mirror of
https://github.com/reactos/wine.git
synced 2024-11-29 22:50:43 +00:00
2657 lines
82 KiB
C
2657 lines
82 KiB
C
/*
|
|
* Variant formatting functions
|
|
*
|
|
* Copyright 2008 Damjan Jovanovic
|
|
* Copyright 2003 Jon Griffiths
|
|
*
|
|
* This library is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU Lesser General Public
|
|
* License as published by the Free Software Foundation; either
|
|
* version 2.1 of the License, or (at your option) any later version.
|
|
*
|
|
* This library is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
* Lesser General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU Lesser General Public
|
|
* License along with this library; if not, write to the Free Software
|
|
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
|
|
*
|
|
* NOTES
|
|
* Since the formatting functions aren't properly documented, I used the
|
|
* Visual Basic documentation as a guide to implementing these functions. This
|
|
* means that some named or user-defined formats may work slightly differently.
|
|
* Please submit a test case if you find a difference.
|
|
*/
|
|
|
|
#include "config.h"
|
|
|
|
#include <string.h>
|
|
#include <stdlib.h>
|
|
#include <stdarg.h>
|
|
#include <stdio.h>
|
|
|
|
#define NONAMELESSUNION
|
|
#define NONAMELESSSTRUCT
|
|
#include "windef.h"
|
|
#include "winbase.h"
|
|
#include "wine/unicode.h"
|
|
#include "winerror.h"
|
|
#include "variant.h"
|
|
#include "wine/debug.h"
|
|
|
|
WINE_DEFAULT_DEBUG_CHANNEL(variant);
|
|
|
|
/* Make sure internal conversions to strings use the '.','+'/'-' and ','
|
|
* format chars from the US locale. This enables us to parse the created
|
|
* strings to determine the number of decimal places, exponent, etc.
|
|
*/
|
|
#define LCID_US MAKELCID(MAKELANGID(LANG_ENGLISH,SUBLANG_ENGLISH_US),SORT_DEFAULT)
|
|
|
|
static const WCHAR szPercent_d[] = { '%','d','\0' };
|
|
static const WCHAR szPercentZeroTwo_d[] = { '%','0','2','d','\0' };
|
|
static const WCHAR szPercentZeroStar_d[] = { '%','0','*','d','\0' };
|
|
|
|
#if 0
|
|
#define dump_tokens(rgb) do { \
|
|
int i_; TRACE("Tokens->{\n"); \
|
|
for (i_ = 0; i_ < rgb[0]; i_++) \
|
|
TRACE("%s0x%02x", i_?",":"",rgb[i_]); \
|
|
TRACE(" }\n"); \
|
|
} while(0)
|
|
#endif
|
|
|
|
/******************************************************************************
|
|
* Variant-Formats {OLEAUT32}
|
|
*
|
|
* NOTES
|
|
* When formatting a variant a variety of format strings may be used to generate
|
|
* different kinds of formatted output. A format string consists of either a named
|
|
* format, or a user-defined format.
|
|
*
|
|
* The following named formats are defined:
|
|
*| Name Description
|
|
*| ---- -----------
|
|
*| General Date Display Date, and time for non-integer values
|
|
*| Short Date Short date format as defined by locale settings
|
|
*| Medium Date Medium date format as defined by locale settings
|
|
*| Long Date Long date format as defined by locale settings
|
|
*| Short Time Short Time format as defined by locale settings
|
|
*| Medium Time Medium time format as defined by locale settings
|
|
*| Long Time Long time format as defined by locale settings
|
|
*| True/False Localised text of "True" or "False"
|
|
*| Yes/No Localised text of "Yes" or "No"
|
|
*| On/Off Localised text of "On" or "Off"
|
|
*| General Number No thousands separator. No decimal points for integers
|
|
*| Currency General currency format using localised characters
|
|
*| Fixed At least one whole and two fractional digits
|
|
*| Standard Same as 'Fixed', but including decimal separators
|
|
*| Percent Multiply by 100 and display a trailing '%' character
|
|
*| Scientific Display with exponent
|
|
*
|
|
* User-defined formats consist of a combination of tokens and literal
|
|
* characters. Literal characters are copied unmodified to the formatted
|
|
* output at the position they occupy in the format string. Any character
|
|
* that is not recognised as a token is treated as a literal. A literal can
|
|
* also be specified by preceding it with a backslash character
|
|
* (e.g. "\L\i\t\e\r\a\l") or enclosing it in double quotes.
|
|
*
|
|
* A user-defined format can have up to 4 sections, depending on the type of
|
|
* format. The following table lists sections and their meaning:
|
|
*| Format Type Sections Meaning
|
|
*| ----------- -------- -------
|
|
*| Number 1 Use the same format for all numbers
|
|
*| Number 2 Use format 1 for positive and 2 for negative numbers
|
|
*| Number 3 Use format 1 for positive, 2 for zero, and 3
|
|
*| for negative numbers.
|
|
*| Number 4 Use format 1 for positive, 2 for zero, 3 for
|
|
*| negative, and 4 for null numbers.
|
|
*| String 1 Use the same format for all strings
|
|
*| String 2 Use format 2 for null and empty strings, otherwise
|
|
*| use format 1.
|
|
*| Date 1 Use the same format for all dates
|
|
*
|
|
* The formatting tokens fall into several categories depending on the type
|
|
* of formatted output. For more information on each type, see
|
|
* VarFormat-Dates(), VarFormat-Strings() and VarFormat-Numbers().
|
|
*
|
|
* SEE ALSO
|
|
* VarTokenizeFormatString(), VarFormatFromTokens(), VarFormat(),
|
|
* VarFormatDateTime(), VarFormatNumber(), VarFormatCurrency().
|
|
*/
|
|
|
|
/******************************************************************************
|
|
* VarFormat-Strings {OLEAUT32}
|
|
*
|
|
* NOTES
|
|
* When formatting a variant as a string, it is first converted to a VT_BSTR.
|
|
* The user-format string defines which characters are copied into which
|
|
* positions in the output string. Literals may be inserted in the format
|
|
* string. When creating the formatted string, excess characters in the string
|
|
* (those not consumed by a token) are appended to the end of the output. If
|
|
* there are more tokens than characters in the string to format, spaces will
|
|
* be inserted at the start of the string if the '@' token was used.
|
|
*
|
|
* By default strings are converted to lowercase, or uppercase if the '>' token
|
|
* is encountered. This applies to the whole string: it is not possible to
|
|
* generate a mixed-case output string.
|
|
*
|
|
* In user-defined string formats, the following tokens are recognised:
|
|
*| Token Description
|
|
*| ----- -----------
|
|
*| '@' Copy a char from the source, or a space if no chars are left.
|
|
*| '&' Copy a char from the source, or write nothing if no chars are left.
|
|
*| '<' Output the whole string as lower-case (the default).
|
|
*| '>' Output the whole string as upper-case.
|
|
*| '!' MSDN indicates that this character should cause right-to-left
|
|
*| copying, however tests show that it is tokenised but not processed.
|
|
*/
|
|
|
|
/*
|
|
* Common format definitions
|
|
*/
|
|
|
|
/* Format types */
|
|
#define FMT_TYPE_UNKNOWN 0x0
|
|
#define FMT_TYPE_GENERAL 0x1
|
|
#define FMT_TYPE_NUMBER 0x2
|
|
#define FMT_TYPE_DATE 0x3
|
|
#define FMT_TYPE_STRING 0x4
|
|
|
|
#define FMT_TO_STRING 0x0 /* If header->size == this, act like VB's Str() fn */
|
|
|
|
typedef struct tagFMT_SHORT_HEADER
|
|
{
|
|
BYTE size; /* Size of tokenised block (including header), or FMT_TO_STRING */
|
|
BYTE type; /* Allowable types (FMT_TYPE_*) */
|
|
BYTE offset[1]; /* Offset of the first (and only) format section */
|
|
} FMT_SHORT_HEADER;
|
|
|
|
typedef struct tagFMT_HEADER
|
|
{
|
|
BYTE size; /* Total size of the whole tokenised block (including header) */
|
|
BYTE type; /* Allowable types (FMT_TYPE_*) */
|
|
BYTE starts[4]; /* Offset of each of the 4 format sections, or 0 if none */
|
|
} FMT_HEADER;
|
|
|
|
#define FmtGetPositive(x) (x->starts[0])
|
|
#define FmtGetNegative(x) (x->starts[1] ? x->starts[1] : x->starts[0])
|
|
#define FmtGetZero(x) (x->starts[2] ? x->starts[2] : x->starts[0])
|
|
#define FmtGetNull(x) (x->starts[3] ? x->starts[3] : x->starts[0])
|
|
|
|
/*
|
|
* String formats
|
|
*/
|
|
|
|
#define FMT_FLAG_LT 0x1 /* Has '<' (lower case) */
|
|
#define FMT_FLAG_GT 0x2 /* Has '>' (upper case) */
|
|
#define FMT_FLAG_RTL 0x4 /* Has '!' (Copy right to left) */
|
|
|
|
typedef struct tagFMT_STRING_HEADER
|
|
{
|
|
BYTE flags; /* LT, GT, RTL */
|
|
BYTE unknown1;
|
|
BYTE unknown2;
|
|
BYTE copy_chars; /* Number of chars to be copied */
|
|
BYTE unknown3;
|
|
} FMT_STRING_HEADER;
|
|
|
|
/*
|
|
* Number formats
|
|
*/
|
|
|
|
#define FMT_FLAG_PERCENT 0x1 /* Has '%' (Percentage) */
|
|
#define FMT_FLAG_EXPONENT 0x2 /* Has 'e' (Exponent/Scientific notation) */
|
|
#define FMT_FLAG_THOUSANDS 0x4 /* Has ',' (Standard use of the thousands separator) */
|
|
#define FMT_FLAG_BOOL 0x20 /* Boolean format */
|
|
|
|
typedef struct tagFMT_NUMBER_HEADER
|
|
{
|
|
BYTE flags; /* PERCENT, EXPONENT, THOUSANDS, BOOL */
|
|
BYTE multiplier; /* Multiplier, 100 for percentages */
|
|
BYTE divisor; /* Divisor, 1000 if '%%' was used */
|
|
BYTE whole; /* Number of digits before the decimal point */
|
|
BYTE fractional; /* Number of digits after the decimal point */
|
|
} FMT_NUMBER_HEADER;
|
|
|
|
/*
|
|
* Date Formats
|
|
*/
|
|
typedef struct tagFMT_DATE_HEADER
|
|
{
|
|
BYTE flags;
|
|
BYTE unknown1;
|
|
BYTE unknown2;
|
|
BYTE unknown3;
|
|
BYTE unknown4;
|
|
} FMT_DATE_HEADER;
|
|
|
|
/*
|
|
* Format token values
|
|
*/
|
|
#define FMT_GEN_COPY 0x00 /* \n, "lit" => 0,pos,len: Copy len chars from input+pos */
|
|
#define FMT_GEN_INLINE 0x01 /* => 1,len,[chars]: Copy len chars from token stream */
|
|
#define FMT_GEN_END 0x02 /* \0,; => 2: End of the tokenised format */
|
|
#define FMT_DATE_TIME_SEP 0x03 /* Time separator char */
|
|
#define FMT_DATE_DATE_SEP 0x04 /* Date separator char */
|
|
#define FMT_DATE_GENERAL 0x05 /* General format date */
|
|
#define FMT_DATE_QUARTER 0x06 /* Quarter of the year from 1-4 */
|
|
#define FMT_DATE_TIME_SYS 0x07 /* System long time format */
|
|
#define FMT_DATE_DAY 0x08 /* Day with no leading 0 */
|
|
#define FMT_DATE_DAY_0 0x09 /* Day with leading 0 */
|
|
#define FMT_DATE_DAY_SHORT 0x0A /* Short day name */
|
|
#define FMT_DATE_DAY_LONG 0x0B /* Long day name */
|
|
#define FMT_DATE_SHORT 0x0C /* Short date format */
|
|
#define FMT_DATE_LONG 0x0D /* Long date format */
|
|
#define FMT_DATE_MEDIUM 0x0E /* Medium date format */
|
|
#define FMT_DATE_DAY_WEEK 0x0F /* First day of the week */
|
|
#define FMT_DATE_WEEK_YEAR 0x10 /* First week of the year */
|
|
#define FMT_DATE_MON 0x11 /* Month with no leading 0 */
|
|
#define FMT_DATE_MON_0 0x12 /* Month with leading 0 */
|
|
#define FMT_DATE_MON_SHORT 0x13 /* Short month name */
|
|
#define FMT_DATE_MON_LONG 0x14 /* Long month name */
|
|
#define FMT_DATE_YEAR_DOY 0x15 /* Day of the year with no leading 0 */
|
|
#define FMT_DATE_YEAR_0 0x16 /* 2 digit year with leading 0 */
|
|
/* NOTE: token 0x17 is not defined, 'yyy' is not valid */
|
|
#define FMT_DATE_YEAR_LONG 0x18 /* 4 digit year */
|
|
#define FMT_DATE_MIN 0x1A /* Minutes with no leading 0 */
|
|
#define FMT_DATE_MIN_0 0x1B /* Minutes with leading 0 */
|
|
#define FMT_DATE_SEC 0x1C /* Seconds with no leading 0 */
|
|
#define FMT_DATE_SEC_0 0x1D /* Seconds with leading 0 */
|
|
#define FMT_DATE_HOUR 0x1E /* Hours with no leading 0 */
|
|
#define FMT_DATE_HOUR_0 0x1F /* Hours with leading 0 */
|
|
#define FMT_DATE_HOUR_12 0x20 /* Hours with no leading 0, 12 hour clock */
|
|
#define FMT_DATE_HOUR_12_0 0x21 /* Hours with leading 0, 12 hour clock */
|
|
#define FMT_DATE_TIME_UNK2 0x23 /* same as FMT_DATE_HOUR_0, for "short time" format */
|
|
/* FIXME: probably missing some here */
|
|
#define FMT_DATE_AMPM_SYS1 0x2E /* AM/PM as defined by system settings */
|
|
#define FMT_DATE_AMPM_UPPER 0x2F /* Upper-case AM or PM */
|
|
#define FMT_DATE_A_UPPER 0x30 /* Upper-case A or P */
|
|
#define FMT_DATE_AMPM_SYS2 0x31 /* AM/PM as defined by system settings */
|
|
#define FMT_DATE_AMPM_LOWER 0x32 /* Lower-case AM or PM */
|
|
#define FMT_DATE_A_LOWER 0x33 /* Lower-case A or P */
|
|
#define FMT_NUM_COPY_ZERO 0x34 /* Copy 1 digit or 0 if no digit */
|
|
#define FMT_NUM_COPY_SKIP 0x35 /* Copy 1 digit or skip if no digit */
|
|
#define FMT_NUM_DECIMAL 0x36 /* Decimal separator */
|
|
#define FMT_NUM_EXP_POS_U 0x37 /* Scientific notation, uppercase, + sign */
|
|
#define FMT_NUM_EXP_NEG_U 0x38 /* Scientific notation, uppercase, - sign */
|
|
#define FMT_NUM_EXP_POS_L 0x39 /* Scientific notation, lowercase, + sign */
|
|
#define FMT_NUM_EXP_NEG_L 0x3A /* Scientific notation, lowercase, - sign */
|
|
#define FMT_NUM_CURRENCY 0x3B /* Currency symbol */
|
|
#define FMT_NUM_TRUE_FALSE 0x3D /* Convert to "True" or "False" */
|
|
#define FMT_NUM_YES_NO 0x3E /* Convert to "Yes" or "No" */
|
|
#define FMT_NUM_ON_OFF 0x3F /* Convert to "On" or "Off" */
|
|
#define FMT_STR_COPY_SPACE 0x40 /* Copy len chars with space if no char */
|
|
#define FMT_STR_COPY_SKIP 0x41 /* Copy len chars or skip if no char */
|
|
/* Wine additions */
|
|
#define FMT_WINE_HOURS_12 0x81 /* Hours using 12 hour clock */
|
|
|
|
/* Named Formats and their tokenised values */
|
|
static const WCHAR szGeneralDate[] = { 'G','e','n','e','r','a','l',' ','D','a','t','e','\0' };
|
|
static const BYTE fmtGeneralDate[0x0a] =
|
|
{
|
|
0x0a,FMT_TYPE_DATE,sizeof(FMT_SHORT_HEADER),
|
|
0x0,0x0,0x0,0x0,0x0,
|
|
FMT_DATE_GENERAL,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szShortDate[] = { 'S','h','o','r','t',' ','D','a','t','e','\0' };
|
|
static const BYTE fmtShortDate[0x0a] =
|
|
{
|
|
0x0a,FMT_TYPE_DATE,sizeof(FMT_SHORT_HEADER),
|
|
0x0,0x0,0x0,0x0,0x0,
|
|
FMT_DATE_SHORT,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szMediumDate[] = { 'M','e','d','i','u','m',' ','D','a','t','e','\0' };
|
|
static const BYTE fmtMediumDate[0x0a] =
|
|
{
|
|
0x0a,FMT_TYPE_DATE,sizeof(FMT_SHORT_HEADER),
|
|
0x0,0x0,0x0,0x0,0x0,
|
|
FMT_DATE_MEDIUM,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szLongDate[] = { 'L','o','n','g',' ','D','a','t','e','\0' };
|
|
static const BYTE fmtLongDate[0x0a] =
|
|
{
|
|
0x0a,FMT_TYPE_DATE,sizeof(FMT_SHORT_HEADER),
|
|
0x0,0x0,0x0,0x0,0x0,
|
|
FMT_DATE_LONG,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szShortTime[] = { 'S','h','o','r','t',' ','T','i','m','e','\0' };
|
|
static const BYTE fmtShortTime[0x0c] =
|
|
{
|
|
0x0c,FMT_TYPE_DATE,sizeof(FMT_SHORT_HEADER),
|
|
0x0,0x0,0x0,0x0,0x0,
|
|
FMT_DATE_TIME_UNK2,FMT_DATE_TIME_SEP,FMT_DATE_MIN_0,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szMediumTime[] = { 'M','e','d','i','u','m',' ','T','i','m','e','\0' };
|
|
static const BYTE fmtMediumTime[0x11] =
|
|
{
|
|
0x11,FMT_TYPE_DATE,sizeof(FMT_SHORT_HEADER),
|
|
0x0,0x0,0x0,0x0,0x0,
|
|
FMT_DATE_HOUR_12_0,FMT_DATE_TIME_SEP,FMT_DATE_MIN_0,
|
|
FMT_GEN_INLINE,0x01,' ','\0',FMT_DATE_AMPM_SYS1,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szLongTime[] = { 'L','o','n','g',' ','T','i','m','e','\0' };
|
|
static const BYTE fmtLongTime[0x0d] =
|
|
{
|
|
0x0a,FMT_TYPE_DATE,sizeof(FMT_SHORT_HEADER),
|
|
0x0,0x0,0x0,0x0,0x0,
|
|
FMT_DATE_TIME_SYS,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szTrueFalse[] = { 'T','r','u','e','/','F','a','l','s','e','\0' };
|
|
static const BYTE fmtTrueFalse[0x0d] =
|
|
{
|
|
0x0d,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x0,0x0,0x0,
|
|
FMT_FLAG_BOOL,0x0,0x0,0x0,0x0,
|
|
FMT_NUM_TRUE_FALSE,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szYesNo[] = { 'Y','e','s','/','N','o','\0' };
|
|
static const BYTE fmtYesNo[0x0d] =
|
|
{
|
|
0x0d,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x0,0x0,0x0,
|
|
FMT_FLAG_BOOL,0x0,0x0,0x0,0x0,
|
|
FMT_NUM_YES_NO,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szOnOff[] = { 'O','n','/','O','f','f','\0' };
|
|
static const BYTE fmtOnOff[0x0d] =
|
|
{
|
|
0x0d,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x0,0x0,0x0,
|
|
FMT_FLAG_BOOL,0x0,0x0,0x0,0x0,
|
|
FMT_NUM_ON_OFF,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szGeneralNumber[] = { 'G','e','n','e','r','a','l',' ','N','u','m','b','e','r','\0' };
|
|
static const BYTE fmtGeneralNumber[sizeof(FMT_HEADER)] =
|
|
{
|
|
sizeof(FMT_HEADER),FMT_TYPE_GENERAL,sizeof(FMT_HEADER),0x0,0x0,0x0
|
|
};
|
|
|
|
static const WCHAR szCurrency[] = { 'C','u','r','r','e','n','c','y','\0' };
|
|
static const BYTE fmtCurrency[0x26] =
|
|
{
|
|
0x26,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x12,0x0,0x0,
|
|
/* Positive numbers */
|
|
FMT_FLAG_THOUSANDS,0xcc,0x0,0x1,0x2,
|
|
FMT_NUM_CURRENCY,FMT_NUM_COPY_ZERO,0x1,FMT_NUM_DECIMAL,FMT_NUM_COPY_ZERO,0x2,
|
|
FMT_GEN_END,
|
|
/* Negative numbers */
|
|
FMT_FLAG_THOUSANDS,0xcc,0x0,0x1,0x2,
|
|
FMT_GEN_INLINE,0x1,'(','\0',FMT_NUM_CURRENCY,FMT_NUM_COPY_ZERO,0x1,
|
|
FMT_NUM_DECIMAL,FMT_NUM_COPY_ZERO,0x2,FMT_GEN_INLINE,0x1,')','\0',
|
|
FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szFixed[] = { 'F','i','x','e','d','\0' };
|
|
static const BYTE fmtFixed[0x11] =
|
|
{
|
|
0x11,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x0,0x0,0x0,
|
|
0x0,0x0,0x0,0x1,0x2,
|
|
FMT_NUM_COPY_ZERO,0x1,FMT_NUM_DECIMAL,FMT_NUM_COPY_ZERO,0x2,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szStandard[] = { 'S','t','a','n','d','a','r','d','\0' };
|
|
static const BYTE fmtStandard[0x11] =
|
|
{
|
|
0x11,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x0,0x0,0x0,
|
|
FMT_FLAG_THOUSANDS,0x0,0x0,0x1,0x2,
|
|
FMT_NUM_COPY_ZERO,0x1,FMT_NUM_DECIMAL,FMT_NUM_COPY_ZERO,0x2,FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szPercent[] = { 'P','e','r','c','e','n','t','\0' };
|
|
static const BYTE fmtPercent[0x15] =
|
|
{
|
|
0x15,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x0,0x0,0x0,
|
|
FMT_FLAG_PERCENT,0x1,0x0,0x1,0x2,
|
|
FMT_NUM_COPY_ZERO,0x1,FMT_NUM_DECIMAL,FMT_NUM_COPY_ZERO,0x2,
|
|
FMT_GEN_INLINE,0x1,'%','\0',FMT_GEN_END
|
|
};
|
|
|
|
static const WCHAR szScientific[] = { 'S','c','i','e','n','t','i','f','i','c','\0' };
|
|
static const BYTE fmtScientific[0x13] =
|
|
{
|
|
0x13,FMT_TYPE_NUMBER,sizeof(FMT_HEADER),0x0,0x0,0x0,
|
|
FMT_FLAG_EXPONENT,0x0,0x0,0x1,0x2,
|
|
FMT_NUM_COPY_ZERO,0x1,FMT_NUM_DECIMAL,FMT_NUM_COPY_ZERO,0x2,FMT_NUM_EXP_POS_U,0x2,FMT_GEN_END
|
|
};
|
|
|
|
typedef struct tagNAMED_FORMAT
|
|
{
|
|
LPCWSTR name;
|
|
const BYTE* format;
|
|
} NAMED_FORMAT;
|
|
|
|
/* Format name to tokenised format. Must be kept sorted by name */
|
|
static const NAMED_FORMAT VARIANT_NamedFormats[] =
|
|
{
|
|
{ szCurrency, fmtCurrency },
|
|
{ szFixed, fmtFixed },
|
|
{ szGeneralDate, fmtGeneralDate },
|
|
{ szGeneralNumber, fmtGeneralNumber },
|
|
{ szLongDate, fmtLongDate },
|
|
{ szLongTime, fmtLongTime },
|
|
{ szMediumDate, fmtMediumDate },
|
|
{ szMediumTime, fmtMediumTime },
|
|
{ szOnOff, fmtOnOff },
|
|
{ szPercent, fmtPercent },
|
|
{ szScientific, fmtScientific },
|
|
{ szShortDate, fmtShortDate },
|
|
{ szShortTime, fmtShortTime },
|
|
{ szStandard, fmtStandard },
|
|
{ szTrueFalse, fmtTrueFalse },
|
|
{ szYesNo, fmtYesNo }
|
|
};
|
|
typedef const NAMED_FORMAT *LPCNAMED_FORMAT;
|
|
|
|
static int FormatCompareFn(const void *l, const void *r)
|
|
{
|
|
return strcmpiW(((LPCNAMED_FORMAT)l)->name, ((LPCNAMED_FORMAT)r)->name);
|
|
}
|
|
|
|
static inline const BYTE *VARIANT_GetNamedFormat(LPCWSTR lpszFormat)
|
|
{
|
|
NAMED_FORMAT key;
|
|
LPCNAMED_FORMAT fmt;
|
|
|
|
key.name = lpszFormat;
|
|
fmt = bsearch(&key, VARIANT_NamedFormats,
|
|
sizeof(VARIANT_NamedFormats)/sizeof(NAMED_FORMAT),
|
|
sizeof(NAMED_FORMAT), FormatCompareFn);
|
|
return fmt ? fmt->format : NULL;
|
|
}
|
|
|
|
/* Return an error if the token for the value will not fit in the destination */
|
|
#define NEED_SPACE(x) if (cbTok < (int)(x)) return TYPE_E_BUFFERTOOSMALL; cbTok -= (x)
|
|
|
|
/* Non-zero if the format is unknown or a given type */
|
|
#define COULD_BE(typ) ((!fmt_number && header->type==FMT_TYPE_UNKNOWN)||header->type==typ)
|
|
|
|
/* State during tokenising */
|
|
#define FMT_STATE_OPEN_COPY 0x1 /* Last token written was a copy */
|
|
#define FMT_STATE_WROTE_DECIMAL 0x2 /* Already wrote a decimal separator */
|
|
#define FMT_STATE_SEEN_HOURS 0x4 /* See the hh specifier */
|
|
#define FMT_STATE_WROTE_MINUTES 0x8 /* Wrote minutes */
|
|
|
|
/**********************************************************************
|
|
* VarTokenizeFormatString [OLEAUT32.140]
|
|
*
|
|
* Convert a format string into tokenised form.
|
|
*
|
|
* PARAMS
|
|
* lpszFormat [I] Format string to tokenise
|
|
* rgbTok [O] Destination for tokenised format
|
|
* cbTok [I] Size of rgbTok in bytes
|
|
* nFirstDay [I] First day of the week (1-7, or 0 for current system default)
|
|
* nFirstWeek [I] How to treat the first week (see notes)
|
|
* lcid [I] Locale Id of the format string
|
|
* pcbActual [O] If non-NULL, filled with the first token generated
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK. rgbTok contains the tokenised format.
|
|
* Failure: E_INVALIDARG, if any argument is invalid.
|
|
* TYPE_E_BUFFERTOOSMALL, if rgbTok is not large enough.
|
|
*
|
|
* NOTES
|
|
* Valid values for the nFirstWeek parameter are:
|
|
*| Value Meaning
|
|
*| ----- -------
|
|
*| 0 Use the current system default
|
|
*| 1 The first week is that containing Jan 1
|
|
*| 2 Four or more days of the first week are in the current year
|
|
*| 3 The first week is 7 days long
|
|
* See Variant-Formats(), VarFormatFromTokens().
|
|
*/
|
|
HRESULT WINAPI VarTokenizeFormatString(LPOLESTR lpszFormat, LPBYTE rgbTok,
|
|
int cbTok, int nFirstDay, int nFirstWeek,
|
|
LCID lcid, int *pcbActual)
|
|
{
|
|
/* Note: none of these strings should be NUL terminated */
|
|
static const WCHAR szTTTTT[] = { 't','t','t','t','t' };
|
|
static const WCHAR szAMPM[] = { 'A','M','P','M' };
|
|
static const WCHAR szampm[] = { 'a','m','p','m' };
|
|
static const WCHAR szAMSlashPM[] = { 'A','M','/','P','M' };
|
|
static const WCHAR szamSlashpm[] = { 'a','m','/','p','m' };
|
|
const BYTE *namedFmt;
|
|
FMT_HEADER *header = (FMT_HEADER*)rgbTok;
|
|
FMT_STRING_HEADER *str_header = (FMT_STRING_HEADER*)(rgbTok + sizeof(FMT_HEADER));
|
|
FMT_NUMBER_HEADER *num_header = (FMT_NUMBER_HEADER*)str_header;
|
|
BYTE* pOut = rgbTok + sizeof(FMT_HEADER) + sizeof(FMT_STRING_HEADER);
|
|
BYTE* pLastHours = NULL;
|
|
BYTE fmt_number = 0;
|
|
DWORD fmt_state = 0;
|
|
LPCWSTR pFormat = lpszFormat;
|
|
|
|
TRACE("(%s,%p,%d,%d,%d,0x%08x,%p)\n", debugstr_w(lpszFormat), rgbTok, cbTok,
|
|
nFirstDay, nFirstWeek, lcid, pcbActual);
|
|
|
|
if (!rgbTok ||
|
|
nFirstDay < 0 || nFirstDay > 7 || nFirstWeek < 0 || nFirstWeek > 3)
|
|
return E_INVALIDARG;
|
|
|
|
if (!lpszFormat || !*lpszFormat)
|
|
{
|
|
/* An empty string means 'general format' */
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*rgbTok = FMT_TO_STRING;
|
|
if (pcbActual)
|
|
*pcbActual = FMT_TO_STRING;
|
|
return S_OK;
|
|
}
|
|
|
|
if (cbTok > 255)
|
|
cbTok = 255; /* Ensure we error instead of wrapping */
|
|
|
|
/* Named formats */
|
|
namedFmt = VARIANT_GetNamedFormat(lpszFormat);
|
|
if (namedFmt)
|
|
{
|
|
NEED_SPACE(namedFmt[0]);
|
|
memcpy(rgbTok, namedFmt, namedFmt[0]);
|
|
TRACE("Using pre-tokenised named format %s\n", debugstr_w(lpszFormat));
|
|
/* FIXME: pcbActual */
|
|
return S_OK;
|
|
}
|
|
|
|
/* Insert header */
|
|
NEED_SPACE(sizeof(FMT_HEADER) + sizeof(FMT_STRING_HEADER));
|
|
memset(header, 0, sizeof(FMT_HEADER));
|
|
memset(str_header, 0, sizeof(FMT_STRING_HEADER));
|
|
|
|
header->starts[fmt_number] = sizeof(FMT_HEADER);
|
|
|
|
while (*pFormat)
|
|
{
|
|
/* --------------
|
|
* General tokens
|
|
* --------------
|
|
*/
|
|
if (*pFormat == ';')
|
|
{
|
|
while (*pFormat == ';')
|
|
{
|
|
TRACE(";\n");
|
|
if (++fmt_number > 3)
|
|
return E_INVALIDARG; /* too many formats */
|
|
pFormat++;
|
|
}
|
|
if (*pFormat)
|
|
{
|
|
TRACE("New header\n");
|
|
NEED_SPACE(sizeof(BYTE) + sizeof(FMT_STRING_HEADER));
|
|
*pOut++ = FMT_GEN_END;
|
|
|
|
header->starts[fmt_number] = pOut - rgbTok;
|
|
str_header = (FMT_STRING_HEADER*)pOut;
|
|
num_header = (FMT_NUMBER_HEADER*)pOut;
|
|
memset(str_header, 0, sizeof(FMT_STRING_HEADER));
|
|
pOut += sizeof(FMT_STRING_HEADER);
|
|
fmt_state = 0;
|
|
pLastHours = NULL;
|
|
}
|
|
}
|
|
else if (*pFormat == '\\')
|
|
{
|
|
/* Escaped character */
|
|
if (pFormat[1])
|
|
{
|
|
NEED_SPACE(3 * sizeof(BYTE));
|
|
pFormat++;
|
|
*pOut++ = FMT_GEN_COPY;
|
|
*pOut++ = pFormat - lpszFormat;
|
|
*pOut++ = 0x1;
|
|
fmt_state |= FMT_STATE_OPEN_COPY;
|
|
TRACE("'\\'\n");
|
|
}
|
|
else
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
pFormat++;
|
|
}
|
|
else if (*pFormat == '"')
|
|
{
|
|
/* Escaped string
|
|
* Note: Native encodes "" as a copy of length zero. That's just dumb, so
|
|
* here we avoid encoding anything in this case.
|
|
*/
|
|
if (!pFormat[1])
|
|
pFormat++;
|
|
else if (pFormat[1] == '"')
|
|
{
|
|
pFormat += 2;
|
|
}
|
|
else
|
|
{
|
|
LPCWSTR start = ++pFormat;
|
|
while (*pFormat && *pFormat != '"')
|
|
pFormat++;
|
|
NEED_SPACE(3 * sizeof(BYTE));
|
|
*pOut++ = FMT_GEN_COPY;
|
|
*pOut++ = start - lpszFormat;
|
|
*pOut++ = pFormat - start;
|
|
if (*pFormat == '"')
|
|
pFormat++;
|
|
TRACE("Quoted string pos %d, len %d\n", pOut[-2], pOut[-1]);
|
|
}
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
/* -------------
|
|
* Number tokens
|
|
* -------------
|
|
*/
|
|
else if (*pFormat == '0' && COULD_BE(FMT_TYPE_NUMBER))
|
|
{
|
|
/* Number formats: Digit from number or '0' if no digits
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_NUMBER;
|
|
NEED_SPACE(2 * sizeof(BYTE));
|
|
*pOut++ = FMT_NUM_COPY_ZERO;
|
|
*pOut = 0x0;
|
|
while (*pFormat == '0')
|
|
{
|
|
*pOut = *pOut + 1;
|
|
pFormat++;
|
|
}
|
|
if (fmt_state & FMT_STATE_WROTE_DECIMAL)
|
|
num_header->fractional += *pOut;
|
|
else
|
|
num_header->whole += *pOut;
|
|
TRACE("%d 0's\n", *pOut);
|
|
pOut++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if (*pFormat == '#' && COULD_BE(FMT_TYPE_NUMBER))
|
|
{
|
|
/* Number formats: Digit from number or blank if no digits
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_NUMBER;
|
|
NEED_SPACE(2 * sizeof(BYTE));
|
|
*pOut++ = FMT_NUM_COPY_SKIP;
|
|
*pOut = 0x0;
|
|
while (*pFormat == '#')
|
|
{
|
|
*pOut = *pOut + 1;
|
|
pFormat++;
|
|
}
|
|
if (fmt_state & FMT_STATE_WROTE_DECIMAL)
|
|
num_header->fractional += *pOut;
|
|
else
|
|
num_header->whole += *pOut;
|
|
TRACE("%d #'s\n", *pOut);
|
|
pOut++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if (*pFormat == '.' && COULD_BE(FMT_TYPE_NUMBER) &&
|
|
!(fmt_state & FMT_STATE_WROTE_DECIMAL))
|
|
{
|
|
/* Number formats: Decimal separator when 1st seen, literal thereafter
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_NUMBER;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*pOut++ = FMT_NUM_DECIMAL;
|
|
fmt_state |= FMT_STATE_WROTE_DECIMAL;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
pFormat++;
|
|
TRACE("decimal sep\n");
|
|
}
|
|
else if ((*pFormat == 'e' || *pFormat == 'E') && (pFormat[1] == '-' ||
|
|
pFormat[1] == '+') && header->type == FMT_TYPE_NUMBER)
|
|
{
|
|
/* Number formats: Exponent specifier
|
|
* Other formats: Literal
|
|
*/
|
|
num_header->flags |= FMT_FLAG_EXPONENT;
|
|
NEED_SPACE(2 * sizeof(BYTE));
|
|
if (*pFormat == 'e') {
|
|
if (pFormat[1] == '+')
|
|
*pOut = FMT_NUM_EXP_POS_L;
|
|
else
|
|
*pOut = FMT_NUM_EXP_NEG_L;
|
|
} else {
|
|
if (pFormat[1] == '+')
|
|
*pOut = FMT_NUM_EXP_POS_U;
|
|
else
|
|
*pOut = FMT_NUM_EXP_NEG_U;
|
|
}
|
|
pFormat += 2;
|
|
*++pOut = 0x0;
|
|
while (*pFormat == '0')
|
|
{
|
|
*pOut = *pOut + 1;
|
|
pFormat++;
|
|
}
|
|
pOut++;
|
|
TRACE("exponent\n");
|
|
}
|
|
/* FIXME: %% => Divide by 1000 */
|
|
else if (*pFormat == ',' && header->type == FMT_TYPE_NUMBER)
|
|
{
|
|
/* Number formats: Use the thousands separator
|
|
* Other formats: Literal
|
|
*/
|
|
num_header->flags |= FMT_FLAG_THOUSANDS;
|
|
pFormat++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
TRACE("thousands sep\n");
|
|
}
|
|
/* -----------
|
|
* Date tokens
|
|
* -----------
|
|
*/
|
|
else if (*pFormat == '/' && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Date separator
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*pOut++ = FMT_DATE_DATE_SEP;
|
|
pFormat++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
TRACE("date sep\n");
|
|
}
|
|
else if (*pFormat == ':' && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Time separator
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*pOut++ = FMT_DATE_TIME_SEP;
|
|
pFormat++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
TRACE("time sep\n");
|
|
}
|
|
else if ((*pFormat == 'a' || *pFormat == 'A') &&
|
|
!strncmpiW(pFormat, szAMPM, sizeof(szAMPM)/sizeof(WCHAR)))
|
|
{
|
|
/* Date formats: System AM/PM designation
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat += sizeof(szAMPM)/sizeof(WCHAR);
|
|
if (!strncmpW(pFormat, szampm, sizeof(szampm)/sizeof(WCHAR)))
|
|
*pOut++ = FMT_DATE_AMPM_SYS2;
|
|
else
|
|
*pOut++ = FMT_DATE_AMPM_SYS1;
|
|
if (pLastHours)
|
|
*pLastHours = *pLastHours + 2;
|
|
TRACE("ampm\n");
|
|
}
|
|
else if (*pFormat == 'a' && pFormat[1] == '/' &&
|
|
(pFormat[2] == 'p' || pFormat[2] == 'P'))
|
|
{
|
|
/* Date formats: lowercase a or p designation
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat += 3;
|
|
*pOut++ = FMT_DATE_A_LOWER;
|
|
if (pLastHours)
|
|
*pLastHours = *pLastHours + 2;
|
|
TRACE("a/p\n");
|
|
}
|
|
else if (*pFormat == 'A' && pFormat[1] == '/' &&
|
|
(pFormat[2] == 'p' || pFormat[2] == 'P'))
|
|
{
|
|
/* Date formats: Uppercase a or p designation
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat += 3;
|
|
*pOut++ = FMT_DATE_A_UPPER;
|
|
if (pLastHours)
|
|
*pLastHours = *pLastHours + 2;
|
|
TRACE("A/P\n");
|
|
}
|
|
else if (*pFormat == 'a' &&
|
|
!strncmpW(pFormat, szamSlashpm, sizeof(szamSlashpm)/sizeof(WCHAR)))
|
|
{
|
|
/* Date formats: lowercase AM or PM designation
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat += sizeof(szamSlashpm)/sizeof(WCHAR);
|
|
*pOut++ = FMT_DATE_AMPM_LOWER;
|
|
if (pLastHours)
|
|
*pLastHours = *pLastHours + 2;
|
|
TRACE("AM/PM\n");
|
|
}
|
|
else if (*pFormat == 'A' &&
|
|
!strncmpW(pFormat, szAMSlashPM, sizeof(szAMSlashPM)/sizeof(WCHAR)))
|
|
{
|
|
/* Date formats: Uppercase AM or PM designation
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat += sizeof(szAMSlashPM)/sizeof(WCHAR);
|
|
*pOut++ = FMT_DATE_AMPM_UPPER;
|
|
TRACE("AM/PM\n");
|
|
}
|
|
else if ((*pFormat == 'c' || *pFormat == 'C') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: General date format
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat += sizeof(szAMSlashPM)/sizeof(WCHAR);
|
|
*pOut++ = FMT_DATE_GENERAL;
|
|
TRACE("gen date\n");
|
|
}
|
|
else if ((*pFormat == 'd' || *pFormat == 'D') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Day specifier
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
int count = -1;
|
|
header->type = FMT_TYPE_DATE;
|
|
while ((*pFormat == 'd' || *pFormat == 'D') && count < 6)
|
|
{
|
|
pFormat++;
|
|
count++;
|
|
}
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*pOut++ = FMT_DATE_DAY + count;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
/* When we find the days token, reset the seen hours state so that
|
|
* 'mm' is again written as month when encountered.
|
|
*/
|
|
fmt_state &= ~FMT_STATE_SEEN_HOURS;
|
|
TRACE("%d d's\n", count + 1);
|
|
}
|
|
else if ((*pFormat == 'h' || *pFormat == 'H') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Hour specifier
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat++;
|
|
/* Record the position of the hours specifier - if we encounter
|
|
* an am/pm specifier we will change the hours from 24 to 12.
|
|
*/
|
|
pLastHours = pOut;
|
|
if (*pFormat == 'h' || *pFormat == 'H')
|
|
{
|
|
pFormat++;
|
|
*pOut++ = FMT_DATE_HOUR_0;
|
|
TRACE("hh\n");
|
|
}
|
|
else
|
|
{
|
|
*pOut++ = FMT_DATE_HOUR;
|
|
TRACE("h\n");
|
|
}
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
/* Note that now we have seen an hours token, the next occurrence of
|
|
* 'mm' indicates minutes, not months.
|
|
*/
|
|
fmt_state |= FMT_STATE_SEEN_HOURS;
|
|
}
|
|
else if ((*pFormat == 'm' || *pFormat == 'M') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Month specifier (or Minute specifier, after hour specifier)
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
int count = -1;
|
|
header->type = FMT_TYPE_DATE;
|
|
while ((*pFormat == 'm' || *pFormat == 'M') && count < 4)
|
|
{
|
|
pFormat++;
|
|
count++;
|
|
}
|
|
NEED_SPACE(sizeof(BYTE));
|
|
if (count <= 1 && fmt_state & FMT_STATE_SEEN_HOURS &&
|
|
!(fmt_state & FMT_STATE_WROTE_MINUTES))
|
|
{
|
|
/* We have seen an hours specifier and not yet written a minutes
|
|
* specifier. Write this as minutes and thereafter as months.
|
|
*/
|
|
*pOut++ = count == 1 ? FMT_DATE_MIN_0 : FMT_DATE_MIN;
|
|
fmt_state |= FMT_STATE_WROTE_MINUTES; /* Hereafter write months */
|
|
}
|
|
else
|
|
*pOut++ = FMT_DATE_MON + count; /* Months */
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
TRACE("%d m's\n", count + 1);
|
|
}
|
|
else if ((*pFormat == 'n' || *pFormat == 'N') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Minute specifier
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat++;
|
|
if (*pFormat == 'n' || *pFormat == 'N')
|
|
{
|
|
pFormat++;
|
|
*pOut++ = FMT_DATE_MIN_0;
|
|
TRACE("nn\n");
|
|
}
|
|
else
|
|
{
|
|
*pOut++ = FMT_DATE_MIN;
|
|
TRACE("n\n");
|
|
}
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if ((*pFormat == 'q' || *pFormat == 'Q') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Quarter specifier
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*pOut++ = FMT_DATE_QUARTER;
|
|
pFormat++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
TRACE("quarter\n");
|
|
}
|
|
else if ((*pFormat == 's' || *pFormat == 'S') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Second specifier
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
NEED_SPACE(sizeof(BYTE));
|
|
pFormat++;
|
|
if (*pFormat == 's' || *pFormat == 'S')
|
|
{
|
|
pFormat++;
|
|
*pOut++ = FMT_DATE_SEC_0;
|
|
TRACE("ss\n");
|
|
}
|
|
else
|
|
{
|
|
*pOut++ = FMT_DATE_SEC;
|
|
TRACE("s\n");
|
|
}
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if ((*pFormat == 't' || *pFormat == 'T') &&
|
|
!strncmpiW(pFormat, szTTTTT, sizeof(szTTTTT)/sizeof(WCHAR)))
|
|
{
|
|
/* Date formats: System time specifier
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
pFormat += sizeof(szTTTTT)/sizeof(WCHAR);
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*pOut++ = FMT_DATE_TIME_SYS;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if ((*pFormat == 'w' || *pFormat == 'W') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Week of the year/Day of the week
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_DATE;
|
|
pFormat++;
|
|
if (*pFormat == 'w' || *pFormat == 'W')
|
|
{
|
|
NEED_SPACE(3 * sizeof(BYTE));
|
|
pFormat++;
|
|
*pOut++ = FMT_DATE_WEEK_YEAR;
|
|
*pOut++ = nFirstDay;
|
|
*pOut++ = nFirstWeek;
|
|
TRACE("ww\n");
|
|
}
|
|
else
|
|
{
|
|
NEED_SPACE(2 * sizeof(BYTE));
|
|
*pOut++ = FMT_DATE_DAY_WEEK;
|
|
*pOut++ = nFirstDay;
|
|
TRACE("w\n");
|
|
}
|
|
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if ((*pFormat == 'y' || *pFormat == 'Y') && COULD_BE(FMT_TYPE_DATE))
|
|
{
|
|
/* Date formats: Day of year/Year specifier
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
int count = -1;
|
|
header->type = FMT_TYPE_DATE;
|
|
while ((*pFormat == 'y' || *pFormat == 'Y') && count < 4)
|
|
{
|
|
pFormat++;
|
|
count++;
|
|
}
|
|
if (count == 2)
|
|
{
|
|
count--; /* 'yyy' has no meaning, despite what MSDN says */
|
|
pFormat--;
|
|
}
|
|
NEED_SPACE(sizeof(BYTE));
|
|
*pOut++ = FMT_DATE_YEAR_DOY + count;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
TRACE("%d y's\n", count + 1);
|
|
}
|
|
/* -------------
|
|
* String tokens
|
|
* -------------
|
|
*/
|
|
else if (*pFormat == '@' && COULD_BE(FMT_TYPE_STRING))
|
|
{
|
|
/* String formats: Character from string or space if no char
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_STRING;
|
|
NEED_SPACE(2 * sizeof(BYTE));
|
|
*pOut++ = FMT_STR_COPY_SPACE;
|
|
*pOut = 0x0;
|
|
while (*pFormat == '@')
|
|
{
|
|
*pOut = *pOut + 1;
|
|
str_header->copy_chars++;
|
|
pFormat++;
|
|
}
|
|
TRACE("%d @'s\n", *pOut);
|
|
pOut++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if (*pFormat == '&' && COULD_BE(FMT_TYPE_STRING))
|
|
{
|
|
/* String formats: Character from string or skip if no char
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_STRING;
|
|
NEED_SPACE(2 * sizeof(BYTE));
|
|
*pOut++ = FMT_STR_COPY_SKIP;
|
|
*pOut = 0x0;
|
|
while (*pFormat == '&')
|
|
{
|
|
*pOut = *pOut + 1;
|
|
str_header->copy_chars++;
|
|
pFormat++;
|
|
}
|
|
TRACE("%d &'s\n", *pOut);
|
|
pOut++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if ((*pFormat == '<' || *pFormat == '>') && COULD_BE(FMT_TYPE_STRING))
|
|
{
|
|
/* String formats: Use upper/lower case
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_STRING;
|
|
if (*pFormat == '<')
|
|
str_header->flags |= FMT_FLAG_LT;
|
|
else
|
|
str_header->flags |= FMT_FLAG_GT;
|
|
TRACE("to %s case\n", *pFormat == '<' ? "lower" : "upper");
|
|
pFormat++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
}
|
|
else if (*pFormat == '!' && COULD_BE(FMT_TYPE_STRING))
|
|
{
|
|
/* String formats: Copy right to left
|
|
* Other formats: Literal
|
|
* Types the format if found
|
|
*/
|
|
header->type = FMT_TYPE_STRING;
|
|
str_header->flags |= FMT_FLAG_RTL;
|
|
pFormat++;
|
|
fmt_state &= ~FMT_STATE_OPEN_COPY;
|
|
TRACE("copy right-to-left\n");
|
|
}
|
|
/* --------
|
|
* Literals
|
|
* --------
|
|
*/
|
|
/* FIXME: [ seems to be ignored */
|
|
else
|
|
{
|
|
if (*pFormat == '%' && header->type == FMT_TYPE_NUMBER)
|
|
{
|
|
/* Number formats: Percentage indicator, also a literal
|
|
* Other formats: Literal
|
|
* Doesn't type the format
|
|
*/
|
|
num_header->flags |= FMT_FLAG_PERCENT;
|
|
}
|
|
|
|
if (fmt_state & FMT_STATE_OPEN_COPY)
|
|
{
|
|
pOut[-1] = pOut[-1] + 1; /* Increase the length of the open copy */
|
|
TRACE("extend copy (char '%c'), length now %d\n", *pFormat, pOut[-1]);
|
|
}
|
|
else
|
|
{
|
|
/* Create a new open copy */
|
|
TRACE("New copy (char '%c')\n", *pFormat);
|
|
NEED_SPACE(3 * sizeof(BYTE));
|
|
*pOut++ = FMT_GEN_COPY;
|
|
*pOut++ = pFormat - lpszFormat;
|
|
*pOut++ = 0x1;
|
|
fmt_state |= FMT_STATE_OPEN_COPY;
|
|
}
|
|
pFormat++;
|
|
}
|
|
}
|
|
|
|
*pOut++ = FMT_GEN_END;
|
|
|
|
header->size = pOut - rgbTok;
|
|
if (pcbActual)
|
|
*pcbActual = header->size;
|
|
|
|
return S_OK;
|
|
}
|
|
|
|
/* Number formatting state flags */
|
|
#define NUM_WROTE_DEC 0x01 /* Written the decimal separator */
|
|
#define NUM_WRITE_ON 0x02 /* Started to write the number */
|
|
#define NUM_WROTE_SIGN 0x04 /* Written the negative sign */
|
|
|
|
/* Format a variant using a number format */
|
|
static HRESULT VARIANT_FormatNumber(LPVARIANT pVarIn, LPOLESTR lpszFormat,
|
|
LPBYTE rgbTok, ULONG dwFlags,
|
|
BSTR *pbstrOut, LCID lcid)
|
|
{
|
|
BYTE rgbDig[256], *prgbDig;
|
|
NUMPARSE np;
|
|
int have_int, need_int = 0, have_frac, need_frac, exponent = 0, pad = 0;
|
|
WCHAR buff[256], *pBuff = buff;
|
|
WCHAR thousandSeparator[32];
|
|
VARIANT vString, vBool;
|
|
DWORD dwState = 0;
|
|
FMT_HEADER *header = (FMT_HEADER*)rgbTok;
|
|
FMT_NUMBER_HEADER *numHeader;
|
|
const BYTE* pToken = NULL;
|
|
HRESULT hRes = S_OK;
|
|
|
|
TRACE("(%p->(%s%s),%s,%p,0x%08x,%p,0x%08x)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), debugstr_w(lpszFormat), rgbTok, dwFlags, pbstrOut,
|
|
lcid);
|
|
|
|
V_VT(&vString) = VT_EMPTY;
|
|
V_VT(&vBool) = VT_BOOL;
|
|
|
|
if (V_TYPE(pVarIn) == VT_EMPTY || V_TYPE(pVarIn) == VT_NULL)
|
|
{
|
|
have_int = have_frac = 0;
|
|
numHeader = (FMT_NUMBER_HEADER*)(rgbTok + FmtGetNull(header));
|
|
V_BOOL(&vBool) = VARIANT_FALSE;
|
|
}
|
|
else
|
|
{
|
|
/* Get a number string from pVarIn, and parse it */
|
|
hRes = VariantChangeTypeEx(&vString, pVarIn, LCID_US, VARIANT_NOUSEROVERRIDE, VT_BSTR);
|
|
if (FAILED(hRes))
|
|
return hRes;
|
|
|
|
np.cDig = sizeof(rgbDig);
|
|
np.dwInFlags = NUMPRS_STD;
|
|
hRes = VarParseNumFromStr(V_BSTR(&vString), LCID_US, 0, &np, rgbDig);
|
|
if (FAILED(hRes))
|
|
return hRes;
|
|
|
|
have_int = np.cDig;
|
|
have_frac = 0;
|
|
exponent = np.nPwr10;
|
|
|
|
/* Figure out which format to use */
|
|
if (np.dwOutFlags & NUMPRS_NEG)
|
|
{
|
|
numHeader = (FMT_NUMBER_HEADER*)(rgbTok + FmtGetNegative(header));
|
|
V_BOOL(&vBool) = VARIANT_TRUE;
|
|
}
|
|
else if (have_int == 1 && !exponent && rgbDig[0] == 0)
|
|
{
|
|
numHeader = (FMT_NUMBER_HEADER*)(rgbTok + FmtGetZero(header));
|
|
V_BOOL(&vBool) = VARIANT_FALSE;
|
|
}
|
|
else
|
|
{
|
|
numHeader = (FMT_NUMBER_HEADER*)(rgbTok + FmtGetPositive(header));
|
|
V_BOOL(&vBool) = VARIANT_TRUE;
|
|
}
|
|
|
|
TRACE("num header: flags = 0x%x, mult=%d, div=%d, whole=%d, fract=%d\n",
|
|
numHeader->flags, numHeader->multiplier, numHeader->divisor,
|
|
numHeader->whole, numHeader->fractional);
|
|
|
|
need_int = numHeader->whole;
|
|
need_frac = numHeader->fractional;
|
|
|
|
if (numHeader->flags & FMT_FLAG_PERCENT &&
|
|
!(have_int == 1 && !exponent && rgbDig[0] == 0))
|
|
exponent += 2;
|
|
|
|
if (numHeader->flags & FMT_FLAG_EXPONENT)
|
|
{
|
|
/* Exponent format: length of the integral number part is fixed and
|
|
specified by the format. */
|
|
pad = need_int - have_int;
|
|
exponent -= pad;
|
|
if (pad < 0)
|
|
{
|
|
have_int = need_int;
|
|
have_frac -= pad;
|
|
pad = 0;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
/* Convert the exponent */
|
|
pad = max(exponent, -have_int);
|
|
exponent -= pad;
|
|
if (pad < 0)
|
|
{
|
|
have_int += pad;
|
|
have_frac = -pad;
|
|
pad = 0;
|
|
}
|
|
if(exponent < 0 && exponent > (-256 + have_int + have_frac))
|
|
{
|
|
/* Remove exponent notation */
|
|
memmove(rgbDig - exponent, rgbDig, have_int + have_frac);
|
|
ZeroMemory(rgbDig, -exponent);
|
|
have_frac -= exponent;
|
|
exponent = 0;
|
|
}
|
|
}
|
|
|
|
/* Rounding the number */
|
|
if (have_frac > need_frac)
|
|
{
|
|
prgbDig = &rgbDig[have_int + need_frac];
|
|
have_frac = need_frac;
|
|
if (*prgbDig >= 5)
|
|
{
|
|
while (prgbDig-- > rgbDig && *prgbDig == 9)
|
|
*prgbDig = 0;
|
|
if (prgbDig < rgbDig)
|
|
{
|
|
/* We reached the first digit and that was also a 9 */
|
|
rgbDig[0] = 1;
|
|
if (numHeader->flags & FMT_FLAG_EXPONENT)
|
|
exponent++;
|
|
else
|
|
{
|
|
rgbDig[have_int + need_frac] = 0;
|
|
if (exponent < 0)
|
|
exponent++;
|
|
else
|
|
have_int++;
|
|
}
|
|
}
|
|
else
|
|
(*prgbDig)++;
|
|
}
|
|
/* We converted trailing digits to zeroes => have_frac has changed */
|
|
while (have_frac > 0 && rgbDig[have_int + have_frac - 1] == 0)
|
|
have_frac--;
|
|
}
|
|
TRACE("have_int=%d,need_int=%d,have_frac=%d,need_frac=%d,pad=%d,exp=%d\n",
|
|
have_int, need_int, have_frac, need_frac, pad, exponent);
|
|
}
|
|
|
|
if (numHeader->flags & FMT_FLAG_THOUSANDS)
|
|
{
|
|
if (!GetLocaleInfoW(lcid, LOCALE_STHOUSAND, thousandSeparator,
|
|
sizeof(thousandSeparator)/sizeof(WCHAR)))
|
|
{
|
|
thousandSeparator[0] = ',';
|
|
thousandSeparator[1] = 0;
|
|
}
|
|
}
|
|
|
|
pToken = (const BYTE*)numHeader + sizeof(FMT_NUMBER_HEADER);
|
|
prgbDig = rgbDig;
|
|
|
|
while (SUCCEEDED(hRes) && *pToken != FMT_GEN_END)
|
|
{
|
|
WCHAR defaultChar = '?';
|
|
DWORD boolFlag, localeValue = 0;
|
|
BOOL shouldAdvance = TRUE;
|
|
|
|
if (pToken - rgbTok > header->size)
|
|
{
|
|
ERR("Ran off the end of the format!\n");
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatNumber_Exit;
|
|
}
|
|
|
|
switch (*pToken)
|
|
{
|
|
case FMT_GEN_COPY:
|
|
TRACE("copy %s\n", debugstr_wn(lpszFormat + pToken[1], pToken[2]));
|
|
memcpy(pBuff, lpszFormat + pToken[1], pToken[2] * sizeof(WCHAR));
|
|
pBuff += pToken[2];
|
|
pToken += 2;
|
|
break;
|
|
|
|
case FMT_GEN_INLINE:
|
|
pToken += 2;
|
|
TRACE("copy %s\n", debugstr_a((LPCSTR)pToken));
|
|
while (*pToken)
|
|
*pBuff++ = *pToken++;
|
|
break;
|
|
|
|
case FMT_NUM_YES_NO:
|
|
boolFlag = VAR_BOOLYESNO;
|
|
goto VARIANT_FormatNumber_Bool;
|
|
|
|
case FMT_NUM_ON_OFF:
|
|
boolFlag = VAR_BOOLONOFF;
|
|
goto VARIANT_FormatNumber_Bool;
|
|
|
|
case FMT_NUM_TRUE_FALSE:
|
|
boolFlag = VAR_LOCALBOOL;
|
|
|
|
VARIANT_FormatNumber_Bool:
|
|
{
|
|
BSTR boolStr = NULL;
|
|
|
|
if (pToken[1] != FMT_GEN_END)
|
|
{
|
|
ERR("Boolean token not at end of format!\n");
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatNumber_Exit;
|
|
}
|
|
hRes = VarBstrFromBool(V_BOOL(&vBool), lcid, boolFlag, &boolStr);
|
|
if (SUCCEEDED(hRes))
|
|
{
|
|
strcpyW(pBuff, boolStr);
|
|
SysFreeString(boolStr);
|
|
while (*pBuff)
|
|
pBuff++;
|
|
}
|
|
}
|
|
break;
|
|
|
|
case FMT_NUM_DECIMAL:
|
|
if ((np.dwOutFlags & NUMPRS_NEG) && !(dwState & NUM_WROTE_SIGN) && !header->starts[1])
|
|
{
|
|
/* last chance for a negative sign in the .# case */
|
|
TRACE("write negative sign\n");
|
|
localeValue = LOCALE_SNEGATIVESIGN;
|
|
defaultChar = '-';
|
|
dwState |= NUM_WROTE_SIGN;
|
|
shouldAdvance = FALSE;
|
|
break;
|
|
}
|
|
TRACE("write decimal separator\n");
|
|
localeValue = LOCALE_SDECIMAL;
|
|
defaultChar = '.';
|
|
dwState |= NUM_WROTE_DEC;
|
|
break;
|
|
|
|
case FMT_NUM_CURRENCY:
|
|
TRACE("write currency symbol\n");
|
|
localeValue = LOCALE_SCURRENCY;
|
|
defaultChar = '$';
|
|
break;
|
|
|
|
case FMT_NUM_EXP_POS_U:
|
|
case FMT_NUM_EXP_POS_L:
|
|
case FMT_NUM_EXP_NEG_U:
|
|
case FMT_NUM_EXP_NEG_L:
|
|
if (*pToken == FMT_NUM_EXP_POS_L || *pToken == FMT_NUM_EXP_NEG_L)
|
|
*pBuff++ = 'e';
|
|
else
|
|
*pBuff++ = 'E';
|
|
if (exponent < 0)
|
|
{
|
|
*pBuff++ = '-';
|
|
sprintfW(pBuff, szPercentZeroStar_d, pToken[1], -exponent);
|
|
}
|
|
else
|
|
{
|
|
if (*pToken == FMT_NUM_EXP_POS_L || *pToken == FMT_NUM_EXP_POS_U)
|
|
*pBuff++ = '+';
|
|
sprintfW(pBuff, szPercentZeroStar_d, pToken[1], exponent);
|
|
}
|
|
while (*pBuff)
|
|
pBuff++;
|
|
pToken++;
|
|
break;
|
|
|
|
case FMT_NUM_COPY_ZERO:
|
|
dwState |= NUM_WRITE_ON;
|
|
/* Fall through */
|
|
|
|
case FMT_NUM_COPY_SKIP:
|
|
TRACE("write %d %sdigits or %s\n", pToken[1],
|
|
dwState & NUM_WROTE_DEC ? "fractional " : "",
|
|
*pToken == FMT_NUM_COPY_ZERO ? "0" : "skip");
|
|
|
|
if (dwState & NUM_WROTE_DEC)
|
|
{
|
|
int count, i;
|
|
|
|
if (!(numHeader->flags & FMT_FLAG_EXPONENT) && exponent < 0)
|
|
{
|
|
/* Pad with 0 before writing the fractional digits */
|
|
pad = max(exponent, -pToken[1]);
|
|
exponent -= pad;
|
|
count = min(have_frac, pToken[1] + pad);
|
|
for (i = 0; i > pad; i--)
|
|
*pBuff++ = '0';
|
|
}
|
|
else
|
|
count = min(have_frac, pToken[1]);
|
|
|
|
pad += pToken[1] - count;
|
|
have_frac -= count;
|
|
while (count--)
|
|
*pBuff++ = '0' + *prgbDig++;
|
|
if (*pToken == FMT_NUM_COPY_ZERO)
|
|
{
|
|
for (; pad > 0; pad--)
|
|
*pBuff++ = '0'; /* Write zeros for missing trailing digits */
|
|
}
|
|
}
|
|
else
|
|
{
|
|
int count, count_max, position;
|
|
|
|
if ((np.dwOutFlags & NUMPRS_NEG) && !(dwState & NUM_WROTE_SIGN) && !header->starts[1])
|
|
{
|
|
TRACE("write negative sign\n");
|
|
localeValue = LOCALE_SNEGATIVESIGN;
|
|
defaultChar = '-';
|
|
dwState |= NUM_WROTE_SIGN;
|
|
shouldAdvance = FALSE;
|
|
break;
|
|
}
|
|
|
|
position = have_int + pad;
|
|
if (dwState & NUM_WRITE_ON)
|
|
position = max(position, need_int);
|
|
need_int -= pToken[1];
|
|
count_max = have_int + pad - need_int;
|
|
if (count_max < 0)
|
|
count_max = 0;
|
|
if (dwState & NUM_WRITE_ON)
|
|
{
|
|
count = pToken[1] - count_max;
|
|
TRACE("write %d leading zeros\n", count);
|
|
while (count-- > 0)
|
|
{
|
|
*pBuff++ = '0';
|
|
if ((numHeader->flags & FMT_FLAG_THOUSANDS) &&
|
|
position > 1 && (--position % 3) == 0)
|
|
{
|
|
int k;
|
|
TRACE("write thousand separator\n");
|
|
for (k = 0; thousandSeparator[k]; k++)
|
|
*pBuff++ = thousandSeparator[k];
|
|
}
|
|
}
|
|
}
|
|
if (*pToken == FMT_NUM_COPY_ZERO || have_int > 1 ||
|
|
(have_int > 0 && *prgbDig > 0))
|
|
{
|
|
count = min(count_max, have_int);
|
|
count_max -= count;
|
|
have_int -= count;
|
|
TRACE("write %d whole number digits\n", count);
|
|
while (count--)
|
|
{
|
|
dwState |= NUM_WRITE_ON;
|
|
*pBuff++ = '0' + *prgbDig++;
|
|
if ((numHeader->flags & FMT_FLAG_THOUSANDS) &&
|
|
position > 1 && (--position % 3) == 0)
|
|
{
|
|
int k;
|
|
TRACE("write thousand separator\n");
|
|
for (k = 0; thousandSeparator[k]; k++)
|
|
*pBuff++ = thousandSeparator[k];
|
|
}
|
|
}
|
|
}
|
|
count = min(count_max, pad);
|
|
pad -= count;
|
|
TRACE("write %d whole trailing 0's\n", count);
|
|
while (count--)
|
|
{
|
|
*pBuff++ = '0';
|
|
if ((numHeader->flags & FMT_FLAG_THOUSANDS) &&
|
|
position > 1 && (--position % 3) == 0)
|
|
{
|
|
int k;
|
|
TRACE("write thousand separator\n");
|
|
for (k = 0; thousandSeparator[k]; k++)
|
|
*pBuff++ = thousandSeparator[k];
|
|
}
|
|
}
|
|
}
|
|
pToken++;
|
|
break;
|
|
|
|
default:
|
|
ERR("Unknown token 0x%02x!\n", *pToken);
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatNumber_Exit;
|
|
}
|
|
if (localeValue)
|
|
{
|
|
if (GetLocaleInfoW(lcid, localeValue, pBuff,
|
|
sizeof(buff)/sizeof(WCHAR)-(pBuff-buff)))
|
|
{
|
|
TRACE("added %s\n", debugstr_w(pBuff));
|
|
while (*pBuff)
|
|
pBuff++;
|
|
}
|
|
else
|
|
{
|
|
TRACE("added %d '%c'\n", defaultChar, defaultChar);
|
|
*pBuff++ = defaultChar;
|
|
}
|
|
}
|
|
if (shouldAdvance)
|
|
pToken++;
|
|
}
|
|
|
|
VARIANT_FormatNumber_Exit:
|
|
VariantClear(&vString);
|
|
*pBuff = '\0';
|
|
TRACE("buff is %s\n", debugstr_w(buff));
|
|
if (SUCCEEDED(hRes))
|
|
{
|
|
*pbstrOut = SysAllocString(buff);
|
|
if (!*pbstrOut)
|
|
hRes = E_OUTOFMEMORY;
|
|
}
|
|
return hRes;
|
|
}
|
|
|
|
/* Format a variant using a date format */
|
|
static HRESULT VARIANT_FormatDate(LPVARIANT pVarIn, LPOLESTR lpszFormat,
|
|
LPBYTE rgbTok, ULONG dwFlags,
|
|
BSTR *pbstrOut, LCID lcid)
|
|
{
|
|
WCHAR buff[256], *pBuff = buff;
|
|
VARIANT vDate;
|
|
UDATE udate;
|
|
FMT_HEADER *header = (FMT_HEADER*)rgbTok;
|
|
FMT_DATE_HEADER *dateHeader;
|
|
const BYTE* pToken = NULL;
|
|
HRESULT hRes;
|
|
|
|
TRACE("(%p->(%s%s),%s,%p,0x%08x,%p,0x%08x)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), debugstr_w(lpszFormat), rgbTok, dwFlags, pbstrOut,
|
|
lcid);
|
|
|
|
V_VT(&vDate) = VT_EMPTY;
|
|
|
|
if (V_TYPE(pVarIn) == VT_EMPTY || V_TYPE(pVarIn) == VT_NULL)
|
|
{
|
|
dateHeader = (FMT_DATE_HEADER*)(rgbTok + FmtGetNegative(header));
|
|
V_DATE(&vDate) = 0;
|
|
}
|
|
else
|
|
{
|
|
USHORT usFlags = dwFlags & VARIANT_CALENDAR_HIJRI ? VAR_CALENDAR_HIJRI : 0;
|
|
|
|
hRes = VariantChangeTypeEx(&vDate, pVarIn, LCID_US, usFlags, VT_DATE);
|
|
if (FAILED(hRes))
|
|
return hRes;
|
|
dateHeader = (FMT_DATE_HEADER*)(rgbTok + FmtGetPositive(header));
|
|
}
|
|
|
|
hRes = VarUdateFromDate(V_DATE(&vDate), 0 /* FIXME: flags? */, &udate);
|
|
if (FAILED(hRes))
|
|
return hRes;
|
|
pToken = (const BYTE*)dateHeader + sizeof(FMT_DATE_HEADER);
|
|
|
|
while (*pToken != FMT_GEN_END)
|
|
{
|
|
DWORD dwVal = 0, localeValue = 0, dwFmt = 0;
|
|
LPCWSTR szPrintFmt = NULL;
|
|
WCHAR defaultChar = '?';
|
|
|
|
if (pToken - rgbTok > header->size)
|
|
{
|
|
ERR("Ran off the end of the format!\n");
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatDate_Exit;
|
|
}
|
|
|
|
switch (*pToken)
|
|
{
|
|
case FMT_GEN_COPY:
|
|
TRACE("copy %s\n", debugstr_wn(lpszFormat + pToken[1], pToken[2]));
|
|
memcpy(pBuff, lpszFormat + pToken[1], pToken[2] * sizeof(WCHAR));
|
|
pBuff += pToken[2];
|
|
pToken += 2;
|
|
break;
|
|
|
|
case FMT_GEN_INLINE:
|
|
pToken += 2;
|
|
TRACE("copy %s\n", debugstr_a((LPCSTR)pToken));
|
|
while (*pToken)
|
|
*pBuff++ = *pToken++;
|
|
break;
|
|
|
|
case FMT_DATE_TIME_SEP:
|
|
TRACE("time separator\n");
|
|
localeValue = LOCALE_STIME;
|
|
defaultChar = ':';
|
|
break;
|
|
|
|
case FMT_DATE_DATE_SEP:
|
|
TRACE("date separator\n");
|
|
localeValue = LOCALE_SDATE;
|
|
defaultChar = '/';
|
|
break;
|
|
|
|
case FMT_DATE_GENERAL:
|
|
{
|
|
BSTR date = NULL;
|
|
WCHAR *pDate;
|
|
hRes = VarBstrFromDate(V_DATE(&vDate), lcid, 0, &date);
|
|
if (FAILED(hRes))
|
|
goto VARIANT_FormatDate_Exit;
|
|
pDate = date;
|
|
while (*pDate)
|
|
*pBuff++ = *pDate++;
|
|
SysFreeString(date);
|
|
}
|
|
break;
|
|
|
|
case FMT_DATE_QUARTER:
|
|
if (udate.st.wMonth <= 3)
|
|
*pBuff++ = '1';
|
|
else if (udate.st.wMonth <= 6)
|
|
*pBuff++ = '2';
|
|
else if (udate.st.wMonth <= 9)
|
|
*pBuff++ = '3';
|
|
else
|
|
*pBuff++ = '4';
|
|
break;
|
|
|
|
case FMT_DATE_TIME_SYS:
|
|
{
|
|
/* FIXME: VARIANT_CALENDAR HIJRI should cause Hijri output */
|
|
BSTR date = NULL;
|
|
WCHAR *pDate;
|
|
hRes = VarBstrFromDate(V_DATE(&vDate), lcid, VAR_TIMEVALUEONLY, &date);
|
|
if (FAILED(hRes))
|
|
goto VARIANT_FormatDate_Exit;
|
|
pDate = date;
|
|
while (*pDate)
|
|
*pBuff++ = *pDate++;
|
|
SysFreeString(date);
|
|
}
|
|
break;
|
|
|
|
case FMT_DATE_DAY:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.st.wDay;
|
|
break;
|
|
|
|
case FMT_DATE_DAY_0:
|
|
szPrintFmt = szPercentZeroTwo_d;
|
|
dwVal = udate.st.wDay;
|
|
break;
|
|
|
|
case FMT_DATE_DAY_SHORT:
|
|
/* FIXME: VARIANT_CALENDAR HIJRI should cause Hijri output */
|
|
TRACE("short day\n");
|
|
localeValue = LOCALE_SABBREVDAYNAME1 + (udate.st.wDayOfWeek + 6)%7;
|
|
defaultChar = '?';
|
|
break;
|
|
|
|
case FMT_DATE_DAY_LONG:
|
|
/* FIXME: VARIANT_CALENDAR HIJRI should cause Hijri output */
|
|
TRACE("long day\n");
|
|
localeValue = LOCALE_SDAYNAME1 + (udate.st.wDayOfWeek + 6)%7;
|
|
defaultChar = '?';
|
|
break;
|
|
|
|
case FMT_DATE_SHORT:
|
|
/* FIXME: VARIANT_CALENDAR HIJRI should cause Hijri output */
|
|
dwFmt = LOCALE_SSHORTDATE;
|
|
break;
|
|
|
|
case FMT_DATE_LONG:
|
|
/* FIXME: VARIANT_CALENDAR HIJRI should cause Hijri output */
|
|
dwFmt = LOCALE_SLONGDATE;
|
|
break;
|
|
|
|
case FMT_DATE_MEDIUM:
|
|
FIXME("Medium date treated as long date\n");
|
|
dwFmt = LOCALE_SLONGDATE;
|
|
break;
|
|
|
|
case FMT_DATE_DAY_WEEK:
|
|
szPrintFmt = szPercent_d;
|
|
if (pToken[1])
|
|
dwVal = udate.st.wDayOfWeek + 2 - pToken[1];
|
|
else
|
|
{
|
|
GetLocaleInfoW(lcid,LOCALE_RETURN_NUMBER|LOCALE_IFIRSTDAYOFWEEK,
|
|
(LPWSTR)&dwVal, sizeof(dwVal)/sizeof(WCHAR));
|
|
dwVal = udate.st.wDayOfWeek + 1 - dwVal;
|
|
}
|
|
pToken++;
|
|
break;
|
|
|
|
case FMT_DATE_WEEK_YEAR:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.wDayOfYear / 7 + 1;
|
|
pToken += 2;
|
|
FIXME("Ignoring nFirstDay of %d, nFirstWeek of %d\n", pToken[0], pToken[1]);
|
|
break;
|
|
|
|
case FMT_DATE_MON:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.st.wMonth;
|
|
break;
|
|
|
|
case FMT_DATE_MON_0:
|
|
szPrintFmt = szPercentZeroTwo_d;
|
|
dwVal = udate.st.wMonth;
|
|
break;
|
|
|
|
case FMT_DATE_MON_SHORT:
|
|
/* FIXME: VARIANT_CALENDAR HIJRI should cause Hijri output */
|
|
TRACE("short month\n");
|
|
localeValue = LOCALE_SABBREVMONTHNAME1 + udate.st.wMonth - 1;
|
|
defaultChar = '?';
|
|
break;
|
|
|
|
case FMT_DATE_MON_LONG:
|
|
/* FIXME: VARIANT_CALENDAR HIJRI should cause Hijri output */
|
|
TRACE("long month\n");
|
|
localeValue = LOCALE_SMONTHNAME1 + udate.st.wMonth - 1;
|
|
defaultChar = '?';
|
|
break;
|
|
|
|
case FMT_DATE_YEAR_DOY:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.wDayOfYear;
|
|
break;
|
|
|
|
case FMT_DATE_YEAR_0:
|
|
szPrintFmt = szPercentZeroTwo_d;
|
|
dwVal = udate.st.wYear % 100;
|
|
break;
|
|
|
|
case FMT_DATE_YEAR_LONG:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.st.wYear;
|
|
break;
|
|
|
|
case FMT_DATE_MIN:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.st.wMinute;
|
|
break;
|
|
|
|
case FMT_DATE_MIN_0:
|
|
szPrintFmt = szPercentZeroTwo_d;
|
|
dwVal = udate.st.wMinute;
|
|
break;
|
|
|
|
case FMT_DATE_SEC:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.st.wSecond;
|
|
break;
|
|
|
|
case FMT_DATE_SEC_0:
|
|
szPrintFmt = szPercentZeroTwo_d;
|
|
dwVal = udate.st.wSecond;
|
|
break;
|
|
|
|
case FMT_DATE_HOUR:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.st.wHour;
|
|
break;
|
|
|
|
case FMT_DATE_HOUR_0:
|
|
case FMT_DATE_TIME_UNK2:
|
|
szPrintFmt = szPercentZeroTwo_d;
|
|
dwVal = udate.st.wHour;
|
|
break;
|
|
|
|
case FMT_DATE_HOUR_12:
|
|
szPrintFmt = szPercent_d;
|
|
dwVal = udate.st.wHour ? udate.st.wHour > 12 ? udate.st.wHour - 12 : udate.st.wHour : 12;
|
|
break;
|
|
|
|
case FMT_DATE_HOUR_12_0:
|
|
szPrintFmt = szPercentZeroTwo_d;
|
|
dwVal = udate.st.wHour ? udate.st.wHour > 12 ? udate.st.wHour - 12 : udate.st.wHour : 12;
|
|
break;
|
|
|
|
case FMT_DATE_AMPM_SYS1:
|
|
case FMT_DATE_AMPM_SYS2:
|
|
localeValue = udate.st.wHour < 12 ? LOCALE_S1159 : LOCALE_S2359;
|
|
defaultChar = '?';
|
|
break;
|
|
|
|
case FMT_DATE_AMPM_UPPER:
|
|
*pBuff++ = udate.st.wHour < 12 ? 'A' : 'P';
|
|
*pBuff++ = 'M';
|
|
break;
|
|
|
|
case FMT_DATE_A_UPPER:
|
|
*pBuff++ = udate.st.wHour < 12 ? 'A' : 'P';
|
|
break;
|
|
|
|
case FMT_DATE_AMPM_LOWER:
|
|
*pBuff++ = udate.st.wHour < 12 ? 'a' : 'p';
|
|
*pBuff++ = 'm';
|
|
break;
|
|
|
|
case FMT_DATE_A_LOWER:
|
|
*pBuff++ = udate.st.wHour < 12 ? 'a' : 'p';
|
|
break;
|
|
|
|
default:
|
|
ERR("Unknown token 0x%02x!\n", *pToken);
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatDate_Exit;
|
|
}
|
|
if (localeValue)
|
|
{
|
|
*pBuff = '\0';
|
|
if (GetLocaleInfoW(lcid, localeValue, pBuff,
|
|
sizeof(buff)/sizeof(WCHAR)-(pBuff-buff)))
|
|
{
|
|
TRACE("added %s\n", debugstr_w(pBuff));
|
|
while (*pBuff)
|
|
pBuff++;
|
|
}
|
|
else
|
|
{
|
|
TRACE("added %d %c\n", defaultChar, defaultChar);
|
|
*pBuff++ = defaultChar;
|
|
}
|
|
}
|
|
else if (dwFmt)
|
|
{
|
|
WCHAR fmt_buff[80];
|
|
|
|
if (!GetLocaleInfoW(lcid, dwFmt, fmt_buff, sizeof(fmt_buff)/sizeof(WCHAR)) ||
|
|
!get_date_format(lcid, 0, &udate.st, fmt_buff, pBuff,
|
|
sizeof(buff)/sizeof(WCHAR)-(pBuff-buff)))
|
|
{
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatDate_Exit;
|
|
}
|
|
while (*pBuff)
|
|
pBuff++;
|
|
}
|
|
else if (szPrintFmt)
|
|
{
|
|
sprintfW(pBuff, szPrintFmt, dwVal);
|
|
while (*pBuff)
|
|
pBuff++;
|
|
}
|
|
pToken++;
|
|
}
|
|
|
|
VARIANT_FormatDate_Exit:
|
|
*pBuff = '\0';
|
|
TRACE("buff is %s\n", debugstr_w(buff));
|
|
if (SUCCEEDED(hRes))
|
|
{
|
|
*pbstrOut = SysAllocString(buff);
|
|
if (!*pbstrOut)
|
|
hRes = E_OUTOFMEMORY;
|
|
}
|
|
return hRes;
|
|
}
|
|
|
|
/* Format a variant using a string format */
|
|
static HRESULT VARIANT_FormatString(LPVARIANT pVarIn, LPOLESTR lpszFormat,
|
|
LPBYTE rgbTok, ULONG dwFlags,
|
|
BSTR *pbstrOut, LCID lcid)
|
|
{
|
|
static WCHAR szEmpty[] = { '\0' };
|
|
WCHAR buff[256], *pBuff = buff;
|
|
WCHAR *pSrc;
|
|
FMT_HEADER *header = (FMT_HEADER*)rgbTok;
|
|
FMT_STRING_HEADER *strHeader;
|
|
const BYTE* pToken = NULL;
|
|
VARIANT vStr;
|
|
int blanks_first;
|
|
BOOL bUpper = FALSE;
|
|
HRESULT hRes = S_OK;
|
|
|
|
TRACE("(%p->(%s%s),%s,%p,0x%08x,%p,0x%08x)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), debugstr_w(lpszFormat), rgbTok, dwFlags, pbstrOut,
|
|
lcid);
|
|
|
|
V_VT(&vStr) = VT_EMPTY;
|
|
|
|
if (V_TYPE(pVarIn) == VT_EMPTY || V_TYPE(pVarIn) == VT_NULL)
|
|
{
|
|
strHeader = (FMT_STRING_HEADER*)(rgbTok + FmtGetNegative(header));
|
|
V_BSTR(&vStr) = szEmpty;
|
|
}
|
|
else
|
|
{
|
|
hRes = VariantChangeTypeEx(&vStr, pVarIn, LCID_US, VARIANT_NOUSEROVERRIDE, VT_BSTR);
|
|
if (FAILED(hRes))
|
|
return hRes;
|
|
|
|
if (V_BSTR(&vStr)[0] == '\0')
|
|
strHeader = (FMT_STRING_HEADER*)(rgbTok + FmtGetNegative(header));
|
|
else
|
|
strHeader = (FMT_STRING_HEADER*)(rgbTok + FmtGetPositive(header));
|
|
}
|
|
pSrc = V_BSTR(&vStr);
|
|
if ((strHeader->flags & (FMT_FLAG_LT|FMT_FLAG_GT)) == FMT_FLAG_GT)
|
|
bUpper = TRUE;
|
|
blanks_first = strHeader->copy_chars - strlenW(pSrc);
|
|
pToken = (const BYTE*)strHeader + sizeof(FMT_DATE_HEADER);
|
|
|
|
while (*pToken != FMT_GEN_END)
|
|
{
|
|
int dwCount = 0;
|
|
|
|
if (pToken - rgbTok > header->size)
|
|
{
|
|
ERR("Ran off the end of the format!\n");
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatString_Exit;
|
|
}
|
|
|
|
switch (*pToken)
|
|
{
|
|
case FMT_GEN_COPY:
|
|
TRACE("copy %s\n", debugstr_wn(lpszFormat + pToken[1], pToken[2]));
|
|
memcpy(pBuff, lpszFormat + pToken[1], pToken[2] * sizeof(WCHAR));
|
|
pBuff += pToken[2];
|
|
pToken += 2;
|
|
break;
|
|
|
|
case FMT_STR_COPY_SPACE:
|
|
case FMT_STR_COPY_SKIP:
|
|
dwCount = pToken[1];
|
|
if (*pToken == FMT_STR_COPY_SPACE && blanks_first > 0)
|
|
{
|
|
TRACE("insert %d initial spaces\n", blanks_first);
|
|
while (dwCount > 0 && blanks_first > 0)
|
|
{
|
|
*pBuff++ = ' ';
|
|
dwCount--;
|
|
blanks_first--;
|
|
}
|
|
}
|
|
TRACE("copy %d chars%s\n", dwCount,
|
|
*pToken == FMT_STR_COPY_SPACE ? " with space" :"");
|
|
while (dwCount > 0 && *pSrc)
|
|
{
|
|
if (bUpper)
|
|
*pBuff++ = toupperW(*pSrc);
|
|
else
|
|
*pBuff++ = tolowerW(*pSrc);
|
|
dwCount--;
|
|
pSrc++;
|
|
}
|
|
if (*pToken == FMT_STR_COPY_SPACE && dwCount > 0)
|
|
{
|
|
TRACE("insert %d spaces\n", dwCount);
|
|
while (dwCount-- > 0)
|
|
*pBuff++ = ' ';
|
|
}
|
|
pToken++;
|
|
break;
|
|
|
|
default:
|
|
ERR("Unknown token 0x%02x!\n", *pToken);
|
|
hRes = E_INVALIDARG;
|
|
goto VARIANT_FormatString_Exit;
|
|
}
|
|
pToken++;
|
|
}
|
|
|
|
VARIANT_FormatString_Exit:
|
|
/* Copy out any remaining chars */
|
|
while (*pSrc)
|
|
{
|
|
if (bUpper)
|
|
*pBuff++ = toupperW(*pSrc);
|
|
else
|
|
*pBuff++ = tolowerW(*pSrc);
|
|
pSrc++;
|
|
}
|
|
VariantClear(&vStr);
|
|
*pBuff = '\0';
|
|
TRACE("buff is %s\n", debugstr_w(buff));
|
|
if (SUCCEEDED(hRes))
|
|
{
|
|
*pbstrOut = SysAllocString(buff);
|
|
if (!*pbstrOut)
|
|
hRes = E_OUTOFMEMORY;
|
|
}
|
|
return hRes;
|
|
}
|
|
|
|
#define NUMBER_VTBITS (VTBIT_I1|VTBIT_UI1|VTBIT_I2|VTBIT_UI2| \
|
|
VTBIT_I4|VTBIT_UI4|VTBIT_I8|VTBIT_UI8| \
|
|
VTBIT_R4|VTBIT_R8|VTBIT_CY|VTBIT_DECIMAL| \
|
|
VTBIT_BOOL|VTBIT_INT|VTBIT_UINT)
|
|
|
|
/**********************************************************************
|
|
* VarFormatFromTokens [OLEAUT32.139]
|
|
*/
|
|
HRESULT WINAPI VarFormatFromTokens(LPVARIANT pVarIn, LPOLESTR lpszFormat,
|
|
LPBYTE rgbTok, ULONG dwFlags,
|
|
BSTR *pbstrOut, LCID lcid)
|
|
{
|
|
FMT_SHORT_HEADER *header = (FMT_SHORT_HEADER *)rgbTok;
|
|
VARIANT vTmp;
|
|
HRESULT hres;
|
|
|
|
TRACE("(%p,%s,%p,%x,%p,0x%08x)\n", pVarIn, debugstr_w(lpszFormat),
|
|
rgbTok, dwFlags, pbstrOut, lcid);
|
|
|
|
if (!pbstrOut)
|
|
return E_INVALIDARG;
|
|
|
|
*pbstrOut = NULL;
|
|
|
|
if (!pVarIn || !rgbTok)
|
|
return E_INVALIDARG;
|
|
|
|
if (V_VT(pVarIn) == VT_NULL)
|
|
return S_OK;
|
|
|
|
if (*rgbTok == FMT_TO_STRING || header->type == FMT_TYPE_GENERAL)
|
|
{
|
|
/* According to MSDN, general format acts somewhat like the 'Str'
|
|
* function in Visual Basic.
|
|
*/
|
|
VarFormatFromTokens_AsStr:
|
|
V_VT(&vTmp) = VT_EMPTY;
|
|
hres = VariantChangeTypeEx(&vTmp, pVarIn, lcid, dwFlags, VT_BSTR);
|
|
*pbstrOut = V_BSTR(&vTmp);
|
|
}
|
|
else
|
|
{
|
|
if (header->type == FMT_TYPE_NUMBER ||
|
|
(header->type == FMT_TYPE_UNKNOWN && ((1 << V_TYPE(pVarIn)) & NUMBER_VTBITS)))
|
|
{
|
|
hres = VARIANT_FormatNumber(pVarIn, lpszFormat, rgbTok, dwFlags, pbstrOut, lcid);
|
|
}
|
|
else if (header->type == FMT_TYPE_DATE ||
|
|
(header->type == FMT_TYPE_UNKNOWN && V_TYPE(pVarIn) == VT_DATE))
|
|
{
|
|
hres = VARIANT_FormatDate(pVarIn, lpszFormat, rgbTok, dwFlags, pbstrOut, lcid);
|
|
}
|
|
else if (header->type == FMT_TYPE_STRING || V_TYPE(pVarIn) == VT_BSTR)
|
|
{
|
|
hres = VARIANT_FormatString(pVarIn, lpszFormat, rgbTok, dwFlags, pbstrOut, lcid);
|
|
}
|
|
else
|
|
{
|
|
ERR("unrecognised format type 0x%02x\n", header->type);
|
|
return E_INVALIDARG;
|
|
}
|
|
/* If the coercion failed, still try to create output, unless the
|
|
* VAR_FORMAT_NOSUBSTITUTE flag is set.
|
|
*/
|
|
if ((hres == DISP_E_OVERFLOW || hres == DISP_E_TYPEMISMATCH) &&
|
|
!(dwFlags & VAR_FORMAT_NOSUBSTITUTE))
|
|
goto VarFormatFromTokens_AsStr;
|
|
}
|
|
|
|
return hres;
|
|
}
|
|
|
|
/**********************************************************************
|
|
* VarFormat [OLEAUT32.87]
|
|
*
|
|
* Format a variant from a format string.
|
|
*
|
|
* PARAMS
|
|
* pVarIn [I] Variant to format
|
|
* lpszFormat [I] Format string (see notes)
|
|
* nFirstDay [I] First day of the week, (See VarTokenizeFormatString() for details)
|
|
* nFirstWeek [I] First week of the year (See VarTokenizeFormatString() for details)
|
|
* dwFlags [I] Flags for the format (VAR_ flags from "oleauto.h")
|
|
* pbstrOut [O] Destination for formatted string.
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK. pbstrOut contains the formatted value.
|
|
* Failure: E_INVALIDARG, if any parameter is invalid.
|
|
* E_OUTOFMEMORY, if enough memory cannot be allocated.
|
|
* DISP_E_TYPEMISMATCH, if the variant cannot be formatted.
|
|
*
|
|
* NOTES
|
|
* - See Variant-Formats for details concerning creating format strings.
|
|
* - This function uses LOCALE_USER_DEFAULT when calling VarTokenizeFormatString()
|
|
* and VarFormatFromTokens().
|
|
*/
|
|
HRESULT WINAPI VarFormat(LPVARIANT pVarIn, LPOLESTR lpszFormat,
|
|
int nFirstDay, int nFirstWeek, ULONG dwFlags,
|
|
BSTR *pbstrOut)
|
|
{
|
|
BYTE buff[256];
|
|
HRESULT hres;
|
|
|
|
TRACE("(%p->(%s%s),%s,%d,%d,0x%08x,%p)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), debugstr_w(lpszFormat), nFirstDay, nFirstWeek,
|
|
dwFlags, pbstrOut);
|
|
|
|
if (!pbstrOut)
|
|
return E_INVALIDARG;
|
|
*pbstrOut = NULL;
|
|
|
|
hres = VarTokenizeFormatString(lpszFormat, buff, sizeof(buff), nFirstDay,
|
|
nFirstWeek, LOCALE_USER_DEFAULT, NULL);
|
|
if (SUCCEEDED(hres))
|
|
hres = VarFormatFromTokens(pVarIn, lpszFormat, buff, dwFlags,
|
|
pbstrOut, LOCALE_USER_DEFAULT);
|
|
TRACE("returning 0x%08x, %s\n", hres, debugstr_w(*pbstrOut));
|
|
return hres;
|
|
}
|
|
|
|
/**********************************************************************
|
|
* VarFormatDateTime [OLEAUT32.97]
|
|
*
|
|
* Format a variant value as a date and/or time.
|
|
*
|
|
* PARAMS
|
|
* pVarIn [I] Variant to format
|
|
* nFormat [I] Format type (see notes)
|
|
* dwFlags [I] Flags for the format (VAR_ flags from "oleauto.h")
|
|
* pbstrOut [O] Destination for formatted string.
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK. pbstrOut contains the formatted value.
|
|
* Failure: E_INVALIDARG, if any parameter is invalid.
|
|
* E_OUTOFMEMORY, if enough memory cannot be allocated.
|
|
* DISP_E_TYPEMISMATCH, if the variant cannot be formatted.
|
|
*
|
|
* NOTES
|
|
* This function uses LOCALE_USER_DEFAULT when determining the date format
|
|
* characters to use.
|
|
* Possible values for the nFormat parameter are:
|
|
*| Value Meaning
|
|
*| ----- -------
|
|
*| 0 General date format
|
|
*| 1 Long date format
|
|
*| 2 Short date format
|
|
*| 3 Long time format
|
|
*| 4 Short time format
|
|
*/
|
|
HRESULT WINAPI VarFormatDateTime(LPVARIANT pVarIn, INT nFormat, ULONG dwFlags, BSTR *pbstrOut)
|
|
{
|
|
static WCHAR szEmpty[] = { '\0' };
|
|
const BYTE* lpFmt = NULL;
|
|
|
|
TRACE("(%p->(%s%s),%d,0x%08x,%p)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), nFormat, dwFlags, pbstrOut);
|
|
|
|
if (!pVarIn || !pbstrOut || nFormat < 0 || nFormat > 4)
|
|
return E_INVALIDARG;
|
|
|
|
switch (nFormat)
|
|
{
|
|
case 0: lpFmt = fmtGeneralDate; break;
|
|
case 1: lpFmt = fmtLongDate; break;
|
|
case 2: lpFmt = fmtShortDate; break;
|
|
case 3: lpFmt = fmtLongTime; break;
|
|
case 4: lpFmt = fmtShortTime; break;
|
|
}
|
|
return VarFormatFromTokens(pVarIn, szEmpty, (BYTE*)lpFmt, dwFlags,
|
|
pbstrOut, LOCALE_USER_DEFAULT);
|
|
}
|
|
|
|
#define GETLOCALENUMBER(type,field) GetLocaleInfoW(LOCALE_USER_DEFAULT, \
|
|
type|LOCALE_RETURN_NUMBER, \
|
|
(LPWSTR)&numfmt.field, \
|
|
sizeof(numfmt.field)/sizeof(WCHAR))
|
|
|
|
/**********************************************************************
|
|
* VarFormatNumber [OLEAUT32.107]
|
|
*
|
|
* Format a variant value as a number.
|
|
*
|
|
* PARAMS
|
|
* pVarIn [I] Variant to format
|
|
* nDigits [I] Number of digits following the decimal point (-1 = user default)
|
|
* nLeading [I] Use a leading zero (-2 = user default, -1 = yes, 0 = no)
|
|
* nParens [I] Use brackets for values < 0 (-2 = user default, -1 = yes, 0 = no)
|
|
* nGrouping [I] Use grouping characters (-2 = user default, -1 = yes, 0 = no)
|
|
* dwFlags [I] Currently unused, set to zero
|
|
* pbstrOut [O] Destination for formatted string.
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK. pbstrOut contains the formatted value.
|
|
* Failure: E_INVALIDARG, if any parameter is invalid.
|
|
* E_OUTOFMEMORY, if enough memory cannot be allocated.
|
|
* DISP_E_TYPEMISMATCH, if the variant cannot be formatted.
|
|
*
|
|
* NOTES
|
|
* This function uses LOCALE_USER_DEFAULT when determining the number format
|
|
* characters to use.
|
|
*/
|
|
HRESULT WINAPI VarFormatNumber(LPVARIANT pVarIn, INT nDigits, INT nLeading, INT nParens,
|
|
INT nGrouping, ULONG dwFlags, BSTR *pbstrOut)
|
|
{
|
|
HRESULT hRet;
|
|
VARIANT vStr;
|
|
|
|
TRACE("(%p->(%s%s),%d,%d,%d,%d,0x%08x,%p)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), nDigits, nLeading, nParens, nGrouping, dwFlags, pbstrOut);
|
|
|
|
if (!pVarIn || !pbstrOut || nDigits > 9)
|
|
return E_INVALIDARG;
|
|
|
|
*pbstrOut = NULL;
|
|
|
|
V_VT(&vStr) = VT_EMPTY;
|
|
hRet = VariantCopyInd(&vStr, pVarIn);
|
|
|
|
if (SUCCEEDED(hRet))
|
|
hRet = VariantChangeTypeEx(&vStr, &vStr, LCID_US, 0, VT_BSTR);
|
|
|
|
if (SUCCEEDED(hRet))
|
|
{
|
|
WCHAR buff[256], decimal[8], thousands[8];
|
|
NUMBERFMTW numfmt;
|
|
|
|
/* Although MSDN makes it clear that the native versions of these functions
|
|
* are implemented using VarTokenizeFormatString()/VarFormatFromTokens(),
|
|
* using NLS gives us the same result.
|
|
*/
|
|
if (nDigits < 0)
|
|
GETLOCALENUMBER(LOCALE_IDIGITS, NumDigits);
|
|
else
|
|
numfmt.NumDigits = nDigits;
|
|
|
|
if (nLeading == -2)
|
|
GETLOCALENUMBER(LOCALE_ILZERO, LeadingZero);
|
|
else if (nLeading == -1)
|
|
numfmt.LeadingZero = 1;
|
|
else
|
|
numfmt.LeadingZero = 0;
|
|
|
|
if (nGrouping == -2)
|
|
{
|
|
WCHAR grouping[16];
|
|
grouping[2] = '\0';
|
|
GetLocaleInfoW(LOCALE_USER_DEFAULT, LOCALE_SDECIMAL, grouping,
|
|
sizeof(grouping)/sizeof(WCHAR));
|
|
numfmt.Grouping = grouping[2] == '2' ? 32 : grouping[0] - '0';
|
|
}
|
|
else if (nGrouping == -1)
|
|
numfmt.Grouping = 3; /* 3 = "n,nnn.nn" */
|
|
else
|
|
numfmt.Grouping = 0; /* 0 = No grouping */
|
|
|
|
if (nParens == -2)
|
|
GETLOCALENUMBER(LOCALE_INEGNUMBER, NegativeOrder);
|
|
else if (nParens == -1)
|
|
numfmt.NegativeOrder = 0; /* 0 = "(xxx)" */
|
|
else
|
|
numfmt.NegativeOrder = 1; /* 1 = "-xxx" */
|
|
|
|
numfmt.lpDecimalSep = decimal;
|
|
GetLocaleInfoW(LOCALE_USER_DEFAULT, LOCALE_SDECIMAL, decimal,
|
|
sizeof(decimal)/sizeof(WCHAR));
|
|
numfmt.lpThousandSep = thousands;
|
|
GetLocaleInfoW(LOCALE_USER_DEFAULT, LOCALE_SDECIMAL, thousands,
|
|
sizeof(thousands)/sizeof(WCHAR));
|
|
|
|
if (GetNumberFormatW(LOCALE_USER_DEFAULT, 0, V_BSTR(&vStr), &numfmt,
|
|
buff, sizeof(buff)/sizeof(WCHAR)))
|
|
{
|
|
*pbstrOut = SysAllocString(buff);
|
|
if (!*pbstrOut)
|
|
hRet = E_OUTOFMEMORY;
|
|
}
|
|
else
|
|
hRet = DISP_E_TYPEMISMATCH;
|
|
|
|
SysFreeString(V_BSTR(&vStr));
|
|
}
|
|
return hRet;
|
|
}
|
|
|
|
/**********************************************************************
|
|
* VarFormatPercent [OLEAUT32.117]
|
|
*
|
|
* Format a variant value as a percentage.
|
|
*
|
|
* PARAMS
|
|
* pVarIn [I] Variant to format
|
|
* nDigits [I] Number of digits following the decimal point (-1 = user default)
|
|
* nLeading [I] Use a leading zero (-2 = user default, -1 = yes, 0 = no)
|
|
* nParens [I] Use brackets for values < 0 (-2 = user default, -1 = yes, 0 = no)
|
|
* nGrouping [I] Use grouping characters (-2 = user default, -1 = yes, 0 = no)
|
|
* dwFlags [I] Currently unused, set to zero
|
|
* pbstrOut [O] Destination for formatted string.
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK. pbstrOut contains the formatted value.
|
|
* Failure: E_INVALIDARG, if any parameter is invalid.
|
|
* E_OUTOFMEMORY, if enough memory cannot be allocated.
|
|
* DISP_E_OVERFLOW, if overflow occurs during the conversion.
|
|
* DISP_E_TYPEMISMATCH, if the variant cannot be formatted.
|
|
*
|
|
* NOTES
|
|
* This function uses LOCALE_USER_DEFAULT when determining the number format
|
|
* characters to use.
|
|
*/
|
|
HRESULT WINAPI VarFormatPercent(LPVARIANT pVarIn, INT nDigits, INT nLeading, INT nParens,
|
|
INT nGrouping, ULONG dwFlags, BSTR *pbstrOut)
|
|
{
|
|
static const WCHAR szPercent[] = { '%','\0' };
|
|
static const WCHAR szPercentBracket[] = { '%',')','\0' };
|
|
WCHAR buff[256];
|
|
HRESULT hRet;
|
|
VARIANT vDbl;
|
|
|
|
TRACE("(%p->(%s%s),%d,%d,%d,%d,0x%08x,%p)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), nDigits, nLeading, nParens, nGrouping,
|
|
dwFlags, pbstrOut);
|
|
|
|
if (!pVarIn || !pbstrOut || nDigits > 9)
|
|
return E_INVALIDARG;
|
|
|
|
*pbstrOut = NULL;
|
|
|
|
V_VT(&vDbl) = VT_EMPTY;
|
|
hRet = VariantCopyInd(&vDbl, pVarIn);
|
|
|
|
if (SUCCEEDED(hRet))
|
|
{
|
|
hRet = VariantChangeTypeEx(&vDbl, &vDbl, LOCALE_USER_DEFAULT, 0, VT_R8);
|
|
|
|
if (SUCCEEDED(hRet))
|
|
{
|
|
if (V_R8(&vDbl) > (R8_MAX / 100.0))
|
|
return DISP_E_OVERFLOW;
|
|
|
|
V_R8(&vDbl) *= 100.0;
|
|
hRet = VarFormatNumber(&vDbl, nDigits, nLeading, nParens,
|
|
nGrouping, dwFlags, pbstrOut);
|
|
|
|
if (SUCCEEDED(hRet))
|
|
{
|
|
DWORD dwLen = strlenW(*pbstrOut);
|
|
BOOL bBracket = (*pbstrOut)[dwLen] == ')' ? TRUE : FALSE;
|
|
|
|
dwLen -= bBracket;
|
|
memcpy(buff, *pbstrOut, dwLen * sizeof(WCHAR));
|
|
strcpyW(buff + dwLen, bBracket ? szPercentBracket : szPercent);
|
|
SysFreeString(*pbstrOut);
|
|
*pbstrOut = SysAllocString(buff);
|
|
if (!*pbstrOut)
|
|
hRet = E_OUTOFMEMORY;
|
|
}
|
|
}
|
|
}
|
|
return hRet;
|
|
}
|
|
|
|
/**********************************************************************
|
|
* VarFormatCurrency [OLEAUT32.127]
|
|
*
|
|
* Format a variant value as a currency.
|
|
*
|
|
* PARAMS
|
|
* pVarIn [I] Variant to format
|
|
* nDigits [I] Number of digits following the decimal point (-1 = user default)
|
|
* nLeading [I] Use a leading zero (-2 = user default, -1 = yes, 0 = no)
|
|
* nParens [I] Use brackets for values < 0 (-2 = user default, -1 = yes, 0 = no)
|
|
* nGrouping [I] Use grouping characters (-2 = user default, -1 = yes, 0 = no)
|
|
* dwFlags [I] Currently unused, set to zero
|
|
* pbstrOut [O] Destination for formatted string.
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK. pbstrOut contains the formatted value.
|
|
* Failure: E_INVALIDARG, if any parameter is invalid.
|
|
* E_OUTOFMEMORY, if enough memory cannot be allocated.
|
|
* DISP_E_TYPEMISMATCH, if the variant cannot be formatted.
|
|
*
|
|
* NOTES
|
|
* This function uses LOCALE_USER_DEFAULT when determining the currency format
|
|
* characters to use.
|
|
*/
|
|
HRESULT WINAPI VarFormatCurrency(LPVARIANT pVarIn, INT nDigits, INT nLeading,
|
|
INT nParens, INT nGrouping, ULONG dwFlags,
|
|
BSTR *pbstrOut)
|
|
{
|
|
HRESULT hRet;
|
|
VARIANT vStr;
|
|
|
|
TRACE("(%p->(%s%s),%d,%d,%d,%d,0x%08x,%p)\n", pVarIn, debugstr_VT(pVarIn),
|
|
debugstr_VF(pVarIn), nDigits, nLeading, nParens, nGrouping, dwFlags, pbstrOut);
|
|
|
|
if (!pVarIn || !pbstrOut || nDigits > 9)
|
|
return E_INVALIDARG;
|
|
|
|
*pbstrOut = NULL;
|
|
|
|
V_VT(&vStr) = VT_EMPTY;
|
|
hRet = VariantCopyInd(&vStr, pVarIn);
|
|
|
|
if (SUCCEEDED(hRet))
|
|
hRet = VariantChangeTypeEx(&vStr, &vStr, LOCALE_USER_DEFAULT, 0, VT_BSTR);
|
|
|
|
if (SUCCEEDED(hRet))
|
|
{
|
|
WCHAR buff[256], decimal[8], thousands[8], currency[8];
|
|
CURRENCYFMTW numfmt;
|
|
|
|
if (nDigits < 0)
|
|
GETLOCALENUMBER(LOCALE_IDIGITS, NumDigits);
|
|
else
|
|
numfmt.NumDigits = nDigits;
|
|
|
|
if (nLeading == -2)
|
|
GETLOCALENUMBER(LOCALE_ILZERO, LeadingZero);
|
|
else if (nLeading == -1)
|
|
numfmt.LeadingZero = 1;
|
|
else
|
|
numfmt.LeadingZero = 0;
|
|
|
|
if (nGrouping == -2)
|
|
{
|
|
WCHAR grouping[16];
|
|
grouping[2] = '\0';
|
|
GetLocaleInfoW(LOCALE_USER_DEFAULT, LOCALE_SDECIMAL, grouping,
|
|
sizeof(grouping)/sizeof(WCHAR));
|
|
numfmt.Grouping = grouping[2] == '2' ? 32 : grouping[0] - '0';
|
|
}
|
|
else if (nGrouping == -1)
|
|
numfmt.Grouping = 3; /* 3 = "n,nnn.nn" */
|
|
else
|
|
numfmt.Grouping = 0; /* 0 = No grouping */
|
|
|
|
if (nParens == -2)
|
|
GETLOCALENUMBER(LOCALE_INEGCURR, NegativeOrder);
|
|
else if (nParens == -1)
|
|
numfmt.NegativeOrder = 0; /* 0 = "(xxx)" */
|
|
else
|
|
numfmt.NegativeOrder = 1; /* 1 = "-xxx" */
|
|
|
|
GETLOCALENUMBER(LOCALE_ICURRENCY, PositiveOrder);
|
|
|
|
numfmt.lpDecimalSep = decimal;
|
|
GetLocaleInfoW(LOCALE_USER_DEFAULT, LOCALE_SDECIMAL, decimal,
|
|
sizeof(decimal)/sizeof(WCHAR));
|
|
numfmt.lpThousandSep = thousands;
|
|
GetLocaleInfoW(LOCALE_USER_DEFAULT, LOCALE_SDECIMAL, thousands,
|
|
sizeof(thousands)/sizeof(WCHAR));
|
|
numfmt.lpCurrencySymbol = currency;
|
|
GetLocaleInfoW(LOCALE_USER_DEFAULT, LOCALE_SDECIMAL, currency,
|
|
sizeof(currency)/sizeof(WCHAR));
|
|
|
|
/* use NLS as per VarFormatNumber() */
|
|
if (GetCurrencyFormatW(LOCALE_USER_DEFAULT, 0, V_BSTR(&vStr), &numfmt,
|
|
buff, sizeof(buff)/sizeof(WCHAR)))
|
|
{
|
|
*pbstrOut = SysAllocString(buff);
|
|
if (!*pbstrOut)
|
|
hRet = E_OUTOFMEMORY;
|
|
}
|
|
else
|
|
hRet = DISP_E_TYPEMISMATCH;
|
|
|
|
SysFreeString(V_BSTR(&vStr));
|
|
}
|
|
return hRet;
|
|
}
|
|
|
|
/**********************************************************************
|
|
* VarMonthName [OLEAUT32.129]
|
|
*
|
|
* Print the specified month as localized name.
|
|
*
|
|
* PARAMS
|
|
* iMonth [I] month number 1..12
|
|
* fAbbrev [I] 0 - full name, !0 - abbreviated name
|
|
* dwFlags [I] flag stuff. only VAR_CALENDAR_HIJRI possible.
|
|
* pbstrOut [O] Destination for month name
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK. pbstrOut contains the name.
|
|
* Failure: E_INVALIDARG, if any parameter is invalid.
|
|
* E_OUTOFMEMORY, if enough memory cannot be allocated.
|
|
*/
|
|
HRESULT WINAPI VarMonthName(INT iMonth, INT fAbbrev, ULONG dwFlags, BSTR *pbstrOut)
|
|
{
|
|
DWORD localeValue;
|
|
INT size;
|
|
|
|
if ((iMonth < 1) || (iMonth > 12))
|
|
return E_INVALIDARG;
|
|
|
|
if (dwFlags)
|
|
FIXME("Does not support dwFlags 0x%x, ignoring.\n", dwFlags);
|
|
|
|
if (fAbbrev)
|
|
localeValue = LOCALE_SABBREVMONTHNAME1 + iMonth - 1;
|
|
else
|
|
localeValue = LOCALE_SMONTHNAME1 + iMonth - 1;
|
|
|
|
size = GetLocaleInfoW(LOCALE_USER_DEFAULT,localeValue, NULL, 0);
|
|
if (!size) {
|
|
ERR("GetLocaleInfo 0x%x failed.\n", localeValue);
|
|
return HRESULT_FROM_WIN32(GetLastError());
|
|
}
|
|
*pbstrOut = SysAllocStringLen(NULL,size - 1);
|
|
if (!*pbstrOut)
|
|
return E_OUTOFMEMORY;
|
|
size = GetLocaleInfoW(LOCALE_USER_DEFAULT,localeValue, *pbstrOut, size);
|
|
if (!size) {
|
|
ERR("GetLocaleInfo of 0x%x failed in 2nd stage?!\n", localeValue);
|
|
SysFreeString(*pbstrOut);
|
|
return HRESULT_FROM_WIN32(GetLastError());
|
|
}
|
|
return S_OK;
|
|
}
|
|
|
|
/**********************************************************************
|
|
* VarWeekdayName [OLEAUT32.129]
|
|
*
|
|
* Print the specified weekday as localized name.
|
|
*
|
|
* PARAMS
|
|
* iWeekday [I] day of week, 1..7, 1="the first day of the week"
|
|
* fAbbrev [I] 0 - full name, !0 - abbreviated name
|
|
* iFirstDay [I] first day of week,
|
|
* 0=system default, 1=Sunday, 2=Monday, .. (contrary to MSDN)
|
|
* dwFlags [I] flag stuff. only VAR_CALENDAR_HIJRI possible.
|
|
* pbstrOut [O] Destination for weekday name.
|
|
*
|
|
* RETURNS
|
|
* Success: S_OK, pbstrOut contains the name.
|
|
* Failure: E_INVALIDARG, if any parameter is invalid.
|
|
* E_OUTOFMEMORY, if enough memory cannot be allocated.
|
|
*/
|
|
HRESULT WINAPI VarWeekdayName(INT iWeekday, INT fAbbrev, INT iFirstDay,
|
|
ULONG dwFlags, BSTR *pbstrOut)
|
|
{
|
|
DWORD localeValue;
|
|
INT size;
|
|
|
|
/* Windows XP oleaut32.dll doesn't allow iWekday==0, contrary to MSDN */
|
|
if (iWeekday < 1 || iWeekday > 7)
|
|
return E_INVALIDARG;
|
|
if (iFirstDay < 0 || iFirstDay > 7)
|
|
return E_INVALIDARG;
|
|
if (!pbstrOut)
|
|
return E_INVALIDARG;
|
|
|
|
if (dwFlags)
|
|
FIXME("Does not support dwFlags 0x%x, ignoring.\n", dwFlags);
|
|
|
|
/* If we have to use the default firstDay, find which one it is */
|
|
if (iFirstDay == 0) {
|
|
DWORD firstDay;
|
|
localeValue = LOCALE_RETURN_NUMBER | LOCALE_IFIRSTDAYOFWEEK;
|
|
size = GetLocaleInfoW(LOCALE_USER_DEFAULT, localeValue,
|
|
(LPWSTR)&firstDay, sizeof(firstDay) / sizeof(WCHAR));
|
|
if (!size) {
|
|
ERR("GetLocaleInfo 0x%x failed.\n", localeValue);
|
|
return HRESULT_FROM_WIN32(GetLastError());
|
|
}
|
|
iFirstDay = firstDay + 2;
|
|
}
|
|
|
|
/* Determine what we need to return */
|
|
localeValue = fAbbrev ? LOCALE_SABBREVDAYNAME1 : LOCALE_SDAYNAME1;
|
|
localeValue += (7 + iWeekday - 1 + iFirstDay - 2) % 7;
|
|
|
|
/* Determine the size of the data, allocate memory and retrieve the data */
|
|
size = GetLocaleInfoW(LOCALE_USER_DEFAULT, localeValue, NULL, 0);
|
|
if (!size) {
|
|
ERR("GetLocaleInfo 0x%x failed.\n", localeValue);
|
|
return HRESULT_FROM_WIN32(GetLastError());
|
|
}
|
|
*pbstrOut = SysAllocStringLen(NULL, size - 1);
|
|
if (!*pbstrOut)
|
|
return E_OUTOFMEMORY;
|
|
size = GetLocaleInfoW(LOCALE_USER_DEFAULT, localeValue, *pbstrOut, size);
|
|
if (!size) {
|
|
ERR("GetLocaleInfo 0x%x failed in 2nd stage?!\n", localeValue);
|
|
SysFreeString(*pbstrOut);
|
|
return HRESULT_FROM_WIN32(GetLastError());
|
|
}
|
|
return S_OK;
|
|
}
|