mirror of
https://github.com/reactos/reactos.git
synced 2024-11-02 12:53:33 +00:00
1347 lines
37 KiB
C
1347 lines
37 KiB
C
/*
|
|
* numbers.c: Implementation of the XSLT number functions
|
|
*
|
|
* Reference:
|
|
* http://www.w3.org/TR/1999/REC-xslt-19991116
|
|
*
|
|
* See Copyright for the status of this software.
|
|
*
|
|
* daniel@veillard.com
|
|
* Bjorn Reese <breese@users.sourceforge.net>
|
|
*/
|
|
|
|
#include "precomp.h"
|
|
|
|
#ifndef FALSE
|
|
# define FALSE (0 == 1)
|
|
# define TRUE (1 == 1)
|
|
#endif
|
|
|
|
#define SYMBOL_QUOTE ((xmlChar)'\'')
|
|
|
|
#define DEFAULT_TOKEN '0'
|
|
#define DEFAULT_SEPARATOR "."
|
|
|
|
#define MAX_TOKENS 1024
|
|
|
|
typedef struct _xsltFormatToken xsltFormatToken;
|
|
typedef xsltFormatToken *xsltFormatTokenPtr;
|
|
struct _xsltFormatToken {
|
|
xmlChar *separator;
|
|
int token;
|
|
int width;
|
|
};
|
|
|
|
typedef struct _xsltFormat xsltFormat;
|
|
typedef xsltFormat *xsltFormatPtr;
|
|
struct _xsltFormat {
|
|
xmlChar *start;
|
|
xsltFormatToken tokens[MAX_TOKENS];
|
|
int nTokens;
|
|
xmlChar *end;
|
|
};
|
|
|
|
static char alpha_upper_list[] = "ABCDEFGHIJKLMNOPQRSTUVWXYZ";
|
|
static char alpha_lower_list[] = "abcdefghijklmnopqrstuvwxyz";
|
|
static xsltFormatToken default_token;
|
|
|
|
/*
|
|
* **** Start temp insert ****
|
|
*
|
|
* The following routine xsltUTF8Charcmp will be replaced with calls to
|
|
* the corresponding libxml routine at a later date (when other
|
|
* inter-library dependencies require it).
|
|
*/
|
|
|
|
/**
|
|
* xsltUTF8Charcmp
|
|
* @utf1: pointer to first UTF8 char
|
|
* @utf2: pointer to second UTF8 char
|
|
*
|
|
* returns result of comparing the two UCS4 values
|
|
* as with xmlStrncmp
|
|
*/
|
|
static int
|
|
xsltUTF8Charcmp(xmlChar *utf1, xmlChar *utf2) {
|
|
int len = xmlUTF8Strsize(utf1, 1);
|
|
|
|
if (len < 1)
|
|
return -1;
|
|
if (utf1 == NULL ) {
|
|
if (utf2 == NULL)
|
|
return 0;
|
|
return -1;
|
|
}
|
|
return xmlStrncmp(utf1, utf2, len);
|
|
}
|
|
|
|
/***** Stop temp insert *****/
|
|
/************************************************************************
|
|
* *
|
|
* Utility functions *
|
|
* *
|
|
************************************************************************/
|
|
|
|
#define IS_SPECIAL(self,letter) \
|
|
((xsltUTF8Charcmp((letter), (self)->zeroDigit) == 0) || \
|
|
(xsltUTF8Charcmp((letter), (self)->digit) == 0) || \
|
|
(xsltUTF8Charcmp((letter), (self)->decimalPoint) == 0) || \
|
|
(xsltUTF8Charcmp((letter), (self)->grouping) == 0) || \
|
|
(xsltUTF8Charcmp((letter), (self)->patternSeparator) == 0))
|
|
|
|
#define IS_DIGIT_ZERO(x) xsltIsDigitZero(x)
|
|
#define IS_DIGIT_ONE(x) xsltIsDigitZero((x)-1)
|
|
|
|
static int
|
|
xsltIsDigitZero(unsigned int ch)
|
|
{
|
|
/*
|
|
* Reference: ftp://ftp.unicode.org/Public/UNIDATA/UnicodeData.txt
|
|
*
|
|
* There a many more digit ranges in newer Unicode versions. These
|
|
* are only the zeros that match Digit in XML 1.0 (IS_DIGIT macro).
|
|
*/
|
|
switch (ch) {
|
|
case 0x0030: case 0x0660: case 0x06F0: case 0x0966:
|
|
case 0x09E6: case 0x0A66: case 0x0AE6: case 0x0B66:
|
|
case 0x0C66: case 0x0CE6: case 0x0D66: case 0x0E50:
|
|
case 0x0ED0: case 0x0F20:
|
|
return TRUE;
|
|
default:
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
static void
|
|
xsltNumberFormatDecimal(xmlBufferPtr buffer,
|
|
double number,
|
|
int digit_zero,
|
|
int width,
|
|
int digitsPerGroup,
|
|
int groupingCharacter,
|
|
int groupingCharacterLen)
|
|
{
|
|
/*
|
|
* This used to be
|
|
* xmlChar temp_string[sizeof(double) * CHAR_BIT * sizeof(xmlChar) + 4];
|
|
* which would be length 68 on x86 arch. It was changed to be a longer,
|
|
* fixed length in order to try to cater for (reasonable) UTF8
|
|
* separators and numeric characters. The max UTF8 char size will be
|
|
* 6 or less, so the value used [500] should be *much* larger than needed
|
|
*/
|
|
xmlChar temp_string[500];
|
|
xmlChar *pointer;
|
|
xmlChar temp_char[6];
|
|
int i;
|
|
int val;
|
|
int len;
|
|
|
|
/* Build buffer from back */
|
|
pointer = &temp_string[sizeof(temp_string)] - 1; /* last char */
|
|
*pointer = 0;
|
|
i = 0;
|
|
while (pointer > temp_string) {
|
|
if ((i >= width) && (fabs(number) < 1.0))
|
|
break; /* for */
|
|
if ((i > 0) && (groupingCharacter != 0) &&
|
|
(digitsPerGroup > 0) &&
|
|
((i % digitsPerGroup) == 0)) {
|
|
if (pointer - groupingCharacterLen < temp_string) {
|
|
i = -1; /* flag error */
|
|
break;
|
|
}
|
|
pointer -= groupingCharacterLen;
|
|
xmlCopyCharMultiByte(pointer, groupingCharacter);
|
|
}
|
|
|
|
val = digit_zero + (int)fmod(number, 10.0);
|
|
if (val < 0x80) { /* shortcut if ASCII */
|
|
if (pointer <= temp_string) { /* Check enough room */
|
|
i = -1;
|
|
break;
|
|
}
|
|
*(--pointer) = val;
|
|
}
|
|
else {
|
|
/*
|
|
* Here we have a multibyte character. It's a little messy,
|
|
* because until we generate the char we don't know how long
|
|
* it is. So, we generate it into the buffer temp_char, then
|
|
* copy from there into temp_string.
|
|
*/
|
|
len = xmlCopyCharMultiByte(temp_char, val);
|
|
if ( (pointer - len) < temp_string ) {
|
|
i = -1;
|
|
break;
|
|
}
|
|
pointer -= len;
|
|
memcpy(pointer, temp_char, len);
|
|
}
|
|
number /= 10.0;
|
|
++i;
|
|
}
|
|
if (i < 0)
|
|
xsltGenericError(xsltGenericErrorContext,
|
|
"xsltNumberFormatDecimal: Internal buffer size exceeded\n");
|
|
xmlBufferCat(buffer, pointer);
|
|
}
|
|
|
|
static void
|
|
xsltNumberFormatAlpha(xsltNumberDataPtr data,
|
|
xmlBufferPtr buffer,
|
|
double number,
|
|
int is_upper)
|
|
{
|
|
char temp_string[sizeof(double) * CHAR_BIT * sizeof(xmlChar) + 1];
|
|
char *pointer;
|
|
int i;
|
|
char *alpha_list;
|
|
double alpha_size = (double)(sizeof(alpha_upper_list) - 1);
|
|
|
|
/*
|
|
* XSLT 1.0 isn't clear on how to handle zero, but XSLT 2.0 says:
|
|
*
|
|
* For all format tokens other than the first kind above (one that
|
|
* consists of decimal digits), there may be implementation-defined
|
|
* lower and upper bounds on the range of numbers that can be
|
|
* formatted using this format token; indeed, for some numbering
|
|
* sequences there may be intrinsic limits. [...] Numbers that fall
|
|
* outside this range must be formatted using the format token 1.
|
|
*
|
|
* The "a" token has an intrinsic lower limit of 1.
|
|
*/
|
|
if (number < 1.0) {
|
|
xsltNumberFormatDecimal(buffer, number, '0', 1,
|
|
data->digitsPerGroup,
|
|
data->groupingCharacter,
|
|
data->groupingCharacterLen);
|
|
return;
|
|
}
|
|
|
|
/* Build buffer from back */
|
|
pointer = &temp_string[sizeof(temp_string)];
|
|
*(--pointer) = 0;
|
|
alpha_list = (is_upper) ? alpha_upper_list : alpha_lower_list;
|
|
|
|
for (i = 1; i < (int)sizeof(temp_string); i++) {
|
|
number--;
|
|
*(--pointer) = alpha_list[((int)fmod(number, alpha_size))];
|
|
number /= alpha_size;
|
|
if (number < 1.0)
|
|
break; /* for */
|
|
}
|
|
xmlBufferCCat(buffer, pointer);
|
|
}
|
|
|
|
static void
|
|
xsltNumberFormatRoman(xsltNumberDataPtr data,
|
|
xmlBufferPtr buffer,
|
|
double number,
|
|
int is_upper)
|
|
{
|
|
/*
|
|
* See discussion in xsltNumberFormatAlpha. Also use a reasonable upper
|
|
* bound to avoid denial of service.
|
|
*/
|
|
if (number < 1.0 || number > 5000.0) {
|
|
xsltNumberFormatDecimal(buffer, number, '0', 1,
|
|
data->digitsPerGroup,
|
|
data->groupingCharacter,
|
|
data->groupingCharacterLen);
|
|
return;
|
|
}
|
|
|
|
/*
|
|
* Based on an example by Jim Walsh
|
|
*/
|
|
while (number >= 1000.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "M" : "m");
|
|
number -= 1000.0;
|
|
}
|
|
if (number >= 900.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "CM" : "cm");
|
|
number -= 900.0;
|
|
}
|
|
while (number >= 500.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "D" : "d");
|
|
number -= 500.0;
|
|
}
|
|
if (number >= 400.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "CD" : "cd");
|
|
number -= 400.0;
|
|
}
|
|
while (number >= 100.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "C" : "c");
|
|
number -= 100.0;
|
|
}
|
|
if (number >= 90.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "XC" : "xc");
|
|
number -= 90.0;
|
|
}
|
|
while (number >= 50.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "L" : "l");
|
|
number -= 50.0;
|
|
}
|
|
if (number >= 40.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "XL" : "xl");
|
|
number -= 40.0;
|
|
}
|
|
while (number >= 10.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "X" : "x");
|
|
number -= 10.0;
|
|
}
|
|
if (number >= 9.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "IX" : "ix");
|
|
number -= 9.0;
|
|
}
|
|
while (number >= 5.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "V" : "v");
|
|
number -= 5.0;
|
|
}
|
|
if (number >= 4.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "IV" : "iv");
|
|
number -= 4.0;
|
|
}
|
|
while (number >= 1.0) {
|
|
xmlBufferCCat(buffer, (is_upper) ? "I" : "i");
|
|
number--;
|
|
}
|
|
}
|
|
|
|
static void
|
|
xsltNumberFormatTokenize(const xmlChar *format,
|
|
xsltFormatPtr tokens)
|
|
{
|
|
int ix = 0;
|
|
int j;
|
|
int val;
|
|
int len;
|
|
|
|
default_token.token = DEFAULT_TOKEN;
|
|
default_token.width = 1;
|
|
default_token.separator = BAD_CAST(DEFAULT_SEPARATOR);
|
|
|
|
|
|
tokens->start = NULL;
|
|
tokens->tokens[0].separator = NULL;
|
|
tokens->end = NULL;
|
|
|
|
/*
|
|
* Insert initial non-alphanumeric token.
|
|
* There is always such a token in the list, even if NULL
|
|
*/
|
|
while (! (IS_LETTER(val=xmlStringCurrentChar(NULL, format+ix, &len)) ||
|
|
IS_DIGIT(val)) ) {
|
|
if (format[ix] == 0) /* if end of format string */
|
|
break; /* while */
|
|
ix += len;
|
|
}
|
|
if (ix > 0)
|
|
tokens->start = xmlStrndup(format, ix);
|
|
|
|
|
|
for (tokens->nTokens = 0; tokens->nTokens < MAX_TOKENS;
|
|
tokens->nTokens++) {
|
|
if (format[ix] == 0)
|
|
break; /* for */
|
|
|
|
/*
|
|
* separator has already been parsed (except for the first
|
|
* number) in tokens->end, recover it.
|
|
*/
|
|
if (tokens->nTokens > 0) {
|
|
tokens->tokens[tokens->nTokens].separator = tokens->end;
|
|
tokens->end = NULL;
|
|
}
|
|
|
|
val = xmlStringCurrentChar(NULL, format+ix, &len);
|
|
if (IS_DIGIT_ONE(val) ||
|
|
IS_DIGIT_ZERO(val)) {
|
|
tokens->tokens[tokens->nTokens].width = 1;
|
|
while (IS_DIGIT_ZERO(val)) {
|
|
tokens->tokens[tokens->nTokens].width++;
|
|
ix += len;
|
|
val = xmlStringCurrentChar(NULL, format+ix, &len);
|
|
}
|
|
if (IS_DIGIT_ONE(val)) {
|
|
tokens->tokens[tokens->nTokens].token = val - 1;
|
|
ix += len;
|
|
val = xmlStringCurrentChar(NULL, format+ix, &len);
|
|
} else {
|
|
tokens->tokens[tokens->nTokens].token = '0';
|
|
tokens->tokens[tokens->nTokens].width = 1;
|
|
}
|
|
} else if ( (val == 'A') ||
|
|
(val == 'a') ||
|
|
(val == 'I') ||
|
|
(val == 'i') ) {
|
|
tokens->tokens[tokens->nTokens].token = val;
|
|
ix += len;
|
|
val = xmlStringCurrentChar(NULL, format+ix, &len);
|
|
} else {
|
|
/* XSLT section 7.7
|
|
* "Any other format token indicates a numbering sequence
|
|
* that starts with that token. If an implementation does
|
|
* not support a numbering sequence that starts with that
|
|
* token, it must use a format token of 1."
|
|
*/
|
|
tokens->tokens[tokens->nTokens].token = '0';
|
|
tokens->tokens[tokens->nTokens].width = 1;
|
|
}
|
|
/*
|
|
* Skip over remaining alphanumeric characters from the Nd
|
|
* (Number, decimal digit), Nl (Number, letter), No (Number,
|
|
* other), Lu (Letter, uppercase), Ll (Letter, lowercase), Lt
|
|
* (Letters, titlecase), Lm (Letters, modifiers), and Lo
|
|
* (Letters, other (uncased)) Unicode categories. This happens
|
|
* to correspond to the Letter and Digit classes from XML (and
|
|
* one wonders why XSLT doesn't refer to these instead).
|
|
*/
|
|
while (IS_LETTER(val) || IS_DIGIT(val)) {
|
|
ix += len;
|
|
val = xmlStringCurrentChar(NULL, format+ix, &len);
|
|
}
|
|
|
|
/*
|
|
* Insert temporary non-alphanumeric final tooken.
|
|
*/
|
|
j = ix;
|
|
while (! (IS_LETTER(val) || IS_DIGIT(val))) {
|
|
if (val == 0)
|
|
break; /* while */
|
|
ix += len;
|
|
val = xmlStringCurrentChar(NULL, format+ix, &len);
|
|
}
|
|
if (ix > j)
|
|
tokens->end = xmlStrndup(&format[j], ix - j);
|
|
}
|
|
}
|
|
|
|
static void
|
|
xsltNumberFormatInsertNumbers(xsltNumberDataPtr data,
|
|
double *numbers,
|
|
int numbers_max,
|
|
xsltFormatPtr tokens,
|
|
xmlBufferPtr buffer)
|
|
{
|
|
int i = 0;
|
|
double number;
|
|
xsltFormatTokenPtr token;
|
|
|
|
/*
|
|
* Handle initial non-alphanumeric token
|
|
*/
|
|
if (tokens->start != NULL)
|
|
xmlBufferCat(buffer, tokens->start);
|
|
|
|
for (i = 0; i < numbers_max; i++) {
|
|
/* Insert number */
|
|
number = numbers[(numbers_max - 1) - i];
|
|
/* Round to nearest like XSLT 2.0 */
|
|
number = floor(number + 0.5);
|
|
/*
|
|
* XSLT 1.0 isn't clear on how to handle negative numbers, but XSLT
|
|
* 2.0 says:
|
|
*
|
|
* It is a non-recoverable dynamic error if any undiscarded item
|
|
* in the atomized sequence supplied as the value of the value
|
|
* attribute of xsl:number cannot be converted to an integer, or
|
|
* if the resulting integer is less than 0 (zero).
|
|
*/
|
|
if (number < 0.0) {
|
|
xsltTransformError(NULL, NULL, NULL,
|
|
"xsl-number : negative value\n");
|
|
/* Recover by treating negative values as zero. */
|
|
number = 0.0;
|
|
}
|
|
if (i < tokens->nTokens) {
|
|
/*
|
|
* The "n"th format token will be used to format the "n"th
|
|
* number in the list
|
|
*/
|
|
token = &(tokens->tokens[i]);
|
|
} else if (tokens->nTokens > 0) {
|
|
/*
|
|
* If there are more numbers than format tokens, then the
|
|
* last format token will be used to format the remaining
|
|
* numbers.
|
|
*/
|
|
token = &(tokens->tokens[tokens->nTokens - 1]);
|
|
} else {
|
|
/*
|
|
* If there are no format tokens, then a format token of
|
|
* 1 is used to format all numbers.
|
|
*/
|
|
token = &default_token;
|
|
}
|
|
|
|
/* Print separator, except for the first number */
|
|
if (i > 0) {
|
|
if (token->separator != NULL)
|
|
xmlBufferCat(buffer, token->separator);
|
|
else
|
|
xmlBufferCCat(buffer, DEFAULT_SEPARATOR);
|
|
}
|
|
|
|
switch (xmlXPathIsInf(number)) {
|
|
case -1:
|
|
xmlBufferCCat(buffer, "-Infinity");
|
|
break;
|
|
case 1:
|
|
xmlBufferCCat(buffer, "Infinity");
|
|
break;
|
|
default:
|
|
if (xmlXPathIsNaN(number)) {
|
|
xmlBufferCCat(buffer, "NaN");
|
|
} else {
|
|
|
|
switch (token->token) {
|
|
case 'A':
|
|
xsltNumberFormatAlpha(data, buffer, number, TRUE);
|
|
break;
|
|
case 'a':
|
|
xsltNumberFormatAlpha(data, buffer, number, FALSE);
|
|
break;
|
|
case 'I':
|
|
xsltNumberFormatRoman(data, buffer, number, TRUE);
|
|
break;
|
|
case 'i':
|
|
xsltNumberFormatRoman(data, buffer, number, FALSE);
|
|
break;
|
|
default:
|
|
if (IS_DIGIT_ZERO(token->token)) {
|
|
xsltNumberFormatDecimal(buffer,
|
|
number,
|
|
token->token,
|
|
token->width,
|
|
data->digitsPerGroup,
|
|
data->groupingCharacter,
|
|
data->groupingCharacterLen);
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Handle final non-alphanumeric token
|
|
*/
|
|
if (tokens->end != NULL)
|
|
xmlBufferCat(buffer, tokens->end);
|
|
|
|
}
|
|
|
|
static int
|
|
xsltTestCompMatchCount(xsltTransformContextPtr context,
|
|
xmlNodePtr node,
|
|
xsltCompMatchPtr countPat,
|
|
xmlNodePtr cur)
|
|
{
|
|
if (countPat != NULL) {
|
|
return xsltTestCompMatchList(context, node, countPat);
|
|
}
|
|
else {
|
|
/*
|
|
* 7.7 Numbering
|
|
*
|
|
* If count attribute is not specified, then it defaults to the
|
|
* pattern that matches any node with the same node type as the
|
|
* current node and, if the current node has an expanded-name, with
|
|
* the same expanded-name as the current node.
|
|
*/
|
|
if (node->type != cur->type)
|
|
return 0;
|
|
if (node->type == XML_NAMESPACE_DECL)
|
|
/*
|
|
* Namespace nodes have no preceding siblings and no parents
|
|
* that are namespace nodes. This means that node == cur.
|
|
*/
|
|
return 1;
|
|
/* TODO: Skip node types without expanded names like text nodes. */
|
|
if (!xmlStrEqual(node->name, cur->name))
|
|
return 0;
|
|
if (node->ns == cur->ns)
|
|
return 1;
|
|
if ((node->ns == NULL) || (cur->ns == NULL))
|
|
return 0;
|
|
return (xmlStrEqual(node->ns->href, cur->ns->href));
|
|
}
|
|
}
|
|
|
|
static int
|
|
xsltNumberFormatGetAnyLevel(xsltTransformContextPtr context,
|
|
xmlNodePtr node,
|
|
xsltCompMatchPtr countPat,
|
|
xsltCompMatchPtr fromPat,
|
|
double *array)
|
|
{
|
|
int amount = 0;
|
|
int cnt = 0;
|
|
xmlNodePtr cur = node;
|
|
|
|
while (cur != NULL) {
|
|
/* process current node */
|
|
if (xsltTestCompMatchCount(context, cur, countPat, node))
|
|
cnt++;
|
|
if ((fromPat != NULL) &&
|
|
xsltTestCompMatchList(context, cur, fromPat)) {
|
|
break; /* while */
|
|
}
|
|
|
|
/* Skip to next preceding or ancestor */
|
|
if ((cur->type == XML_DOCUMENT_NODE) ||
|
|
#ifdef LIBXML_DOCB_ENABLED
|
|
(cur->type == XML_DOCB_DOCUMENT_NODE) ||
|
|
#endif
|
|
(cur->type == XML_HTML_DOCUMENT_NODE))
|
|
break; /* while */
|
|
|
|
if (cur->type == XML_NAMESPACE_DECL) {
|
|
/*
|
|
* The XPath module stores the parent of a namespace node in
|
|
* the ns->next field.
|
|
*/
|
|
cur = (xmlNodePtr) ((xmlNsPtr) cur)->next;
|
|
} else if (cur->type == XML_ATTRIBUTE_NODE) {
|
|
cur = cur->parent;
|
|
} else {
|
|
while ((cur->prev != NULL) && ((cur->prev->type == XML_DTD_NODE) ||
|
|
(cur->prev->type == XML_XINCLUDE_START) ||
|
|
(cur->prev->type == XML_XINCLUDE_END)))
|
|
cur = cur->prev;
|
|
if (cur->prev != NULL) {
|
|
for (cur = cur->prev; cur->last != NULL; cur = cur->last);
|
|
} else {
|
|
cur = cur->parent;
|
|
}
|
|
}
|
|
}
|
|
|
|
array[amount++] = (double) cnt;
|
|
|
|
return(amount);
|
|
}
|
|
|
|
static int
|
|
xsltNumberFormatGetMultipleLevel(xsltTransformContextPtr context,
|
|
xmlNodePtr node,
|
|
xsltCompMatchPtr countPat,
|
|
xsltCompMatchPtr fromPat,
|
|
double *array,
|
|
int max)
|
|
{
|
|
int amount = 0;
|
|
int cnt;
|
|
xmlNodePtr oldCtxtNode;
|
|
xmlNodePtr ancestor;
|
|
xmlNodePtr preceding;
|
|
xmlXPathParserContextPtr parser;
|
|
|
|
oldCtxtNode = context->xpathCtxt->node;
|
|
parser = xmlXPathNewParserContext(NULL, context->xpathCtxt);
|
|
if (parser) {
|
|
/* ancestor-or-self::*[count] */
|
|
ancestor = node;
|
|
while ((ancestor != NULL) && (ancestor->type != XML_DOCUMENT_NODE)) {
|
|
if ((fromPat != NULL) &&
|
|
xsltTestCompMatchList(context, ancestor, fromPat))
|
|
break; /* for */
|
|
|
|
/*
|
|
* The xmlXPathNext* iterators require that the context node is
|
|
* set to the start node. Calls to xsltTestCompMatch* may also
|
|
* leave the context node in an undefined state, so make sure
|
|
* that the context node is reset before each iterator invocation.
|
|
*/
|
|
|
|
if (xsltTestCompMatchCount(context, ancestor, countPat, node)) {
|
|
/* count(preceding-sibling::*) */
|
|
cnt = 1;
|
|
context->xpathCtxt->node = ancestor;
|
|
preceding = xmlXPathNextPrecedingSibling(parser, ancestor);
|
|
while (preceding != NULL) {
|
|
if (xsltTestCompMatchCount(context, preceding, countPat,
|
|
node))
|
|
cnt++;
|
|
context->xpathCtxt->node = ancestor;
|
|
preceding =
|
|
xmlXPathNextPrecedingSibling(parser, preceding);
|
|
}
|
|
array[amount++] = (double)cnt;
|
|
if (amount >= max)
|
|
break; /* for */
|
|
}
|
|
context->xpathCtxt->node = node;
|
|
ancestor = xmlXPathNextAncestor(parser, ancestor);
|
|
}
|
|
xmlXPathFreeParserContext(parser);
|
|
}
|
|
context->xpathCtxt->node = oldCtxtNode;
|
|
return amount;
|
|
}
|
|
|
|
static int
|
|
xsltNumberFormatGetValue(xmlXPathContextPtr context,
|
|
xmlNodePtr node,
|
|
const xmlChar *value,
|
|
double *number)
|
|
{
|
|
int amount = 0;
|
|
xmlBufferPtr pattern;
|
|
xmlXPathObjectPtr obj;
|
|
|
|
pattern = xmlBufferCreate();
|
|
if (pattern != NULL) {
|
|
xmlBufferCCat(pattern, "number(");
|
|
xmlBufferCat(pattern, value);
|
|
xmlBufferCCat(pattern, ")");
|
|
context->node = node;
|
|
obj = xmlXPathEvalExpression(xmlBufferContent(pattern),
|
|
context);
|
|
if (obj != NULL) {
|
|
*number = obj->floatval;
|
|
amount++;
|
|
xmlXPathFreeObject(obj);
|
|
}
|
|
xmlBufferFree(pattern);
|
|
}
|
|
return amount;
|
|
}
|
|
|
|
/**
|
|
* xsltNumberFormat:
|
|
* @ctxt: the XSLT transformation context
|
|
* @data: the formatting information
|
|
* @node: the data to format
|
|
*
|
|
* Convert one number.
|
|
*/
|
|
void
|
|
xsltNumberFormat(xsltTransformContextPtr ctxt,
|
|
xsltNumberDataPtr data,
|
|
xmlNodePtr node)
|
|
{
|
|
xmlBufferPtr output = NULL;
|
|
int amount, i;
|
|
double number;
|
|
xsltFormat tokens;
|
|
|
|
if (data->format != NULL) {
|
|
xsltNumberFormatTokenize(data->format, &tokens);
|
|
}
|
|
else {
|
|
xmlChar *format;
|
|
|
|
/* The format needs to be recomputed each time */
|
|
if (data->has_format == 0)
|
|
return;
|
|
format = xsltEvalAttrValueTemplate(ctxt, data->node,
|
|
(const xmlChar *) "format",
|
|
XSLT_NAMESPACE);
|
|
if (format == NULL)
|
|
return;
|
|
xsltNumberFormatTokenize(format, &tokens);
|
|
xmlFree(format);
|
|
}
|
|
|
|
output = xmlBufferCreate();
|
|
if (output == NULL)
|
|
goto XSLT_NUMBER_FORMAT_END;
|
|
|
|
/*
|
|
* Evaluate the XPath expression to find the value(s)
|
|
*/
|
|
if (data->value) {
|
|
amount = xsltNumberFormatGetValue(ctxt->xpathCtxt,
|
|
node,
|
|
data->value,
|
|
&number);
|
|
if (amount == 1) {
|
|
xsltNumberFormatInsertNumbers(data,
|
|
&number,
|
|
1,
|
|
&tokens,
|
|
output);
|
|
}
|
|
|
|
} else if (data->level) {
|
|
|
|
if (xmlStrEqual(data->level, (const xmlChar *) "single")) {
|
|
amount = xsltNumberFormatGetMultipleLevel(ctxt,
|
|
node,
|
|
data->countPat,
|
|
data->fromPat,
|
|
&number,
|
|
1);
|
|
if (amount == 1) {
|
|
xsltNumberFormatInsertNumbers(data,
|
|
&number,
|
|
1,
|
|
&tokens,
|
|
output);
|
|
}
|
|
} else if (xmlStrEqual(data->level, (const xmlChar *) "multiple")) {
|
|
double numarray[1024];
|
|
int max = sizeof(numarray)/sizeof(numarray[0]);
|
|
amount = xsltNumberFormatGetMultipleLevel(ctxt,
|
|
node,
|
|
data->countPat,
|
|
data->fromPat,
|
|
numarray,
|
|
max);
|
|
if (amount > 0) {
|
|
xsltNumberFormatInsertNumbers(data,
|
|
numarray,
|
|
amount,
|
|
&tokens,
|
|
output);
|
|
}
|
|
} else if (xmlStrEqual(data->level, (const xmlChar *) "any")) {
|
|
amount = xsltNumberFormatGetAnyLevel(ctxt,
|
|
node,
|
|
data->countPat,
|
|
data->fromPat,
|
|
&number);
|
|
if (amount > 0) {
|
|
xsltNumberFormatInsertNumbers(data,
|
|
&number,
|
|
1,
|
|
&tokens,
|
|
output);
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Unlike `match` patterns, `count` and `from` patterns can contain
|
|
* variable references, so we have to clear the pattern match
|
|
* cache if the "direct" matching algorithm was used.
|
|
*/
|
|
if (data->countPat != NULL)
|
|
xsltCompMatchClearCache(ctxt, data->countPat);
|
|
if (data->fromPat != NULL)
|
|
xsltCompMatchClearCache(ctxt, data->fromPat);
|
|
}
|
|
/* Insert number as text node */
|
|
xsltCopyTextString(ctxt, ctxt->insert, xmlBufferContent(output), 0);
|
|
|
|
xmlBufferFree(output);
|
|
|
|
XSLT_NUMBER_FORMAT_END:
|
|
if (tokens.start != NULL)
|
|
xmlFree(tokens.start);
|
|
if (tokens.end != NULL)
|
|
xmlFree(tokens.end);
|
|
for (i = 0;i < tokens.nTokens;i++) {
|
|
if (tokens.tokens[i].separator != NULL)
|
|
xmlFree(tokens.tokens[i].separator);
|
|
}
|
|
}
|
|
|
|
static int
|
|
xsltFormatNumberPreSuffix(xsltDecimalFormatPtr self, xmlChar **format, xsltFormatNumberInfoPtr info)
|
|
{
|
|
/* will hold total length of prefix/suffix without quote characters */
|
|
int count=0;
|
|
int len;
|
|
|
|
while (1) {
|
|
/*
|
|
* prefix / suffix ends at end of string or at
|
|
* first 'special' character
|
|
*/
|
|
if (**format == 0)
|
|
return count;
|
|
/* if next character 'escaped' just count it */
|
|
if (**format == SYMBOL_QUOTE) {
|
|
if (*++(*format) == 0)
|
|
return -1;
|
|
}
|
|
else if (IS_SPECIAL(self, *format))
|
|
return count;
|
|
/*
|
|
* else treat percent/per-mille as special cases,
|
|
* depending on whether +ve or -ve
|
|
*/
|
|
else {
|
|
/*
|
|
* for +ve prefix/suffix, allow only a
|
|
* single occurence of either
|
|
*/
|
|
if (xsltUTF8Charcmp(*format, self->percent) == 0) {
|
|
if (info->is_multiplier_set)
|
|
return -1;
|
|
info->multiplier = 100;
|
|
info->is_multiplier_set = TRUE;
|
|
} else if (xsltUTF8Charcmp(*format, self->permille) == 0) {
|
|
if (info->is_multiplier_set)
|
|
return -1;
|
|
info->multiplier = 1000;
|
|
info->is_multiplier_set = TRUE;
|
|
}
|
|
}
|
|
|
|
if ((len=xmlUTF8Strsize(*format, 1)) < 1)
|
|
return -1;
|
|
count += len;
|
|
*format += len;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* xsltFormatNumberConversion:
|
|
* @self: the decimal format
|
|
* @format: the format requested
|
|
* @number: the value to format
|
|
* @result: the place to output the result
|
|
*
|
|
* format-number() uses the JDK 1.1 DecimalFormat class:
|
|
*
|
|
* http://java.sun.com/products/jdk/1.1/docs/api/java.text.DecimalFormat.html
|
|
*
|
|
* Structure:
|
|
*
|
|
* pattern := subpattern{;subpattern}
|
|
* subpattern := {prefix}integer{.fraction}{suffix}
|
|
* prefix := '\\u0000'..'\\uFFFD' - specialCharacters
|
|
* suffix := '\\u0000'..'\\uFFFD' - specialCharacters
|
|
* integer := '#'* '0'* '0'
|
|
* fraction := '0'* '#'*
|
|
*
|
|
* Notation:
|
|
* X* 0 or more instances of X
|
|
* (X | Y) either X or Y.
|
|
* X..Y any character from X up to Y, inclusive.
|
|
* S - T characters in S, except those in T
|
|
*
|
|
* Special Characters:
|
|
*
|
|
* Symbol Meaning
|
|
* 0 a digit
|
|
* # a digit, zero shows as absent
|
|
* . placeholder for decimal separator
|
|
* , placeholder for grouping separator.
|
|
* ; separates formats.
|
|
* - default negative prefix.
|
|
* % multiply by 100 and show as percentage
|
|
* ? multiply by 1000 and show as per mille
|
|
* X any other characters can be used in the prefix or suffix
|
|
* ' used to quote special characters in a prefix or suffix.
|
|
*
|
|
* Returns a possible XPath error
|
|
*/
|
|
xmlXPathError
|
|
xsltFormatNumberConversion(xsltDecimalFormatPtr self,
|
|
xmlChar *format,
|
|
double number,
|
|
xmlChar **result)
|
|
{
|
|
xmlXPathError status = XPATH_EXPRESSION_OK;
|
|
xmlBufferPtr buffer;
|
|
xmlChar *the_format, *prefix = NULL, *suffix = NULL;
|
|
xmlChar *nprefix, *nsuffix = NULL;
|
|
int prefix_length, suffix_length = 0, nprefix_length, nsuffix_length;
|
|
double scale;
|
|
int j, len;
|
|
int self_grouping_len;
|
|
xsltFormatNumberInfo format_info;
|
|
/*
|
|
* delayed_multiplier allows a 'trailing' percent or
|
|
* permille to be treated as suffix
|
|
*/
|
|
int delayed_multiplier = 0;
|
|
/* flag to show no -ve format present for -ve number */
|
|
char default_sign = 0;
|
|
/* flag to show error found, should use default format */
|
|
char found_error = 0;
|
|
|
|
if (xmlStrlen(format) <= 0) {
|
|
xsltTransformError(NULL, NULL, NULL,
|
|
"xsltFormatNumberConversion : "
|
|
"Invalid format (0-length)\n");
|
|
}
|
|
*result = NULL;
|
|
switch (xmlXPathIsInf(number)) {
|
|
case -1:
|
|
if (self->minusSign == NULL)
|
|
*result = xmlStrdup(BAD_CAST "-");
|
|
else
|
|
*result = xmlStrdup(self->minusSign);
|
|
/* Intentional fall-through */
|
|
case 1:
|
|
if ((self == NULL) || (self->infinity == NULL))
|
|
*result = xmlStrcat(*result, BAD_CAST "Infinity");
|
|
else
|
|
*result = xmlStrcat(*result, self->infinity);
|
|
return(status);
|
|
default:
|
|
if (xmlXPathIsNaN(number)) {
|
|
if ((self == NULL) || (self->noNumber == NULL))
|
|
*result = xmlStrdup(BAD_CAST "NaN");
|
|
else
|
|
*result = xmlStrdup(self->noNumber);
|
|
return(status);
|
|
}
|
|
}
|
|
|
|
buffer = xmlBufferCreate();
|
|
if (buffer == NULL) {
|
|
return XPATH_MEMORY_ERROR;
|
|
}
|
|
|
|
format_info.integer_hash = 0;
|
|
format_info.integer_digits = 0;
|
|
format_info.frac_digits = 0;
|
|
format_info.frac_hash = 0;
|
|
format_info.group = -1;
|
|
format_info.multiplier = 1;
|
|
format_info.add_decimal = FALSE;
|
|
format_info.is_multiplier_set = FALSE;
|
|
format_info.is_negative_pattern = FALSE;
|
|
|
|
the_format = format;
|
|
|
|
/*
|
|
* First we process the +ve pattern to get percent / permille,
|
|
* as well as main format
|
|
*/
|
|
prefix = the_format;
|
|
prefix_length = xsltFormatNumberPreSuffix(self, &the_format, &format_info);
|
|
if (prefix_length < 0) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
|
|
/*
|
|
* Here we process the "number" part of the format. It gets
|
|
* a little messy because of the percent/per-mille - if that
|
|
* appears at the end, it may be part of the suffix instead
|
|
* of part of the number, so the variable delayed_multiplier
|
|
* is used to handle it
|
|
*/
|
|
self_grouping_len = xmlStrlen(self->grouping);
|
|
while ((*the_format != 0) &&
|
|
(xsltUTF8Charcmp(the_format, self->decimalPoint) != 0) &&
|
|
(xsltUTF8Charcmp(the_format, self->patternSeparator) != 0)) {
|
|
|
|
if (delayed_multiplier != 0) {
|
|
format_info.multiplier = delayed_multiplier;
|
|
format_info.is_multiplier_set = TRUE;
|
|
delayed_multiplier = 0;
|
|
}
|
|
if (xsltUTF8Charcmp(the_format, self->digit) == 0) {
|
|
if (format_info.integer_digits > 0) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
format_info.integer_hash++;
|
|
if (format_info.group >= 0)
|
|
format_info.group++;
|
|
} else if (xsltUTF8Charcmp(the_format, self->zeroDigit) == 0) {
|
|
format_info.integer_digits++;
|
|
if (format_info.group >= 0)
|
|
format_info.group++;
|
|
} else if ((self_grouping_len > 0) &&
|
|
(!xmlStrncmp(the_format, self->grouping, self_grouping_len))) {
|
|
/* Reset group count */
|
|
format_info.group = 0;
|
|
the_format += self_grouping_len;
|
|
continue;
|
|
} else if (xsltUTF8Charcmp(the_format, self->percent) == 0) {
|
|
if (format_info.is_multiplier_set) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
delayed_multiplier = 100;
|
|
} else if (xsltUTF8Charcmp(the_format, self->permille) == 0) {
|
|
if (format_info.is_multiplier_set) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
delayed_multiplier = 1000;
|
|
} else
|
|
break; /* while */
|
|
|
|
if ((len=xmlUTF8Strsize(the_format, 1)) < 1) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
the_format += len;
|
|
|
|
}
|
|
|
|
/* We have finished the integer part, now work on fraction */
|
|
if ( (*the_format != 0) &&
|
|
(xsltUTF8Charcmp(the_format, self->decimalPoint) == 0) ) {
|
|
format_info.add_decimal = TRUE;
|
|
if ((len = xmlUTF8Strsize(the_format, 1)) < 1) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
the_format += len; /* Skip over the decimal */
|
|
}
|
|
|
|
while (*the_format != 0) {
|
|
|
|
if (xsltUTF8Charcmp(the_format, self->zeroDigit) == 0) {
|
|
if (format_info.frac_hash != 0) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
format_info.frac_digits++;
|
|
} else if (xsltUTF8Charcmp(the_format, self->digit) == 0) {
|
|
format_info.frac_hash++;
|
|
} else if (xsltUTF8Charcmp(the_format, self->percent) == 0) {
|
|
if (format_info.is_multiplier_set) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
delayed_multiplier = 100;
|
|
if ((len = xmlUTF8Strsize(the_format, 1)) < 1) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
the_format += len;
|
|
continue; /* while */
|
|
} else if (xsltUTF8Charcmp(the_format, self->permille) == 0) {
|
|
if (format_info.is_multiplier_set) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
delayed_multiplier = 1000;
|
|
if ((len = xmlUTF8Strsize(the_format, 1)) < 1) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
the_format += len;
|
|
continue; /* while */
|
|
} else if (xsltUTF8Charcmp(the_format, self->grouping) != 0) {
|
|
break; /* while */
|
|
}
|
|
if ((len = xmlUTF8Strsize(the_format, 1)) < 1) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
the_format += len;
|
|
if (delayed_multiplier != 0) {
|
|
format_info.multiplier = delayed_multiplier;
|
|
delayed_multiplier = 0;
|
|
format_info.is_multiplier_set = TRUE;
|
|
}
|
|
}
|
|
|
|
/*
|
|
* If delayed_multiplier is set after processing the
|
|
* "number" part, should be in suffix
|
|
*/
|
|
if (delayed_multiplier != 0) {
|
|
the_format -= len;
|
|
delayed_multiplier = 0;
|
|
}
|
|
|
|
suffix = the_format;
|
|
suffix_length = xsltFormatNumberPreSuffix(self, &the_format, &format_info);
|
|
if ( (suffix_length < 0) ||
|
|
((*the_format != 0) &&
|
|
(xsltUTF8Charcmp(the_format, self->patternSeparator) != 0)) ) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
|
|
/*
|
|
* We have processed the +ve prefix, number part and +ve suffix.
|
|
* If the number is -ve, we must substitute the -ve prefix / suffix
|
|
*/
|
|
if (number < 0) {
|
|
/*
|
|
* Note that j is the number of UTF8 chars before the separator,
|
|
* not the number of bytes! (bug 151975)
|
|
*/
|
|
j = xmlUTF8Strloc(format, self->patternSeparator);
|
|
if (j < 0) {
|
|
/* No -ve pattern present, so use default signing */
|
|
default_sign = 1;
|
|
}
|
|
else {
|
|
/* Skip over pattern separator (accounting for UTF8) */
|
|
the_format = (xmlChar *)xmlUTF8Strpos(format, j + 1);
|
|
/*
|
|
* Flag changes interpretation of percent/permille
|
|
* in -ve pattern
|
|
*/
|
|
format_info.is_negative_pattern = TRUE;
|
|
format_info.is_multiplier_set = FALSE;
|
|
|
|
/* First do the -ve prefix */
|
|
nprefix = the_format;
|
|
nprefix_length = xsltFormatNumberPreSuffix(self,
|
|
&the_format, &format_info);
|
|
if (nprefix_length<0) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
|
|
while (*the_format != 0) {
|
|
if ( (xsltUTF8Charcmp(the_format, (self)->percent) == 0) ||
|
|
(xsltUTF8Charcmp(the_format, (self)->permille)== 0) ) {
|
|
if (format_info.is_multiplier_set) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
format_info.is_multiplier_set = TRUE;
|
|
delayed_multiplier = 1;
|
|
}
|
|
else if (IS_SPECIAL(self, the_format))
|
|
delayed_multiplier = 0;
|
|
else
|
|
break; /* while */
|
|
if ((len = xmlUTF8Strsize(the_format, 1)) < 1) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
the_format += len;
|
|
}
|
|
if (delayed_multiplier != 0) {
|
|
format_info.is_multiplier_set = FALSE;
|
|
the_format -= len;
|
|
}
|
|
|
|
/* Finally do the -ve suffix */
|
|
if (*the_format != 0) {
|
|
nsuffix = the_format;
|
|
nsuffix_length = xsltFormatNumberPreSuffix(self,
|
|
&the_format, &format_info);
|
|
if (nsuffix_length < 0) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
}
|
|
else
|
|
nsuffix_length = 0;
|
|
if (*the_format != 0) {
|
|
found_error = 1;
|
|
goto OUTPUT_NUMBER;
|
|
}
|
|
/*
|
|
* Here's another Java peculiarity:
|
|
* if -ve prefix/suffix == +ve ones, discard & use default
|
|
*/
|
|
if ((nprefix_length != prefix_length) ||
|
|
(nsuffix_length != suffix_length) ||
|
|
((nprefix_length > 0) &&
|
|
(xmlStrncmp(nprefix, prefix, prefix_length) !=0 )) ||
|
|
((nsuffix_length > 0) &&
|
|
(xmlStrncmp(nsuffix, suffix, suffix_length) !=0 ))) {
|
|
prefix = nprefix;
|
|
prefix_length = nprefix_length;
|
|
suffix = nsuffix;
|
|
suffix_length = nsuffix_length;
|
|
} /* else {
|
|
default_sign = 1;
|
|
}
|
|
*/
|
|
}
|
|
}
|
|
|
|
OUTPUT_NUMBER:
|
|
if (found_error != 0) {
|
|
xsltTransformError(NULL, NULL, NULL,
|
|
"xsltFormatNumberConversion : "
|
|
"error in format string '%s', using default\n", format);
|
|
default_sign = (number < 0.0) ? 1 : 0;
|
|
prefix_length = suffix_length = 0;
|
|
format_info.integer_hash = 0;
|
|
format_info.integer_digits = 1;
|
|
format_info.frac_digits = 1;
|
|
format_info.frac_hash = 4;
|
|
format_info.group = -1;
|
|
format_info.multiplier = 1;
|
|
format_info.add_decimal = TRUE;
|
|
}
|
|
|
|
/* Ready to output our number. First see if "default sign" is required */
|
|
if (default_sign != 0)
|
|
xmlBufferAdd(buffer, self->minusSign, xmlUTF8Strsize(self->minusSign, 1));
|
|
|
|
/* Put the prefix into the buffer */
|
|
for (j = 0; j < prefix_length; ) {
|
|
if (*prefix == SYMBOL_QUOTE)
|
|
prefix++;
|
|
len = xmlUTF8Strsize(prefix, 1);
|
|
xmlBufferAdd(buffer, prefix, len);
|
|
prefix += len;
|
|
j += len;
|
|
}
|
|
|
|
/* Next do the integer part of the number */
|
|
number = fabs(number) * (double)format_info.multiplier;
|
|
scale = pow(10.0, (double)(format_info.frac_digits + format_info.frac_hash));
|
|
number = floor((scale * number + 0.5)) / scale;
|
|
if ((self->grouping != NULL) &&
|
|
(self->grouping[0] != 0)) {
|
|
int gchar;
|
|
|
|
len = xmlStrlen(self->grouping);
|
|
gchar = xsltGetUTF8Char(self->grouping, &len);
|
|
xsltNumberFormatDecimal(buffer, floor(number), self->zeroDigit[0],
|
|
format_info.integer_digits,
|
|
format_info.group,
|
|
gchar, len);
|
|
} else
|
|
xsltNumberFormatDecimal(buffer, floor(number), self->zeroDigit[0],
|
|
format_info.integer_digits,
|
|
format_info.group,
|
|
',', 1);
|
|
|
|
/* Special case: java treats '.#' like '.0', '.##' like '.0#', etc. */
|
|
if ((format_info.integer_digits + format_info.integer_hash +
|
|
format_info.frac_digits == 0) && (format_info.frac_hash > 0)) {
|
|
++format_info.frac_digits;
|
|
--format_info.frac_hash;
|
|
}
|
|
|
|
/* Add leading zero, if required */
|
|
if ((floor(number) == 0) &&
|
|
(format_info.integer_digits + format_info.frac_digits == 0)) {
|
|
xmlBufferAdd(buffer, self->zeroDigit, xmlUTF8Strsize(self->zeroDigit, 1));
|
|
}
|
|
|
|
/* Next the fractional part, if required */
|
|
if (format_info.frac_digits + format_info.frac_hash == 0) {
|
|
if (format_info.add_decimal)
|
|
xmlBufferAdd(buffer, self->decimalPoint,
|
|
xmlUTF8Strsize(self->decimalPoint, 1));
|
|
}
|
|
else {
|
|
number -= floor(number);
|
|
if ((number != 0) || (format_info.frac_digits != 0)) {
|
|
xmlBufferAdd(buffer, self->decimalPoint,
|
|
xmlUTF8Strsize(self->decimalPoint, 1));
|
|
number = floor(scale * number + 0.5);
|
|
for (j = format_info.frac_hash; j > 0; j--) {
|
|
if (fmod(number, 10.0) >= 1.0)
|
|
break; /* for */
|
|
number /= 10.0;
|
|
}
|
|
xsltNumberFormatDecimal(buffer, floor(number), self->zeroDigit[0],
|
|
format_info.frac_digits + j,
|
|
0, 0, 0);
|
|
}
|
|
}
|
|
/* Put the suffix into the buffer */
|
|
for (j = 0; j < suffix_length; ) {
|
|
if (*suffix == SYMBOL_QUOTE)
|
|
suffix++;
|
|
len = xmlUTF8Strsize(suffix, 1);
|
|
xmlBufferAdd(buffer, suffix, len);
|
|
suffix += len;
|
|
j += len;
|
|
}
|
|
|
|
*result = xmlStrdup(xmlBufferContent(buffer));
|
|
xmlBufferFree(buffer);
|
|
return status;
|
|
}
|
|
|