal.lo) {
if (expectedLen != NULL)
*expectedLen = facet->val->value.decimal.lo;
return (XML_SCHEMAV_CVC_MAXLENGTH_VALID);
}
} else
/*
* NOTE: That we can pass NULL as xmlSchemaValPtr to
* xmlSchemaValidateFacet, since the remaining facet types
* are: XML_SCHEMA_FACET_PATTERN, XML_SCHEMA_FACET_ENUMERATION.
*/
return(xmlSchemaValidateFacet(NULL, facet, value, NULL));
return (0);
}
/**
* xmlSchemaValidateLengthFacet:
* @type: the built-in type
* @facet: the facet to check
* @value: the lexical repr. of the value to be validated
* @val: the precomputed value
* @ws: the whitespace type of the value
* @length: the actual length of the value
*
* Checka a value against a "length", "minLength" and "maxLength"
* facet; sets @length to the computed length of @value.
*
* Returns 0 if the value is valid, a positive error code
* otherwise and -1 in case of an internal or API error.
*/
static int
xmlSchemaValidateLengthFacetInternal(xmlSchemaFacetPtr facet,
xmlSchemaValType valType,
const xmlChar *value,
xmlSchemaValPtr val,
unsigned long *length,
xmlSchemaWhitespaceValueType ws)
{
unsigned int len = 0;
if ((length == NULL) || (facet == NULL))
return (-1);
*length = 0;
if ((facet->type != XML_SCHEMA_FACET_LENGTH) &&
(facet->type != XML_SCHEMA_FACET_MAXLENGTH) &&
(facet->type != XML_SCHEMA_FACET_MINLENGTH))
return (-1);
/*
* TODO: length, maxLength and minLength must be of type
* nonNegativeInteger only. Check if decimal is used somehow.
*/
if ((facet->val == NULL) ||
((facet->val->type != XML_SCHEMAS_DECIMAL) &&
(facet->val->type != XML_SCHEMAS_NNINTEGER)) ||
(facet->val->value.decimal.frac != 0)) {
return(-1);
}
if ((val != NULL) && (val->type == XML_SCHEMAS_HEXBINARY))
len = val->value.hex.total;
else if ((val != NULL) && (val->type == XML_SCHEMAS_BASE64BINARY))
len = val->value.base64.total;
else {
switch (valType) {
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_NORMSTRING:
if (ws == XML_SCHEMA_WHITESPACE_UNKNOWN) {
/*
* This is to ensure API compatibility with the old
* xmlSchemaValidateLengthFacet(). Anyway, this was and
* is not the correct handling.
* TODO: Get rid of this case somehow.
*/
if (valType == XML_SCHEMAS_STRING)
len = xmlUTF8Strlen(value);
else
len = xmlSchemaNormLen(value);
} else if (value != NULL) {
if (ws == XML_SCHEMA_WHITESPACE_COLLAPSE)
len = xmlSchemaNormLen(value);
else
/*
* Should be OK for "preserve" as well.
*/
len = xmlUTF8Strlen(value);
}
break;
case XML_SCHEMAS_IDREF:
case XML_SCHEMAS_TOKEN:
case XML_SCHEMAS_LANGUAGE:
case XML_SCHEMAS_NMTOKEN:
case XML_SCHEMAS_NAME:
case XML_SCHEMAS_NCNAME:
case XML_SCHEMAS_ID:
/*
* FIXME: What exactly to do with anyURI?
*/
case XML_SCHEMAS_ANYURI:
if (value != NULL)
len = xmlSchemaNormLen(value);
break;
case XML_SCHEMAS_QNAME:
case XML_SCHEMAS_NOTATION:
/*
* For QName and NOTATION, those facets are
* deprecated and should be ignored.
*/
return (0);
default:
TODO
}
}
*length = (unsigned long) len;
/*
* TODO: Return the whole expected value, i.e. "lo", "mi" and "hi".
*/
if (facet->type == XML_SCHEMA_FACET_LENGTH) {
if (len != facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_LENGTH_VALID);
} else if (facet->type == XML_SCHEMA_FACET_MINLENGTH) {
if (len < facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_MINLENGTH_VALID);
} else {
if (len > facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_MAXLENGTH_VALID);
}
return (0);
}
/**
* xmlSchemaValidateLengthFacet:
* @type: the built-in type
* @facet: the facet to check
* @value: the lexical repr. of the value to be validated
* @val: the precomputed value
* @length: the actual length of the value
*
* Checka a value against a "length", "minLength" and "maxLength"
* facet; sets @length to the computed length of @value.
*
* Returns 0 if the value is valid, a positive error code
* otherwise and -1 in case of an internal or API error.
*/
int
xmlSchemaValidateLengthFacet(xmlSchemaTypePtr type,
xmlSchemaFacetPtr facet,
const xmlChar *value,
xmlSchemaValPtr val,
unsigned long *length)
{
if (type == NULL)
return(-1);
return (xmlSchemaValidateLengthFacetInternal(facet,
type->builtInType, value, val, length,
XML_SCHEMA_WHITESPACE_UNKNOWN));
}
/**
* xmlSchemaValidateLengthFacetWhtsp:
* @facet: the facet to check
* @valType: the built-in type
* @value: the lexical repr. of the value to be validated
* @val: the precomputed value
* @ws: the whitespace type of the value
* @length: the actual length of the value
*
* Checka a value against a "length", "minLength" and "maxLength"
* facet; sets @length to the computed length of @value.
*
* Returns 0 if the value is valid, a positive error code
* otherwise and -1 in case of an internal or API error.
*/
int
xmlSchemaValidateLengthFacetWhtsp(xmlSchemaFacetPtr facet,
xmlSchemaValType valType,
const xmlChar *value,
xmlSchemaValPtr val,
unsigned long *length,
xmlSchemaWhitespaceValueType ws)
{
return (xmlSchemaValidateLengthFacetInternal(facet, valType, value, val,
length, ws));
}
/**
* xmlSchemaValidateFacetInternal:
* @facet: the facet to check
* @fws: the whitespace type of the facet's value
* @valType: the built-in type of the value
* @value: the lexical repr of the value to validate
* @val: the precomputed value
* @ws: the whitespace type of the value
*
* Check a value against a facet condition
*
* Returns 0 if the element is schemas valid, a positive error code
* number otherwise and -1 in case of internal or API error.
*/
static int
xmlSchemaValidateFacetInternal(xmlSchemaFacetPtr facet,
xmlSchemaWhitespaceValueType fws,
xmlSchemaValType valType,
const xmlChar *value,
xmlSchemaValPtr val,
xmlSchemaWhitespaceValueType ws)
{
int ret;
if (facet == NULL)
return(-1);
switch (facet->type) {
case XML_SCHEMA_FACET_PATTERN:
/*
* NOTE that for patterns, the @value needs to be the normalized
* value, *not* the lexical initial value or the canonical value.
*/
if (value == NULL)
return(-1);
/*
* If string-derived type, regexp must be tested on the value space of
* the datatype.
* See https://www.w3.org/TR/xmlschema-2/#rf-pattern
*/
if (val &&
val->value.str &&
((val->type >= XML_SCHEMAS_STRING &&
val->type <= XML_SCHEMAS_NORMSTRING) ||
(val->type >= XML_SCHEMAS_TOKEN &&
val->type <= XML_SCHEMAS_ENTITIES &&
val->type != XML_SCHEMAS_QNAME))) {
value = val->value.str;
}
ret = xmlRegexpExec(facet->regexp, value);
if (ret == 1)
return(0);
if (ret == 0)
return(XML_SCHEMAV_CVC_PATTERN_VALID);
return(ret);
case XML_SCHEMA_FACET_MAXEXCLUSIVE:
ret = xmlSchemaCompareValues(val, facet->val);
if (ret == -2)
return(-1);
if (ret == -1)
return(0);
return(XML_SCHEMAV_CVC_MAXEXCLUSIVE_VALID);
case XML_SCHEMA_FACET_MAXINCLUSIVE:
ret = xmlSchemaCompareValues(val, facet->val);
if (ret == -2)
return(-1);
if ((ret == -1) || (ret == 0))
return(0);
return(XML_SCHEMAV_CVC_MAXINCLUSIVE_VALID);
case XML_SCHEMA_FACET_MINEXCLUSIVE:
ret = xmlSchemaCompareValues(val, facet->val);
if (ret == -2)
return(-1);
if (ret == 1)
return(0);
return(XML_SCHEMAV_CVC_MINEXCLUSIVE_VALID);
case XML_SCHEMA_FACET_MININCLUSIVE:
ret = xmlSchemaCompareValues(val, facet->val);
if (ret == -2)
return(-1);
if ((ret == 1) || (ret == 0))
return(0);
return(XML_SCHEMAV_CVC_MININCLUSIVE_VALID);
case XML_SCHEMA_FACET_WHITESPACE:
/* TODO whitespaces */
/*
* NOTE: Whitespace should be handled to normalize
* the value to be validated against a the facets;
* not to normalize the value in-between.
*/
return(0);
case XML_SCHEMA_FACET_ENUMERATION:
if (ws == XML_SCHEMA_WHITESPACE_UNKNOWN) {
/*
* This is to ensure API compatibility with the old
* xmlSchemaValidateFacet().
* TODO: Get rid of this case.
*/
if ((facet->value != NULL) &&
(xmlStrEqual(facet->value, value)))
return(0);
} else {
ret = xmlSchemaCompareValuesWhtspExt(facet->val->type,
facet->val, facet->value, fws, valType, val,
value, ws);
if (ret == -2)
return(-1);
if (ret == 0)
return(0);
}
return(XML_SCHEMAV_CVC_ENUMERATION_VALID);
case XML_SCHEMA_FACET_LENGTH:
/*
* SPEC (1.3) "if {primitive type definition} is QName or NOTATION,
* then any {value} is facet-valid."
*/
if ((valType == XML_SCHEMAS_QNAME) ||
(valType == XML_SCHEMAS_NOTATION))
return (0);
/* Falls through. */
case XML_SCHEMA_FACET_MAXLENGTH:
case XML_SCHEMA_FACET_MINLENGTH: {
unsigned int len = 0;
if ((valType == XML_SCHEMAS_QNAME) ||
(valType == XML_SCHEMAS_NOTATION))
return (0);
/*
* TODO: length, maxLength and minLength must be of type
* nonNegativeInteger only. Check if decimal is used somehow.
*/
if ((facet->val == NULL) ||
((facet->val->type != XML_SCHEMAS_DECIMAL) &&
(facet->val->type != XML_SCHEMAS_NNINTEGER)) ||
(facet->val->value.decimal.frac != 0)) {
return(-1);
}
if ((val != NULL) && (val->type == XML_SCHEMAS_HEXBINARY))
len = val->value.hex.total;
else if ((val != NULL) && (val->type == XML_SCHEMAS_BASE64BINARY))
len = val->value.base64.total;
else {
switch (valType) {
case XML_SCHEMAS_STRING:
case XML_SCHEMAS_NORMSTRING:
if (ws == XML_SCHEMA_WHITESPACE_UNKNOWN) {
/*
* This is to ensure API compatibility with the old
* xmlSchemaValidateFacet(). Anyway, this was and
* is not the correct handling.
* TODO: Get rid of this case somehow.
*/
if (valType == XML_SCHEMAS_STRING)
len = xmlUTF8Strlen(value);
else
len = xmlSchemaNormLen(value);
} else if (value != NULL) {
if (ws == XML_SCHEMA_WHITESPACE_COLLAPSE)
len = xmlSchemaNormLen(value);
else
/*
* Should be OK for "preserve" as well.
*/
len = xmlUTF8Strlen(value);
}
break;
case XML_SCHEMAS_IDREF:
case XML_SCHEMAS_TOKEN:
case XML_SCHEMAS_LANGUAGE:
case XML_SCHEMAS_NMTOKEN:
case XML_SCHEMAS_NAME:
case XML_SCHEMAS_NCNAME:
case XML_SCHEMAS_ID:
case XML_SCHEMAS_ANYURI:
if (value != NULL)
len = xmlSchemaNormLen(value);
break;
default:
TODO
}
}
if (facet->type == XML_SCHEMA_FACET_LENGTH) {
if (len != facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_LENGTH_VALID);
} else if (facet->type == XML_SCHEMA_FACET_MINLENGTH) {
if (len < facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_MINLENGTH_VALID);
} else {
if (len > facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_MAXLENGTH_VALID);
}
break;
}
case XML_SCHEMA_FACET_TOTALDIGITS:
case XML_SCHEMA_FACET_FRACTIONDIGITS:
if ((facet->val == NULL) ||
((facet->val->type != XML_SCHEMAS_PINTEGER) &&
(facet->val->type != XML_SCHEMAS_NNINTEGER)) ||
(facet->val->value.decimal.frac != 0)) {
return(-1);
}
if ((val == NULL) ||
((val->type != XML_SCHEMAS_DECIMAL) &&
(val->type != XML_SCHEMAS_INTEGER) &&
(val->type != XML_SCHEMAS_NPINTEGER) &&
(val->type != XML_SCHEMAS_NINTEGER) &&
(val->type != XML_SCHEMAS_NNINTEGER) &&
(val->type != XML_SCHEMAS_PINTEGER) &&
(val->type != XML_SCHEMAS_INT) &&
(val->type != XML_SCHEMAS_UINT) &&
(val->type != XML_SCHEMAS_LONG) &&
(val->type != XML_SCHEMAS_ULONG) &&
(val->type != XML_SCHEMAS_SHORT) &&
(val->type != XML_SCHEMAS_USHORT) &&
(val->type != XML_SCHEMAS_BYTE) &&
(val->type != XML_SCHEMAS_UBYTE))) {
return(-1);
}
if (facet->type == XML_SCHEMA_FACET_TOTALDIGITS) {
if (val->value.decimal.total > facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_TOTALDIGITS_VALID);
} else if (facet->type == XML_SCHEMA_FACET_FRACTIONDIGITS) {
if (val->value.decimal.frac > facet->val->value.decimal.lo)
return(XML_SCHEMAV_CVC_FRACTIONDIGITS_VALID);
}
break;
default:
TODO
}
return(0);
}
/**
* xmlSchemaValidateFacet:
* @base: the base type
* @facet: the facet to check
* @value: the lexical repr of the value to validate
* @val: the precomputed value
*
* Check a value against a facet condition
*
* Returns 0 if the element is schemas valid, a positive error code
* number otherwise and -1 in case of internal or API error.
*/
int
xmlSchemaValidateFacet(xmlSchemaTypePtr base,
xmlSchemaFacetPtr facet,
const xmlChar *value,
xmlSchemaValPtr val)
{
/*
* This tries to ensure API compatibility regarding the old
* xmlSchemaValidateFacet() and the new xmlSchemaValidateFacetInternal() and
* xmlSchemaValidateFacetWhtsp().
*/
if (val != NULL)
return(xmlSchemaValidateFacetInternal(facet,
XML_SCHEMA_WHITESPACE_UNKNOWN, val->type, value, val,
XML_SCHEMA_WHITESPACE_UNKNOWN));
else if (base != NULL)
return(xmlSchemaValidateFacetInternal(facet,
XML_SCHEMA_WHITESPACE_UNKNOWN, base->builtInType, value, val,
XML_SCHEMA_WHITESPACE_UNKNOWN));
return(-1);
}
/**
* xmlSchemaValidateFacetWhtsp:
* @facet: the facet to check
* @fws: the whitespace type of the facet's value
* @valType: the built-in type of the value
* @value: the lexical (or normalized for pattern) repr of the value to validate
* @val: the precomputed value
* @ws: the whitespace type of the value
*
* Check a value against a facet condition. This takes value normalization
* according to the specified whitespace types into account.
* Note that @value needs to be the *normalized* value if the facet
* is of type "pattern".
*
* Returns 0 if the element is schemas valid, a positive error code
* number otherwise and -1 in case of internal or API error.
*/
int
xmlSchemaValidateFacetWhtsp(xmlSchemaFacetPtr facet,
xmlSchemaWhitespaceValueType fws,
xmlSchemaValType valType,
const xmlChar *value,
xmlSchemaValPtr val,
xmlSchemaWhitespaceValueType ws)
{
return(xmlSchemaValidateFacetInternal(facet, fws, valType,
value, val, ws));
}
#if 0
#ifndef DBL_DIG
#define DBL_DIG 16
#endif
#ifndef DBL_EPSILON
#define DBL_EPSILON 1E-9
#endif
#define INTEGER_DIGITS DBL_DIG
#define FRACTION_DIGITS (DBL_DIG + 1)
#define EXPONENT_DIGITS (3 + 2)
/**
* xmlXPathFormatNumber:
* @number: number to format
* @buffer: output buffer
* @buffersize: size of output buffer
*
* Convert the number into a string representation.
*/
static void
xmlSchemaFormatFloat(double number, char buffer[], int buffersize)
{
switch (xmlXPathIsInf(number)) {
case 1:
if (buffersize > (int)sizeof("INF"))
snprintf(buffer, buffersize, "INF");
break;
case -1:
if (buffersize > (int)sizeof("-INF"))
snprintf(buffer, buffersize, "-INF");
break;
default:
if (xmlXPathIsNaN(number)) {
if (buffersize > (int)sizeof("NaN"))
snprintf(buffer, buffersize, "NaN");
} else if (number == 0) {
snprintf(buffer, buffersize, "0.0E0");
} else {
/* 3 is sign, decimal point, and terminating zero */
char work[DBL_DIG + EXPONENT_DIGITS + 3];
int integer_place, fraction_place;
char *ptr;
char *after_fraction;
double absolute_value;
int size;
absolute_value = fabs(number);
/*
* Result is in work, and after_fraction points
* just past the fractional part.
* Use scientific notation
*/
integer_place = DBL_DIG + EXPONENT_DIGITS + 1;
fraction_place = DBL_DIG - 1;
snprintf(work, sizeof(work),"%*.*e",
integer_place, fraction_place, number);
after_fraction = strchr(work + DBL_DIG, 'e');
/* Remove fractional trailing zeroes */
ptr = after_fraction;
while (*(--ptr) == '0')
;
if (*ptr != '.')
ptr++;
while ((*ptr++ = *after_fraction++) != 0);
/* Finally copy result back to caller */
size = strlen(work) + 1;
if (size > buffersize) {
work[buffersize - 1] = 0;
size = buffersize;
}
memmove(buffer, work, size);
}
break;
}
}
#endif
/**
* xmlSchemaGetCanonValue:
* @val: the precomputed value
* @retValue: the returned value
*
* Get the canonical lexical representation of the value.
* The caller has to FREE the returned retValue.
*
* WARNING: Some value types are not supported yet, resulting
* in a @retValue of "???".
*
* TODO: XML Schema 1.0 does not define canonical representations
* for: duration, gYearMonth, gYear, gMonthDay, gMonth, gDay,
* anyURI, QName, NOTATION. This will be fixed in XML Schema 1.1.
*
*
* Returns 0 if the value could be built, 1 if the value type is
* not supported yet and -1 in case of API errors.
*/
int
xmlSchemaGetCanonValue(xmlSchemaValPtr val, const xmlChar **retValue)
{
if ((retValue == NULL) || (val == NULL))
return (-1);
*retValue = NULL;
switch (val->type) {
case XML_SCHEMAS_STRING:
if (val->value.str == NULL)
*retValue = BAD_CAST xmlStrdup(BAD_CAST "");
else
*retValue =
BAD_CAST xmlStrdup((const xmlChar *) val->value.str);
break;
case XML_SCHEMAS_NORMSTRING:
if (val->value.str == NULL)
*retValue = BAD_CAST xmlStrdup(BAD_CAST "");
else {
*retValue = xmlSchemaWhiteSpaceReplace(
(const xmlChar *) val->value.str);
if ((*retValue) == NULL)
*retValue = BAD_CAST xmlStrdup(
(const xmlChar *) val->value.str);
}
break;
case XML_SCHEMAS_TOKEN:
case XML_SCHEMAS_LANGUAGE:
case XML_SCHEMAS_NMTOKEN:
case XML_SCHEMAS_NAME:
case XML_SCHEMAS_NCNAME:
case XML_SCHEMAS_ID:
case XML_SCHEMAS_IDREF:
case XML_SCHEMAS_ENTITY:
case XML_SCHEMAS_NOTATION: /* Unclear */
case XML_SCHEMAS_ANYURI: /* Unclear */
if (val->value.str == NULL)
return (-1);
*retValue =
BAD_CAST xmlSchemaCollapseString(BAD_CAST val->value.str);
if (*retValue == NULL)
*retValue =
BAD_CAST xmlStrdup((const xmlChar *) val->value.str);
break;
case XML_SCHEMAS_QNAME:
/* TODO: Unclear in XML Schema 1.0. */
if (val->value.qname.uri == NULL) {
*retValue = BAD_CAST xmlStrdup(BAD_CAST val->value.qname.name);
return (0);
} else {
*retValue = BAD_CAST xmlStrdup(BAD_CAST "{");
*retValue = BAD_CAST xmlStrcat((xmlChar *) (*retValue),
BAD_CAST val->value.qname.uri);
*retValue = BAD_CAST xmlStrcat((xmlChar *) (*retValue),
BAD_CAST "}");
*retValue = BAD_CAST xmlStrcat((xmlChar *) (*retValue),
BAD_CAST val->value.qname.uri);
}
break;
case XML_SCHEMAS_DECIMAL:
/*
* TODO: Lookout for a more simple implementation.
*/
if ((val->value.decimal.total == 1) &&
(val->value.decimal.lo == 0)) {
*retValue = xmlStrdup(BAD_CAST "0.0");
} else {
xmlSchemaValDecimal dec = val->value.decimal;
int bufsize;
char *buf = NULL, *offs;
/* Add room for the decimal point as well. */
bufsize = dec.total + 2;
if (dec.sign)
bufsize++;
/* Add room for leading/trailing zero. */
if ((dec.frac == 0) || (dec.frac == dec.total))
bufsize++;
buf = xmlMalloc(bufsize);
if (buf == NULL)
return(-1);
offs = buf;
if (dec.sign)
*offs++ = '-';
if (dec.frac == dec.total) {
*offs++ = '0';
*offs++ = '.';
}
if (dec.hi != 0)
snprintf(offs, bufsize - (offs - buf),
"%lu%lu%lu", dec.hi, dec.mi, dec.lo);
else if (dec.mi != 0)
snprintf(offs, bufsize - (offs - buf),
"%lu%lu", dec.mi, dec.lo);
else
snprintf(offs, bufsize - (offs - buf),
"%lu", dec.lo);
if (dec.frac != 0) {
if (dec.frac != dec.total) {
int diff = dec.total - dec.frac;
/*
* Insert the decimal point.
*/
memmove(offs + diff + 1, offs + diff, dec.frac +1);
offs[diff] = '.';
} else {
unsigned int i = 0;
/*
* Insert missing zeroes behind the decimal point.
*/
while (*(offs + i) != 0)
i++;
if (i < dec.total) {
memmove(offs + (dec.total - i), offs, i +1);
memset(offs, '0', dec.total - i);
}
}
} else {
/*
* Append decimal point and zero.
*/
offs = buf + bufsize - 1;
*offs-- = 0;
*offs-- = '0';
*offs-- = '.';
}
*retValue = BAD_CAST buf;
}
break;
case XML_SCHEMAS_INTEGER:
case XML_SCHEMAS_PINTEGER:
case XML_SCHEMAS_NPINTEGER:
case XML_SCHEMAS_NINTEGER:
case XML_SCHEMAS_NNINTEGER:
case XML_SCHEMAS_LONG:
case XML_SCHEMAS_BYTE:
case XML_SCHEMAS_SHORT:
case XML_SCHEMAS_INT:
case XML_SCHEMAS_UINT:
case XML_SCHEMAS_ULONG:
case XML_SCHEMAS_USHORT:
case XML_SCHEMAS_UBYTE:
if ((val->value.decimal.total == 1) &&
(val->value.decimal.lo == 0))
*retValue = xmlStrdup(BAD_CAST "0");
else {
xmlSchemaValDecimal dec = val->value.decimal;
int bufsize = dec.total + 1;
/* Add room for the decimal point as well. */
if (dec.sign)
bufsize++;
*retValue = xmlMalloc(bufsize);
if (*retValue == NULL)
return(-1);
if (dec.hi != 0) {
if (dec.sign)
snprintf((char *) *retValue, bufsize,
"-%lu%lu%lu", dec.hi, dec.mi, dec.lo);
else
snprintf((char *) *retValue, bufsize,
"%lu%lu%lu", dec.hi, dec.mi, dec.lo);
} else if (dec.mi != 0) {
if (dec.sign)
snprintf((char *) *retValue, bufsize,
"-%lu%lu", dec.mi, dec.lo);
else
snprintf((char *) *retValue, bufsize,
"%lu%lu", dec.mi, dec.lo);
} else {
if (dec.sign)
snprintf((char *) *retValue, bufsize, "-%lu", dec.lo);
else
snprintf((char *) *retValue, bufsize, "%lu", dec.lo);
}
}
break;
case XML_SCHEMAS_BOOLEAN:
if (val->value.b)
*retValue = BAD_CAST xmlStrdup(BAD_CAST "true");
else
*retValue = BAD_CAST xmlStrdup(BAD_CAST "false");
break;
case XML_SCHEMAS_DURATION: {
char buf[100];
unsigned long year;
unsigned long mon, day, hour = 0, min = 0;
double sec = 0, left;
/* TODO: Unclear in XML Schema 1.0 */
/*
* TODO: This results in a normalized output of the value
* - which is NOT conformant to the spec -
* since the exact values of each property are not
* recoverable. Think about extending the structure to
* provide a field for every property.
*/
year = (unsigned long) FQUOTIENT(labs(val->value.dur.mon), 12);
mon = labs(val->value.dur.mon) - 12 * year;
day = (unsigned long) FQUOTIENT(fabs(val->value.dur.sec), 86400);
left = fabs(val->value.dur.sec) - day * 86400;
if (left > 0) {
hour = (unsigned long) FQUOTIENT(left, 3600);
left = left - (hour * 3600);
if (left > 0) {
min = (unsigned long) FQUOTIENT(left, 60);
sec = left - (min * 60);
}
}
if ((val->value.dur.mon < 0) || (val->value.dur.sec < 0))
snprintf(buf, 100, "P%luY%luM%luDT%luH%luM%.14gS",
year, mon, day, hour, min, sec);
else
snprintf(buf, 100, "-P%luY%luM%luDT%luH%luM%.14gS",
year, mon, day, hour, min, sec);
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
case XML_SCHEMAS_GYEAR: {
char buf[30];
/* TODO: Unclear in XML Schema 1.0 */
/* TODO: What to do with the timezone? */
snprintf(buf, 30, "%04ld", val->value.date.year);
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
case XML_SCHEMAS_GMONTH: {
/* TODO: Unclear in XML Schema 1.0 */
/* TODO: What to do with the timezone? */
*retValue = xmlMalloc(6);
if (*retValue == NULL)
return(-1);
snprintf((char *) *retValue, 6, "--%02u",
val->value.date.mon);
}
break;
case XML_SCHEMAS_GDAY: {
/* TODO: Unclear in XML Schema 1.0 */
/* TODO: What to do with the timezone? */
*retValue = xmlMalloc(6);
if (*retValue == NULL)
return(-1);
snprintf((char *) *retValue, 6, "---%02u",
val->value.date.day);
}
break;
case XML_SCHEMAS_GMONTHDAY: {
/* TODO: Unclear in XML Schema 1.0 */
/* TODO: What to do with the timezone? */
*retValue = xmlMalloc(8);
if (*retValue == NULL)
return(-1);
snprintf((char *) *retValue, 8, "--%02u-%02u",
val->value.date.mon, val->value.date.day);
}
break;
case XML_SCHEMAS_GYEARMONTH: {
char buf[35];
/* TODO: Unclear in XML Schema 1.0 */
/* TODO: What to do with the timezone? */
if (val->value.date.year < 0)
snprintf(buf, 35, "-%04ld-%02u",
labs(val->value.date.year),
val->value.date.mon);
else
snprintf(buf, 35, "%04ld-%02u",
val->value.date.year, val->value.date.mon);
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
case XML_SCHEMAS_TIME:
{
char buf[30];
if (val->value.date.tz_flag) {
xmlSchemaValPtr norm;
norm = xmlSchemaDateNormalize(val, 0);
if (norm == NULL)
return (-1);
/*
* TODO: Check if "%.14g" is portable.
*/
snprintf(buf, 30,
"%02u:%02u:%02.14gZ",
norm->value.date.hour,
norm->value.date.min,
norm->value.date.sec);
xmlSchemaFreeValue(norm);
} else {
snprintf(buf, 30,
"%02u:%02u:%02.14g",
val->value.date.hour,
val->value.date.min,
val->value.date.sec);
}
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
case XML_SCHEMAS_DATE:
{
char buf[30];
if (val->value.date.tz_flag) {
xmlSchemaValPtr norm;
norm = xmlSchemaDateNormalize(val, 0);
if (norm == NULL)
return (-1);
/*
* TODO: Append the canonical value of the
* recoverable timezone and not "Z".
*/
snprintf(buf, 30,
"%04ld-%02u-%02uZ",
norm->value.date.year, norm->value.date.mon,
norm->value.date.day);
xmlSchemaFreeValue(norm);
} else {
snprintf(buf, 30,
"%04ld-%02u-%02u",
val->value.date.year, val->value.date.mon,
val->value.date.day);
}
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
case XML_SCHEMAS_DATETIME:
{
char buf[50];
if (val->value.date.tz_flag) {
xmlSchemaValPtr norm;
norm = xmlSchemaDateNormalize(val, 0);
if (norm == NULL)
return (-1);
/*
* TODO: Check if "%.14g" is portable.
*/
snprintf(buf, 50,
"%04ld-%02u-%02uT%02u:%02u:%02.14gZ",
norm->value.date.year, norm->value.date.mon,
norm->value.date.day, norm->value.date.hour,
norm->value.date.min, norm->value.date.sec);
xmlSchemaFreeValue(norm);
} else {
snprintf(buf, 50,
"%04ld-%02u-%02uT%02u:%02u:%02.14g",
val->value.date.year, val->value.date.mon,
val->value.date.day, val->value.date.hour,
val->value.date.min, val->value.date.sec);
}
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
case XML_SCHEMAS_HEXBINARY:
*retValue = BAD_CAST xmlStrdup(BAD_CAST val->value.hex.str);
break;
case XML_SCHEMAS_BASE64BINARY:
/*
* TODO: Is the following spec piece implemented?:
* SPEC: "Note: For some values the canonical form defined
* above does not conform to [RFC 2045], which requires breaking
* with linefeeds at appropriate intervals."
*/
*retValue = BAD_CAST xmlStrdup(BAD_CAST val->value.base64.str);
break;
case XML_SCHEMAS_FLOAT: {
char buf[30];
/*
* |m| < 16777216, -149 <= e <= 104.
* TODO: Handle, NaN, INF, -INF. The format is not
* yet conformant. The c type float does not cover
* the whole range.
*/
snprintf(buf, 30, "%01.14e", val->value.f);
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
case XML_SCHEMAS_DOUBLE: {
char buf[40];
/* |m| < 9007199254740992, -1075 <= e <= 970 */
/*
* TODO: Handle, NaN, INF, -INF. The format is not
* yet conformant. The c type float does not cover
* the whole range.
*/
snprintf(buf, 40, "%01.14e", val->value.d);
*retValue = BAD_CAST xmlStrdup(BAD_CAST buf);
}
break;
default:
*retValue = BAD_CAST xmlStrdup(BAD_CAST "???");
return (1);
}
if (*retValue == NULL)
return(-1);
return (0);
}
/**
* xmlSchemaGetCanonValueWhtsp:
* @val: the precomputed value
* @retValue: the returned value
* @ws: the whitespace type of the value
*
* Get the canonical representation of the value.
* The caller has to free the returned @retValue.
*
* Returns 0 if the value could be built, 1 if the value type is
* not supported yet and -1 in case of API errors.
*/
int
xmlSchemaGetCanonValueWhtsp(xmlSchemaValPtr val,
const xmlChar **retValue,
xmlSchemaWhitespaceValueType ws)
{
if ((retValue == NULL) || (val == NULL))
return (-1);
if ((ws == XML_SCHEMA_WHITESPACE_UNKNOWN) ||
(ws > XML_SCHEMA_WHITESPACE_COLLAPSE))
return (-1);
*retValue = NULL;
switch (val->type) {
case XML_SCHEMAS_STRING:
if (val->value.str == NULL)
*retValue = BAD_CAST xmlStrdup(BAD_CAST "");
else if (ws == XML_SCHEMA_WHITESPACE_COLLAPSE)
*retValue = xmlSchemaCollapseString(val->value.str);
else if (ws == XML_SCHEMA_WHITESPACE_REPLACE)
*retValue = xmlSchemaWhiteSpaceReplace(val->value.str);
if ((*retValue) == NULL)
*retValue = BAD_CAST xmlStrdup(val->value.str);
break;
case XML_SCHEMAS_NORMSTRING:
if (val->value.str == NULL)
*retValue = BAD_CAST xmlStrdup(BAD_CAST "");
else {
if (ws == XML_SCHEMA_WHITESPACE_COLLAPSE)
*retValue = xmlSchemaCollapseString(val->value.str);
else
*retValue = xmlSchemaWhiteSpaceReplace(val->value.str);
if ((*retValue) == NULL)
*retValue = BAD_CAST xmlStrdup(val->value.str);
}
break;
default:
return (xmlSchemaGetCanonValue(val, retValue));
}
return (0);
}
/**
* xmlSchemaGetValType:
* @val: a schemas value
*
* Accessor for the type of a value
*
* Returns the xmlSchemaValType of the value
*/
xmlSchemaValType
xmlSchemaGetValType(xmlSchemaValPtr val)
{
if (val == NULL)
return(XML_SCHEMAS_UNKNOWN);
return (val->type);
}
#endif /* LIBXML_SCHEMAS_ENABLED */