mirror of
https://github.com/moparisthebest/Simba
synced 2024-11-27 03:22:17 -05:00
1821 lines
47 KiB
ObjectPascal
1821 lines
47 KiB
ObjectPascal
{==============================================================================|
|
||
| Project : Ararat Synapse | 004.014.000 |
|
||
|==============================================================================|
|
||
| Content: support procedures and functions |
|
||
|==============================================================================|
|
||
| Copyright (c)1999-2010, Lukas Gebauer |
|
||
| All rights reserved. |
|
||
| |
|
||
| Redistribution and use in source and binary forms, with or without |
|
||
| modification, are permitted provided that the following conditions are met: |
|
||
| |
|
||
| Redistributions of source code must retain the above copyright notice, this |
|
||
| list of conditions and the following disclaimer. |
|
||
| |
|
||
| Redistributions in binary form must reproduce the above copyright notice, |
|
||
| this list of conditions and the following disclaimer in the documentation |
|
||
| and/or other materials provided with the distribution. |
|
||
| |
|
||
| Neither the name of Lukas Gebauer nor the names of its contributors may |
|
||
| be used to endorse or promote products derived from this software without |
|
||
| specific prior written permission. |
|
||
| |
|
||
| THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
|
||
| AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
|
||
| IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
|
||
| ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR |
|
||
| ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
|
||
| DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
|
||
| SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
|
||
| CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
|
||
| LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
|
||
| OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH |
|
||
| DAMAGE. |
|
||
|==============================================================================|
|
||
| The Initial Developer of the Original Code is Lukas Gebauer (Czech Republic).|
|
||
| Portions created by Lukas Gebauer are Copyright (c) 1999-2010. |
|
||
| Portions created by Hernan Sanchez are Copyright (c) 2000. |
|
||
| All Rights Reserved. |
|
||
|==============================================================================|
|
||
| Contributor(s): |
|
||
| Hernan Sanchez (hernan.sanchez@iname.com) |
|
||
|==============================================================================|
|
||
| History: see HISTORY.HTM from distribution package |
|
||
| (Found at URL: http://www.ararat.cz/synapse/) |
|
||
|==============================================================================}
|
||
|
||
{:@abstract(Support procedures and functions)}
|
||
|
||
{$IFDEF FPC}
|
||
{$MODE DELPHI}
|
||
{$ENDIF}
|
||
{$Q-}
|
||
{$R-}
|
||
{$H+}
|
||
|
||
//old Delphi does not have MSWINDOWS define.
|
||
{$IFDEF WIN32}
|
||
{$IFNDEF MSWINDOWS}
|
||
{$DEFINE MSWINDOWS}
|
||
{$ENDIF}
|
||
{$ENDIF}
|
||
|
||
{$IFDEF UNICODE}
|
||
{$WARN IMPLICIT_STRING_CAST OFF}
|
||
{$WARN IMPLICIT_STRING_CAST_LOSS OFF}
|
||
{$WARN SUSPICIOUS_TYPECAST OFF}
|
||
{$ENDIF}
|
||
|
||
unit synautil;
|
||
|
||
interface
|
||
|
||
uses
|
||
{$IFDEF MSWINDOWS}
|
||
Windows,
|
||
{$ELSE}
|
||
{$IFDEF FPC}
|
||
UnixUtil, Unix, BaseUnix,
|
||
{$ELSE}
|
||
Libc,
|
||
{$ENDIF}
|
||
{$ENDIF}
|
||
{$IFDEF CIL}
|
||
System.IO,
|
||
{$ENDIF}
|
||
SysUtils, Classes, SynaFpc;
|
||
|
||
{$IFDEF VER100}
|
||
type
|
||
int64 = integer;
|
||
{$ENDIF}
|
||
|
||
{:Return your timezone bias from UTC time in minutes.}
|
||
function TimeZoneBias: integer;
|
||
|
||
{:Return your timezone bias from UTC time in string representation like "+0200".}
|
||
function TimeZone: string;
|
||
|
||
{:Returns current time in format defined in RFC-822. Useful for SMTP messages,
|
||
but other protocols use this time format as well. Results contains the timezone
|
||
specification. Four digit year is used to break any Y2K concerns. (Example
|
||
'Fri, 15 Oct 1999 21:14:56 +0200')}
|
||
function Rfc822DateTime(t: TDateTime): string;
|
||
|
||
{:Returns date and time in format defined in C compilers in format "mmm dd hh:nn:ss"}
|
||
function CDateTime(t: TDateTime): string;
|
||
|
||
{:Returns date and time in format defined in format 'yymmdd hhnnss'}
|
||
function SimpleDateTime(t: TDateTime): string;
|
||
|
||
{:Returns date and time in format defined in ANSI C compilers in format
|
||
"ddd mmm d hh:nn:ss yyyy" }
|
||
function AnsiCDateTime(t: TDateTime): string;
|
||
|
||
{:Decode three-letter string with name of month to their month number. If string
|
||
not match any month name, then is returned 0. For parsing are used predefined
|
||
names for English, French and German and names from system locale too.}
|
||
function GetMonthNumber(Value: String): integer;
|
||
|
||
{:Return decoded time from given string. Time must be witch separator ':'. You
|
||
can use "hh:mm" or "hh:mm:ss".}
|
||
function GetTimeFromStr(Value: string): TDateTime;
|
||
|
||
{:Decode string in format "m-d-y" to TDateTime type.}
|
||
function GetDateMDYFromStr(Value: string): TDateTime;
|
||
|
||
{:Decode various string representations of date and time to Tdatetime type.
|
||
This function do all timezone corrections too! This function can decode lot of
|
||
formats like:
|
||
@longcode(#
|
||
ddd, d mmm yyyy hh:mm:ss
|
||
ddd, d mmm yy hh:mm:ss
|
||
ddd, mmm d yyyy hh:mm:ss
|
||
ddd mmm dd hh:mm:ss yyyy #)
|
||
|
||
and more with lot of modifications, include:
|
||
@longcode(#
|
||
Sun, 06 Nov 1994 08:49:37 GMT ; RFC 822, updated by RFC 1123
|
||
Sunday, 06-Nov-94 08:49:37 GMT ; RFC 850, obsoleted by RFC 1036
|
||
Sun Nov 6 08:49:37 1994 ; ANSI C's asctime() Format
|
||
#)
|
||
Timezone corrections known lot of symbolic timezone names (like CEST, EDT, etc.)
|
||
or numeric representation (like +0200). By convention defined in RFC timezone
|
||
+0000 is GMT and -0000 is current your system timezone.}
|
||
function DecodeRfcDateTime(Value: string): TDateTime;
|
||
|
||
{:Return current system date and time in UTC timezone.}
|
||
function GetUTTime: TDateTime;
|
||
|
||
{:Set Newdt as current system date and time in UTC timezone. This function work
|
||
only if you have administrator rights!}
|
||
function SetUTTime(Newdt: TDateTime): Boolean;
|
||
|
||
{:Return current value of system timer with precizion 1 millisecond. Good for
|
||
measure time difference.}
|
||
function GetTick: LongWord;
|
||
|
||
{:Return difference between two timestamps. It working fine only for differences
|
||
smaller then maxint. (difference must be smaller then 24 days.)}
|
||
function TickDelta(TickOld, TickNew: LongWord): LongWord;
|
||
|
||
{:Return two characters, which ordinal values represents the value in byte
|
||
format. (High-endian)}
|
||
function CodeInt(Value: Word): Ansistring;
|
||
|
||
{:Decodes two characters located at "Index" offset position of the "Value"
|
||
string to Word values.}
|
||
function DecodeInt(const Value: Ansistring; Index: Integer): Word;
|
||
|
||
{:Return four characters, which ordinal values represents the value in byte
|
||
format. (High-endian)}
|
||
function CodeLongInt(Value: LongInt): Ansistring;
|
||
|
||
{:Decodes four characters located at "Index" offset position of the "Value"
|
||
string to LongInt values.}
|
||
function DecodeLongInt(const Value: Ansistring; Index: Integer): LongInt;
|
||
|
||
{:Dump binary buffer stored in a string to a result string.}
|
||
function DumpStr(const Buffer: Ansistring): string;
|
||
|
||
{:Dump binary buffer stored in a string to a result string. All bytes with code
|
||
of character is written as character, not as hexadecimal value.}
|
||
function DumpExStr(const Buffer: Ansistring): string;
|
||
|
||
{:Dump binary buffer stored in a string to a file with DumpFile filename.}
|
||
procedure Dump(const Buffer: AnsiString; DumpFile: string);
|
||
|
||
{:Dump binary buffer stored in a string to a file with DumpFile filename. All
|
||
bytes with code of character is written as character, not as hexadecimal value.}
|
||
procedure DumpEx(const Buffer: AnsiString; DumpFile: string);
|
||
|
||
{:Like TrimLeft, but remove only spaces, not control characters!}
|
||
function TrimSPLeft(const S: string): string;
|
||
|
||
{:Like TrimRight, but remove only spaces, not control characters!}
|
||
function TrimSPRight(const S: string): string;
|
||
|
||
{:Like Trim, but remove only spaces, not control characters!}
|
||
function TrimSP(const S: string): string;
|
||
|
||
{:Returns a portion of the "Value" string located to the left of the "Delimiter"
|
||
string. If a delimiter is not found, results is original string.}
|
||
function SeparateLeft(const Value, Delimiter: string): string;
|
||
|
||
{:Returns the portion of the "Value" string located to the right of the
|
||
"Delimiter" string. If a delimiter is not found, results is original string.}
|
||
function SeparateRight(const Value, Delimiter: string): string;
|
||
|
||
{:Returns parameter value from string in format:
|
||
parameter1="value1"; parameter2=value2}
|
||
function GetParameter(const Value, Parameter: string): string;
|
||
|
||
{:parse value string with elements differed by Delimiter into stringlist.}
|
||
procedure ParseParametersEx(Value, Delimiter: string; const Parameters: TStrings);
|
||
|
||
{:parse value string with elements differed by ';' into stringlist.}
|
||
procedure ParseParameters(Value: string; const Parameters: TStrings);
|
||
|
||
{:Index of string in stringlist with same beginning as Value is returned.}
|
||
function IndexByBegin(Value: string; const List: TStrings): integer;
|
||
|
||
{:Returns only the e-mail portion of an address from the full address format.
|
||
i.e. returns 'nobody@@somewhere.com' from '"someone" <nobody@@somewhere.com>'}
|
||
function GetEmailAddr(const Value: string): string;
|
||
|
||
{:Returns only the description part from a full address format. i.e. returns
|
||
'someone' from '"someone" <nobody@@somewhere.com>'}
|
||
function GetEmailDesc(Value: string): string;
|
||
|
||
{:Returns a string with hexadecimal digits representing the corresponding values
|
||
of the bytes found in "Value" string.}
|
||
function StrToHex(const Value: Ansistring): string;
|
||
|
||
{:Returns a string of binary "Digits" representing "Value".}
|
||
function IntToBin(Value: Integer; Digits: Byte): string;
|
||
|
||
{:Returns an integer equivalent of the binary string in "Value".
|
||
(i.e. ('10001010') returns 138)}
|
||
function BinToInt(const Value: string): Integer;
|
||
|
||
{:Parses a URL to its various components.}
|
||
function ParseURL(URL: string; var Prot, User, Pass, Host, Port, Path,
|
||
Para: string): string;
|
||
|
||
{:Replaces all "Search" string values found within "Value" string, with the
|
||
"Replace" string value.}
|
||
function ReplaceString(Value, Search, Replace: AnsiString): AnsiString;
|
||
|
||
{:It is like RPos, but search is from specified possition.}
|
||
function RPosEx(const Sub, Value: string; From: integer): Integer;
|
||
|
||
{:It is like POS function, but from right side of Value string.}
|
||
function RPos(const Sub, Value: String): Integer;
|
||
|
||
{:Like @link(fetch), but working with binary strings, not with text.}
|
||
function FetchBin(var Value: string; const Delimiter: string): string;
|
||
|
||
{:Fetch string from left of Value string.}
|
||
function Fetch(var Value: string; const Delimiter: string): string;
|
||
|
||
{:Fetch string from left of Value string. This function ignore delimitesr inside
|
||
quotations.}
|
||
function FetchEx(var Value: string; const Delimiter, Quotation: string): string;
|
||
|
||
{:If string is binary string (contains non-printable characters), then is
|
||
returned true.}
|
||
function IsBinaryString(const Value: AnsiString): Boolean;
|
||
|
||
{:return position of string terminator in string. If terminator found, then is
|
||
returned in terminator parameter.
|
||
Possible line terminators are: CRLF, LFCR, CR, LF}
|
||
function PosCRLF(const Value: AnsiString; var Terminator: AnsiString): integer;
|
||
|
||
{:Delete empty strings from end of stringlist.}
|
||
Procedure StringsTrim(const value: TStrings);
|
||
|
||
{:Like Pos function, buf from given string possition.}
|
||
function PosFrom(const SubStr, Value: String; From: integer): integer;
|
||
|
||
{$IFNDEF CIL}
|
||
{:Increase pointer by value.}
|
||
function IncPoint(const p: pointer; Value: integer): pointer;
|
||
{$ENDIF}
|
||
|
||
{:Get string between PairBegin and PairEnd. This function respect nesting.
|
||
For example:
|
||
@longcode(#
|
||
Value is: 'Hi! (hello(yes!))'
|
||
pairbegin is: '('
|
||
pairend is: ')'
|
||
In this case result is: 'hello(yes!)'#)}
|
||
function GetBetween(const PairBegin, PairEnd, Value: string): string;
|
||
|
||
{:Return count of Chr in Value string.}
|
||
function CountOfChar(const Value: string; Chr: char): integer;
|
||
|
||
{:Remove quotation from Value string. If Value is not quoted, then return same
|
||
string without any modification. }
|
||
function UnquoteStr(const Value: string; Quote: Char): string;
|
||
|
||
{:Quote Value string. If Value contains some Quote chars, then it is doubled.}
|
||
function QuoteStr(const Value: string; Quote: Char): string;
|
||
|
||
{:Convert lines in stringlist from 'name: value' form to 'name=value' form.}
|
||
procedure HeadersToList(const Value: TStrings);
|
||
|
||
{:Convert lines in stringlist from 'name=value' form to 'name: value' form.}
|
||
procedure ListToHeaders(const Value: TStrings);
|
||
|
||
{:swap bytes in integer.}
|
||
function SwapBytes(Value: integer): integer;
|
||
|
||
{:read string with requested length form stream.}
|
||
function ReadStrFromStream(const Stream: TStream; len: integer): AnsiString;
|
||
|
||
{:write string to stream.}
|
||
procedure WriteStrToStream(const Stream: TStream; Value: AnsiString);
|
||
|
||
{:Return filename of new temporary file in Dir (if empty, then default temporary
|
||
directory is used) and with optional filename prefix.}
|
||
function GetTempFile(const Dir, prefix: AnsiString): AnsiString;
|
||
|
||
{:Return padded string. If length is greater, string is truncated. If length is
|
||
smaller, string is padded by Pad character.}
|
||
function PadString(const Value: AnsiString; len: integer; Pad: AnsiChar): AnsiString;
|
||
|
||
{:Read header from "Value" stringlist beginning at "Index" position. If header
|
||
is Splitted into multiple lines, then this procedure de-split it into one line.}
|
||
function NormalizeHeader(Value: TStrings; var Index: Integer): string;
|
||
|
||
var
|
||
{:can be used for your own months strings for @link(getmonthnumber)}
|
||
CustomMonthNames: array[1..12] of string;
|
||
|
||
implementation
|
||
|
||
{==============================================================================}
|
||
|
||
const
|
||
MyDayNames: array[1..7] of AnsiString =
|
||
('Sun', 'Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat');
|
||
var
|
||
MyMonthNames: array[0..6, 1..12] of String =
|
||
(
|
||
('Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', //rewrited by system locales
|
||
'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'),
|
||
('Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', //English
|
||
'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'),
|
||
('jan', 'f<>v', 'mar', 'avr', 'mai', 'jun', //French
|
||
'jul', 'ao<61>', 'sep', 'oct', 'nov', 'd<>c'),
|
||
('jan', 'fev', 'mar', 'avr', 'mai', 'jun', //French#2
|
||
'jul', 'aou', 'sep', 'oct', 'nov', 'dec'),
|
||
('Jan', 'Feb', 'Mar', 'Apr', 'Mai', 'Jun', //German
|
||
'Jul', 'Aug', 'Sep', 'Okt', 'Nov', 'Dez'),
|
||
('Jan', 'Feb', 'M<>r', 'Apr', 'Mai', 'Jun', //German#2
|
||
'Jul', 'Aug', 'Sep', 'Okt', 'Nov', 'Dez'),
|
||
('Led', '<27>no', 'B<>e', 'Dub', 'Kv<4B>', '<27>en', //Czech
|
||
'<27>ec', 'Srp', 'Z<><5A>', '<27><>j', 'Lis', 'Pro')
|
||
);
|
||
|
||
|
||
{==============================================================================}
|
||
|
||
function TimeZoneBias: integer;
|
||
{$IFNDEF MSWINDOWS}
|
||
{$IFNDEF FPC}
|
||
var
|
||
t: TTime_T;
|
||
UT: TUnixTime;
|
||
begin
|
||
__time(@T);
|
||
localtime_r(@T, UT);
|
||
Result := ut.__tm_gmtoff div 60;
|
||
{$ELSE}
|
||
begin
|
||
Result := TZSeconds div 60;
|
||
{$ENDIF}
|
||
{$ELSE}
|
||
var
|
||
zoneinfo: TTimeZoneInformation;
|
||
bias: Integer;
|
||
begin
|
||
case GetTimeZoneInformation(Zoneinfo) of
|
||
2:
|
||
bias := zoneinfo.Bias + zoneinfo.DaylightBias;
|
||
1:
|
||
bias := zoneinfo.Bias + zoneinfo.StandardBias;
|
||
else
|
||
bias := zoneinfo.Bias;
|
||
end;
|
||
Result := bias * (-1);
|
||
{$ENDIF}
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function TimeZone: string;
|
||
var
|
||
bias: Integer;
|
||
h, m: Integer;
|
||
begin
|
||
bias := TimeZoneBias;
|
||
if bias >= 0 then
|
||
Result := '+'
|
||
else
|
||
Result := '-';
|
||
bias := Abs(bias);
|
||
h := bias div 60;
|
||
m := bias mod 60;
|
||
Result := Result + Format('%.2d%.2d', [h, m]);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function Rfc822DateTime(t: TDateTime): string;
|
||
var
|
||
wYear, wMonth, wDay: word;
|
||
begin
|
||
DecodeDate(t, wYear, wMonth, wDay);
|
||
Result := Format('%s, %d %s %s %s', [MyDayNames[DayOfWeek(t)], wDay,
|
||
MyMonthNames[1, wMonth], FormatDateTime('yyyy hh":"nn":"ss', t), TimeZone]);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function CDateTime(t: TDateTime): string;
|
||
var
|
||
wYear, wMonth, wDay: word;
|
||
begin
|
||
DecodeDate(t, wYear, wMonth, wDay);
|
||
Result:= Format('%s %2d %s', [MyMonthNames[1, wMonth], wDay,
|
||
FormatDateTime('hh":"nn":"ss', t)]);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function SimpleDateTime(t: TDateTime): string;
|
||
begin
|
||
Result := FormatDateTime('yymmdd hhnnss', t);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function AnsiCDateTime(t: TDateTime): string;
|
||
var
|
||
wYear, wMonth, wDay: word;
|
||
begin
|
||
DecodeDate(t, wYear, wMonth, wDay);
|
||
Result := Format('%s %s %d %s', [MyDayNames[DayOfWeek(t)], MyMonthNames[1, wMonth],
|
||
wDay, FormatDateTime('hh":"nn":"ss yyyy ', t)]);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function DecodeTimeZone(Value: string; var Zone: integer): Boolean;
|
||
var
|
||
x: integer;
|
||
zh, zm: integer;
|
||
s: string;
|
||
begin
|
||
Result := false;
|
||
s := Value;
|
||
if (Pos('+', s) = 1) or (Pos('-',s) = 1) then
|
||
begin
|
||
if s = '-0000' then
|
||
Zone := TimeZoneBias
|
||
else
|
||
if Length(s) > 4 then
|
||
begin
|
||
zh := StrToIntdef(s[2] + s[3], 0);
|
||
zm := StrToIntdef(s[4] + s[5], 0);
|
||
zone := zh * 60 + zm;
|
||
if s[1] = '-' then
|
||
zone := zone * (-1);
|
||
end;
|
||
Result := True;
|
||
end
|
||
else
|
||
begin
|
||
x := 32767;
|
||
if s = 'NZDT' then x := 13;
|
||
if s = 'IDLE' then x := 12;
|
||
if s = 'NZST' then x := 12;
|
||
if s = 'NZT' then x := 12;
|
||
if s = 'EADT' then x := 11;
|
||
if s = 'GST' then x := 10;
|
||
if s = 'JST' then x := 9;
|
||
if s = 'CCT' then x := 8;
|
||
if s = 'WADT' then x := 8;
|
||
if s = 'WAST' then x := 7;
|
||
if s = 'ZP6' then x := 6;
|
||
if s = 'ZP5' then x := 5;
|
||
if s = 'ZP4' then x := 4;
|
||
if s = 'BT' then x := 3;
|
||
if s = 'EET' then x := 2;
|
||
if s = 'MEST' then x := 2;
|
||
if s = 'MESZ' then x := 2;
|
||
if s = 'SST' then x := 2;
|
||
if s = 'FST' then x := 2;
|
||
if s = 'CEST' then x := 2;
|
||
if s = 'CET' then x := 1;
|
||
if s = 'FWT' then x := 1;
|
||
if s = 'MET' then x := 1;
|
||
if s = 'MEWT' then x := 1;
|
||
if s = 'SWT' then x := 1;
|
||
if s = 'UT' then x := 0;
|
||
if s = 'UTC' then x := 0;
|
||
if s = 'GMT' then x := 0;
|
||
if s = 'WET' then x := 0;
|
||
if s = 'WAT' then x := -1;
|
||
if s = 'BST' then x := -1;
|
||
if s = 'AT' then x := -2;
|
||
if s = 'ADT' then x := -3;
|
||
if s = 'AST' then x := -4;
|
||
if s = 'EDT' then x := -4;
|
||
if s = 'EST' then x := -5;
|
||
if s = 'CDT' then x := -5;
|
||
if s = 'CST' then x := -6;
|
||
if s = 'MDT' then x := -6;
|
||
if s = 'MST' then x := -7;
|
||
if s = 'PDT' then x := -7;
|
||
if s = 'PST' then x := -8;
|
||
if s = 'YDT' then x := -8;
|
||
if s = 'YST' then x := -9;
|
||
if s = 'HDT' then x := -9;
|
||
if s = 'AHST' then x := -10;
|
||
if s = 'CAT' then x := -10;
|
||
if s = 'HST' then x := -10;
|
||
if s = 'EAST' then x := -10;
|
||
if s = 'NT' then x := -11;
|
||
if s = 'IDLW' then x := -12;
|
||
if x <> 32767 then
|
||
begin
|
||
zone := x * 60;
|
||
Result := True;
|
||
end;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function GetMonthNumber(Value: String): integer;
|
||
var
|
||
n: integer;
|
||
function TestMonth(Value: String; Index: Integer): Boolean;
|
||
var
|
||
n: integer;
|
||
begin
|
||
Result := False;
|
||
for n := 0 to 6 do
|
||
if Value = AnsiUppercase(MyMonthNames[n, Index]) then
|
||
begin
|
||
Result := True;
|
||
Break;
|
||
end;
|
||
end;
|
||
begin
|
||
Result := 0;
|
||
Value := AnsiUppercase(Value);
|
||
for n := 1 to 12 do
|
||
if TestMonth(Value, n) or (Value = AnsiUppercase(CustomMonthNames[n])) then
|
||
begin
|
||
Result := n;
|
||
Break;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function GetTimeFromStr(Value: string): TDateTime;
|
||
var
|
||
x: integer;
|
||
begin
|
||
x := rpos(':', Value);
|
||
if (x > 0) and ((Length(Value) - x) > 2) then
|
||
Value := Copy(Value, 1, x + 2);
|
||
Value := ReplaceString(Value, ':', TimeSeparator);
|
||
Result := -1;
|
||
try
|
||
Result := StrToTime(Value);
|
||
except
|
||
on Exception do ;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function GetDateMDYFromStr(Value: string): TDateTime;
|
||
var
|
||
wYear, wMonth, wDay: word;
|
||
s: string;
|
||
begin
|
||
Result := 0;
|
||
s := Fetch(Value, '-');
|
||
wMonth := StrToIntDef(s, 12);
|
||
s := Fetch(Value, '-');
|
||
wDay := StrToIntDef(s, 30);
|
||
wYear := StrToIntDef(Value, 1899);
|
||
if wYear < 1000 then
|
||
if (wYear > 99) then
|
||
wYear := wYear + 1900
|
||
else
|
||
if wYear > 50 then
|
||
wYear := wYear + 1900
|
||
else
|
||
wYear := wYear + 2000;
|
||
try
|
||
Result := EncodeDate(wYear, wMonth, wDay);
|
||
except
|
||
on Exception do ;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function DecodeRfcDateTime(Value: string): TDateTime;
|
||
var
|
||
day, month, year: Word;
|
||
zone: integer;
|
||
x, y: integer;
|
||
s: string;
|
||
t: TDateTime;
|
||
begin
|
||
// ddd, d mmm yyyy hh:mm:ss
|
||
// ddd, d mmm yy hh:mm:ss
|
||
// ddd, mmm d yyyy hh:mm:ss
|
||
// ddd mmm dd hh:mm:ss yyyy
|
||
// Sun, 06 Nov 1994 08:49:37 GMT ; RFC 822, updated by RFC 1123
|
||
// Sunday, 06-Nov-94 08:49:37 GMT ; RFC 850, obsoleted by RFC 1036
|
||
// Sun Nov 6 08:49:37 1994 ; ANSI C's asctime() Format
|
||
|
||
Result := 0;
|
||
if Value = '' then
|
||
Exit;
|
||
day := 0;
|
||
month := 0;
|
||
year := 0;
|
||
zone := 0;
|
||
Value := ReplaceString(Value, ' -', ' #');
|
||
Value := ReplaceString(Value, '-', ' ');
|
||
Value := ReplaceString(Value, ' #', ' -');
|
||
while Value <> '' do
|
||
begin
|
||
s := Fetch(Value, ' ');
|
||
s := uppercase(s);
|
||
// timezone
|
||
if DecodetimeZone(s, x) then
|
||
begin
|
||
zone := x;
|
||
continue;
|
||
end;
|
||
x := StrToIntDef(s, 0);
|
||
// day or year
|
||
if x > 0 then
|
||
if (x < 32) and (day = 0) then
|
||
begin
|
||
day := x;
|
||
continue;
|
||
end
|
||
else
|
||
begin
|
||
if (year = 0) and ((month > 0) or (x > 12)) then
|
||
begin
|
||
year := x;
|
||
if year < 32 then
|
||
year := year + 2000;
|
||
if year < 1000 then
|
||
year := year + 1900;
|
||
continue;
|
||
end;
|
||
end;
|
||
// time
|
||
if rpos(':', s) > Pos(':', s) then
|
||
begin
|
||
t := GetTimeFromStr(s);
|
||
if t <> -1 then
|
||
Result := t;
|
||
continue;
|
||
end;
|
||
//timezone daylight saving time
|
||
if s = 'DST' then
|
||
begin
|
||
zone := zone + 60;
|
||
continue;
|
||
end;
|
||
// month
|
||
y := GetMonthNumber(s);
|
||
if (y > 0) and (month = 0) then
|
||
month := y;
|
||
end;
|
||
if year = 0 then
|
||
year := 1980;
|
||
if month < 1 then
|
||
month := 1;
|
||
if month > 12 then
|
||
month := 12;
|
||
if day < 1 then
|
||
day := 1;
|
||
x := MonthDays[IsLeapYear(year), month];
|
||
if day > x then
|
||
day := x;
|
||
Result := Result + Encodedate(year, month, day);
|
||
zone := zone - TimeZoneBias;
|
||
x := zone div 1440;
|
||
Result := Result - x;
|
||
zone := zone mod 1440;
|
||
t := EncodeTime(Abs(zone) div 60, Abs(zone) mod 60, 0, 0);
|
||
if zone < 0 then
|
||
t := 0 - t;
|
||
Result := Result - t;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function GetUTTime: TDateTime;
|
||
{$IFDEF MSWINDOWS}
|
||
{$IFNDEF FPC}
|
||
var
|
||
st: TSystemTime;
|
||
begin
|
||
GetSystemTime(st);
|
||
result := SystemTimeToDateTime(st);
|
||
{$ELSE}
|
||
var
|
||
st: SysUtils.TSystemTime;
|
||
stw: Windows.TSystemTime;
|
||
begin
|
||
GetSystemTime(stw);
|
||
st.Year := stw.wYear;
|
||
st.Month := stw.wMonth;
|
||
st.Day := stw.wDay;
|
||
st.Hour := stw.wHour;
|
||
st.Minute := stw.wMinute;
|
||
st.Second := stw.wSecond;
|
||
st.Millisecond := stw.wMilliseconds;
|
||
result := SystemTimeToDateTime(st);
|
||
{$ENDIF}
|
||
{$ELSE}
|
||
{$IFNDEF FPC}
|
||
var
|
||
TV: TTimeVal;
|
||
begin
|
||
gettimeofday(TV, nil);
|
||
Result := UnixDateDelta + (TV.tv_sec + TV.tv_usec / 1000000) / 86400;
|
||
{$ELSE}
|
||
var
|
||
TV: TimeVal;
|
||
begin
|
||
fpgettimeofday(@TV, nil);
|
||
Result := UnixDateDelta + (TV.tv_sec + TV.tv_usec / 1000000) / 86400;
|
||
{$ENDIF}
|
||
{$ENDIF}
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function SetUTTime(Newdt: TDateTime): Boolean;
|
||
{$IFDEF MSWINDOWS}
|
||
{$IFNDEF FPC}
|
||
var
|
||
st: TSystemTime;
|
||
begin
|
||
DateTimeToSystemTime(newdt,st);
|
||
Result := SetSystemTime(st);
|
||
{$ELSE}
|
||
var
|
||
st: SysUtils.TSystemTime;
|
||
stw: Windows.TSystemTime;
|
||
begin
|
||
DateTimeToSystemTime(newdt,st);
|
||
stw.wYear := st.Year;
|
||
stw.wMonth := st.Month;
|
||
stw.wDay := st.Day;
|
||
stw.wHour := st.Hour;
|
||
stw.wMinute := st.Minute;
|
||
stw.wSecond := st.Second;
|
||
stw.wMilliseconds := st.Millisecond;
|
||
Result := SetSystemTime(stw);
|
||
{$ENDIF}
|
||
{$ELSE}
|
||
{$IFNDEF FPC}
|
||
var
|
||
TV: TTimeVal;
|
||
d: double;
|
||
TZ: Ttimezone;
|
||
PZ: PTimeZone;
|
||
begin
|
||
TZ.tz_minuteswest := 0;
|
||
TZ.tz_dsttime := 0;
|
||
PZ := @TZ;
|
||
gettimeofday(TV, PZ);
|
||
d := (newdt - UnixDateDelta) * 86400;
|
||
TV.tv_sec := trunc(d);
|
||
TV.tv_usec := trunc(frac(d) * 1000000);
|
||
Result := settimeofday(TV, TZ) <> -1;
|
||
{$ELSE}
|
||
var
|
||
TV: TimeVal;
|
||
d: double;
|
||
begin
|
||
d := (newdt - UnixDateDelta) * 86400;
|
||
TV.tv_sec := trunc(d);
|
||
TV.tv_usec := trunc(frac(d) * 1000000);
|
||
Result := fpsettimeofday(@TV, nil) <> -1;
|
||
{$ENDIF}
|
||
{$ENDIF}
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
{$IFNDEF MSWINDOWS}
|
||
function GetTick: LongWord;
|
||
var
|
||
Stamp: TTimeStamp;
|
||
begin
|
||
Stamp := DateTimeToTimeStamp(Now);
|
||
Result := Stamp.Time;
|
||
end;
|
||
{$ELSE}
|
||
function GetTick: LongWord;
|
||
var
|
||
tick, freq: TLargeInteger;
|
||
{$IFDEF VER100}
|
||
x: TLargeInteger;
|
||
{$ENDIF}
|
||
begin
|
||
if Windows.QueryPerformanceFrequency(freq) then
|
||
begin
|
||
Windows.QueryPerformanceCounter(tick);
|
||
{$IFDEF VER100}
|
||
x.QuadPart := (tick.QuadPart / freq.QuadPart) * 1000;
|
||
Result := x.LowPart;
|
||
{$ELSE}
|
||
Result := Trunc((tick / freq) * 1000) and High(LongWord)
|
||
{$ENDIF}
|
||
end
|
||
else
|
||
Result := Windows.GetTickCount;
|
||
end;
|
||
{$ENDIF}
|
||
|
||
{==============================================================================}
|
||
|
||
function TickDelta(TickOld, TickNew: LongWord): LongWord;
|
||
begin
|
||
//if DWord is signed type (older Deplhi),
|
||
// then it not work properly on differencies larger then maxint!
|
||
Result := 0;
|
||
if TickOld <> TickNew then
|
||
begin
|
||
if TickNew < TickOld then
|
||
begin
|
||
TickNew := TickNew + LongWord(MaxInt) + 1;
|
||
TickOld := TickOld + LongWord(MaxInt) + 1;
|
||
end;
|
||
Result := TickNew - TickOld;
|
||
if TickNew < TickOld then
|
||
if Result > 0 then
|
||
Result := 0 - Result;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function CodeInt(Value: Word): Ansistring;
|
||
begin
|
||
setlength(result, 2);
|
||
result[1] := AnsiChar(Value div 256);
|
||
result[2] := AnsiChar(Value mod 256);
|
||
// Result := AnsiChar(Value div 256) + AnsiChar(Value mod 256)
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function DecodeInt(const Value: Ansistring; Index: Integer): Word;
|
||
var
|
||
x, y: Byte;
|
||
begin
|
||
if Length(Value) > Index then
|
||
x := Ord(Value[Index])
|
||
else
|
||
x := 0;
|
||
if Length(Value) >= (Index + 1) then
|
||
y := Ord(Value[Index + 1])
|
||
else
|
||
y := 0;
|
||
Result := x * 256 + y;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function CodeLongInt(Value: Longint): Ansistring;
|
||
var
|
||
x, y: word;
|
||
begin
|
||
// this is fix for negative numbers on systems where longint = integer
|
||
x := (Value shr 16) and integer($ffff);
|
||
y := Value and integer($ffff);
|
||
setlength(result, 4);
|
||
result[1] := AnsiChar(x div 256);
|
||
result[2] := AnsiChar(x mod 256);
|
||
result[3] := AnsiChar(y div 256);
|
||
result[4] := AnsiChar(y mod 256);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function DecodeLongInt(const Value: Ansistring; Index: Integer): LongInt;
|
||
var
|
||
x, y: Byte;
|
||
xl, yl: Byte;
|
||
begin
|
||
if Length(Value) > Index then
|
||
x := Ord(Value[Index])
|
||
else
|
||
x := 0;
|
||
if Length(Value) >= (Index + 1) then
|
||
y := Ord(Value[Index + 1])
|
||
else
|
||
y := 0;
|
||
if Length(Value) >= (Index + 2) then
|
||
xl := Ord(Value[Index + 2])
|
||
else
|
||
xl := 0;
|
||
if Length(Value) >= (Index + 3) then
|
||
yl := Ord(Value[Index + 3])
|
||
else
|
||
yl := 0;
|
||
Result := ((x * 256 + y) * 65536) + (xl * 256 + yl);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function DumpStr(const Buffer: Ansistring): string;
|
||
var
|
||
n: Integer;
|
||
begin
|
||
Result := '';
|
||
for n := 1 to Length(Buffer) do
|
||
Result := Result + ' +#$' + IntToHex(Ord(Buffer[n]), 2);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function DumpExStr(const Buffer: Ansistring): string;
|
||
var
|
||
n: Integer;
|
||
x: Byte;
|
||
begin
|
||
Result := '';
|
||
for n := 1 to Length(Buffer) do
|
||
begin
|
||
x := Ord(Buffer[n]);
|
||
if x in [65..90, 97..122] then
|
||
Result := Result + ' +''' + char(x) + ''''
|
||
else
|
||
Result := Result + ' +#$' + IntToHex(Ord(Buffer[n]), 2);
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
procedure Dump(const Buffer: AnsiString; DumpFile: string);
|
||
var
|
||
f: Text;
|
||
begin
|
||
AssignFile(f, DumpFile);
|
||
if FileExists(DumpFile) then
|
||
DeleteFile(DumpFile);
|
||
Rewrite(f);
|
||
try
|
||
Writeln(f, DumpStr(Buffer));
|
||
finally
|
||
CloseFile(f);
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
procedure DumpEx(const Buffer: AnsiString; DumpFile: string);
|
||
var
|
||
f: Text;
|
||
begin
|
||
AssignFile(f, DumpFile);
|
||
if FileExists(DumpFile) then
|
||
DeleteFile(DumpFile);
|
||
Rewrite(f);
|
||
try
|
||
Writeln(f, DumpExStr(Buffer));
|
||
finally
|
||
CloseFile(f);
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function TrimSPLeft(const S: string): string;
|
||
var
|
||
I, L: Integer;
|
||
begin
|
||
Result := '';
|
||
if S = '' then
|
||
Exit;
|
||
L := Length(S);
|
||
I := 1;
|
||
while (I <= L) and (S[I] = ' ') do
|
||
Inc(I);
|
||
Result := Copy(S, I, Maxint);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function TrimSPRight(const S: string): string;
|
||
var
|
||
I: Integer;
|
||
begin
|
||
Result := '';
|
||
if S = '' then
|
||
Exit;
|
||
I := Length(S);
|
||
while (I > 0) and (S[I] = ' ') do
|
||
Dec(I);
|
||
Result := Copy(S, 1, I);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function TrimSP(const S: string): string;
|
||
begin
|
||
Result := TrimSPLeft(s);
|
||
Result := TrimSPRight(Result);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function SeparateLeft(const Value, Delimiter: string): string;
|
||
var
|
||
x: Integer;
|
||
begin
|
||
x := Pos(Delimiter, Value);
|
||
if x < 1 then
|
||
Result := Value
|
||
else
|
||
Result := Copy(Value, 1, x - 1);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function SeparateRight(const Value, Delimiter: string): string;
|
||
var
|
||
x: Integer;
|
||
begin
|
||
x := Pos(Delimiter, Value);
|
||
if x > 0 then
|
||
x := x + Length(Delimiter) - 1;
|
||
Result := Copy(Value, x + 1, Length(Value) - x);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function GetParameter(const Value, Parameter: string): string;
|
||
var
|
||
s: string;
|
||
v: string;
|
||
begin
|
||
Result := '';
|
||
v := Value;
|
||
while v <> '' do
|
||
begin
|
||
s := Trim(FetchEx(v, ';', '"'));
|
||
if Pos(Uppercase(parameter), Uppercase(s)) = 1 then
|
||
begin
|
||
Delete(s, 1, Length(Parameter));
|
||
s := Trim(s);
|
||
if s = '' then
|
||
Break;
|
||
if s[1] = '=' then
|
||
begin
|
||
Result := Trim(SeparateRight(s, '='));
|
||
Result := UnquoteStr(Result, '"');
|
||
break;
|
||
end;
|
||
end;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
procedure ParseParametersEx(Value, Delimiter: string; const Parameters: TStrings);
|
||
var
|
||
s: string;
|
||
begin
|
||
Parameters.Clear;
|
||
while Value <> '' do
|
||
begin
|
||
s := Trim(FetchEx(Value, Delimiter, '"'));
|
||
Parameters.Add(s);
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
procedure ParseParameters(Value: string; const Parameters: TStrings);
|
||
begin
|
||
ParseParametersEx(Value, ';', Parameters);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function IndexByBegin(Value: string; const List: TStrings): integer;
|
||
var
|
||
n: integer;
|
||
s: string;
|
||
begin
|
||
Result := -1;
|
||
Value := uppercase(Value);
|
||
for n := 0 to List.Count -1 do
|
||
begin
|
||
s := UpperCase(List[n]);
|
||
if Pos(Value, s) = 1 then
|
||
begin
|
||
Result := n;
|
||
Break;
|
||
end;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function GetEmailAddr(const Value: string): string;
|
||
var
|
||
s: string;
|
||
begin
|
||
s := SeparateRight(Value, '<');
|
||
s := SeparateLeft(s, '>');
|
||
Result := Trim(s);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function GetEmailDesc(Value: string): string;
|
||
var
|
||
s: string;
|
||
begin
|
||
Value := Trim(Value);
|
||
s := SeparateRight(Value, '"');
|
||
if s <> Value then
|
||
s := SeparateLeft(s, '"')
|
||
else
|
||
begin
|
||
s := SeparateLeft(Value, '<');
|
||
if s = Value then
|
||
begin
|
||
s := SeparateRight(Value, '(');
|
||
if s <> Value then
|
||
s := SeparateLeft(s, ')')
|
||
else
|
||
s := '';
|
||
end;
|
||
end;
|
||
Result := Trim(s);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function StrToHex(const Value: Ansistring): string;
|
||
var
|
||
n: Integer;
|
||
begin
|
||
Result := '';
|
||
for n := 1 to Length(Value) do
|
||
Result := Result + IntToHex(Byte(Value[n]), 2);
|
||
Result := LowerCase(Result);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function IntToBin(Value: Integer; Digits: Byte): string;
|
||
var
|
||
x, y, n: Integer;
|
||
begin
|
||
Result := '';
|
||
x := Value;
|
||
repeat
|
||
y := x mod 2;
|
||
x := x div 2;
|
||
if y > 0 then
|
||
Result := '1' + Result
|
||
else
|
||
Result := '0' + Result;
|
||
until x = 0;
|
||
x := Length(Result);
|
||
for n := x to Digits - 1 do
|
||
Result := '0' + Result;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function BinToInt(const Value: string): Integer;
|
||
var
|
||
n: Integer;
|
||
begin
|
||
Result := 0;
|
||
for n := 1 to Length(Value) do
|
||
begin
|
||
if Value[n] = '0' then
|
||
Result := Result * 2
|
||
else
|
||
if Value[n] = '1' then
|
||
Result := Result * 2 + 1
|
||
else
|
||
Break;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function ParseURL(URL: string; var Prot, User, Pass, Host, Port, Path,
|
||
Para: string): string;
|
||
var
|
||
x, y: Integer;
|
||
sURL: string;
|
||
s: string;
|
||
s1, s2: string;
|
||
begin
|
||
Prot := 'http';
|
||
User := '';
|
||
Pass := '';
|
||
Port := '80';
|
||
Para := '';
|
||
|
||
x := Pos('://', URL);
|
||
if x > 0 then
|
||
begin
|
||
Prot := SeparateLeft(URL, '://');
|
||
sURL := SeparateRight(URL, '://');
|
||
end
|
||
else
|
||
sURL := URL;
|
||
if UpperCase(Prot) = 'HTTPS' then
|
||
Port := '443';
|
||
if UpperCase(Prot) = 'FTP' then
|
||
Port := '21';
|
||
x := Pos('@', sURL);
|
||
y := Pos('/', sURL);
|
||
if (x > 0) and ((x < y) or (y < 1))then
|
||
begin
|
||
s := SeparateLeft(sURL, '@');
|
||
sURL := SeparateRight(sURL, '@');
|
||
x := Pos(':', s);
|
||
if x > 0 then
|
||
begin
|
||
User := SeparateLeft(s, ':');
|
||
Pass := SeparateRight(s, ':');
|
||
end
|
||
else
|
||
User := s;
|
||
end;
|
||
x := Pos('/', sURL);
|
||
if x > 0 then
|
||
begin
|
||
s1 := SeparateLeft(sURL, '/');
|
||
s2 := SeparateRight(sURL, '/');
|
||
end
|
||
else
|
||
begin
|
||
s1 := sURL;
|
||
s2 := '';
|
||
end;
|
||
if Pos('[', s1) = 1 then
|
||
begin
|
||
Host := Separateleft(s1, ']');
|
||
Delete(Host, 1, 1);
|
||
s1 := SeparateRight(s1, ']');
|
||
if Pos(':', s1) = 1 then
|
||
Port := SeparateRight(s1, ':');
|
||
end
|
||
else
|
||
begin
|
||
x := Pos(':', s1);
|
||
if x > 0 then
|
||
begin
|
||
Host := SeparateLeft(s1, ':');
|
||
Port := SeparateRight(s1, ':');
|
||
end
|
||
else
|
||
Host := s1;
|
||
end;
|
||
Result := '/' + s2;
|
||
x := Pos('?', s2);
|
||
if x > 0 then
|
||
begin
|
||
Path := '/' + SeparateLeft(s2, '?');
|
||
Para := SeparateRight(s2, '?');
|
||
end
|
||
else
|
||
Path := '/' + s2;
|
||
if Host = '' then
|
||
Host := 'localhost';
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function ReplaceString(Value, Search, Replace: AnsiString): AnsiString;
|
||
var
|
||
x, l, ls, lr: Integer;
|
||
begin
|
||
if (Value = '') or (Search = '') then
|
||
begin
|
||
Result := Value;
|
||
Exit;
|
||
end;
|
||
ls := Length(Search);
|
||
lr := Length(Replace);
|
||
Result := '';
|
||
x := Pos(Search, Value);
|
||
while x > 0 do
|
||
begin
|
||
{$IFNDEF CIL}
|
||
l := Length(Result);
|
||
SetLength(Result, l + x - 1);
|
||
Move(Pointer(Value)^, Pointer(@Result[l + 1])^, x - 1);
|
||
{$ELSE}
|
||
Result:=Result+Copy(Value,1,x-1);
|
||
{$ENDIF}
|
||
{$IFNDEF CIL}
|
||
l := Length(Result);
|
||
SetLength(Result, l + lr);
|
||
Move(Pointer(Replace)^, Pointer(@Result[l + 1])^, lr);
|
||
{$ELSE}
|
||
Result:=Result+Replace;
|
||
{$ENDIF}
|
||
Delete(Value, 1, x - 1 + ls);
|
||
x := Pos(Search, Value);
|
||
end;
|
||
Result := Result + Value;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function RPosEx(const Sub, Value: string; From: integer): Integer;
|
||
var
|
||
n: Integer;
|
||
l: Integer;
|
||
begin
|
||
result := 0;
|
||
l := Length(Sub);
|
||
for n := From - l + 1 downto 1 do
|
||
begin
|
||
if Copy(Value, n, l) = Sub then
|
||
begin
|
||
result := n;
|
||
break;
|
||
end;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function RPos(const Sub, Value: String): Integer;
|
||
begin
|
||
Result := RPosEx(Sub, Value, Length(Value));
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function FetchBin(var Value: string; const Delimiter: string): string;
|
||
var
|
||
s: string;
|
||
begin
|
||
Result := SeparateLeft(Value, Delimiter);
|
||
s := SeparateRight(Value, Delimiter);
|
||
if s = Value then
|
||
Value := ''
|
||
else
|
||
Value := s;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function Fetch(var Value: string; const Delimiter: string): string;
|
||
begin
|
||
Result := FetchBin(Value, Delimiter);
|
||
Result := TrimSP(Result);
|
||
Value := TrimSP(Value);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function FetchEx(var Value: string; const Delimiter, Quotation: string): string;
|
||
var
|
||
b: Boolean;
|
||
begin
|
||
Result := '';
|
||
b := False;
|
||
while Length(Value) > 0 do
|
||
begin
|
||
if b then
|
||
begin
|
||
if Pos(Quotation, Value) = 1 then
|
||
b := False;
|
||
Result := Result + Value[1];
|
||
Delete(Value, 1, 1);
|
||
end
|
||
else
|
||
begin
|
||
if Pos(Delimiter, Value) = 1 then
|
||
begin
|
||
Delete(Value, 1, Length(delimiter));
|
||
break;
|
||
end;
|
||
b := Pos(Quotation, Value) = 1;
|
||
Result := Result + Value[1];
|
||
Delete(Value, 1, 1);
|
||
end;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function IsBinaryString(const Value: AnsiString): Boolean;
|
||
var
|
||
n: integer;
|
||
begin
|
||
Result := False;
|
||
for n := 1 to Length(Value) do
|
||
if Value[n] in [#0..#8, #10..#31] then
|
||
//ignore null-terminated strings
|
||
if not ((n = Length(value)) and (Value[n] = AnsiChar(#0))) then
|
||
begin
|
||
Result := True;
|
||
Break;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function PosCRLF(const Value: AnsiString; var Terminator: AnsiString): integer;
|
||
var
|
||
n, l: integer;
|
||
begin
|
||
Result := -1;
|
||
Terminator := '';
|
||
l := length(value);
|
||
for n := 1 to l do
|
||
if value[n] in [#$0d, #$0a] then
|
||
begin
|
||
Result := n;
|
||
Terminator := Value[n];
|
||
if n <> l then
|
||
case value[n] of
|
||
#$0d:
|
||
if value[n + 1] = #$0a then
|
||
Terminator := #$0d + #$0a;
|
||
#$0a:
|
||
if value[n + 1] = #$0d then
|
||
Terminator := #$0a + #$0d;
|
||
end;
|
||
Break;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
Procedure StringsTrim(const Value: TStrings);
|
||
var
|
||
n: integer;
|
||
begin
|
||
for n := Value.Count - 1 downto 0 do
|
||
if Value[n] = '' then
|
||
Value.Delete(n)
|
||
else
|
||
Break;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function PosFrom(const SubStr, Value: String; From: integer): integer;
|
||
var
|
||
ls,lv: integer;
|
||
begin
|
||
Result := 0;
|
||
ls := Length(SubStr);
|
||
lv := Length(Value);
|
||
if (ls = 0) or (lv = 0) then
|
||
Exit;
|
||
if From < 1 then
|
||
From := 1;
|
||
while (ls + from - 1) <= (lv) do
|
||
begin
|
||
{$IFNDEF CIL}
|
||
if CompareMem(@SubStr[1],@Value[from],ls) then
|
||
{$ELSE}
|
||
if SubStr = copy(Value, from, ls) then
|
||
{$ENDIF}
|
||
begin
|
||
result := from;
|
||
break;
|
||
end
|
||
else
|
||
inc(from);
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
{$IFNDEF CIL}
|
||
function IncPoint(const p: pointer; Value: integer): pointer;
|
||
begin
|
||
Result := PAnsiChar(p) + Value;
|
||
end;
|
||
{$ENDIF}
|
||
|
||
{==============================================================================}
|
||
//improved by 'DoggyDawg'
|
||
function GetBetween(const PairBegin, PairEnd, Value: string): string;
|
||
var
|
||
n: integer;
|
||
x: integer;
|
||
s: string;
|
||
lenBegin: integer;
|
||
lenEnd: integer;
|
||
str: string;
|
||
max: integer;
|
||
begin
|
||
lenBegin := Length(PairBegin);
|
||
lenEnd := Length(PairEnd);
|
||
n := Length(Value);
|
||
if (Value = PairBegin + PairEnd) then
|
||
begin
|
||
Result := '';//nothing between
|
||
exit;
|
||
end;
|
||
if (n < lenBegin + lenEnd) then
|
||
begin
|
||
Result := Value;
|
||
exit;
|
||
end;
|
||
s := SeparateRight(Value, PairBegin);
|
||
if (s = Value) then
|
||
begin
|
||
Result := Value;
|
||
exit;
|
||
end;
|
||
n := Pos(PairEnd, s);
|
||
if (n = 0) then
|
||
begin
|
||
Result := Value;
|
||
exit;
|
||
end;
|
||
Result := '';
|
||
x := 1;
|
||
max := Length(s) - lenEnd + 1;
|
||
for n := 1 to max do
|
||
begin
|
||
str := copy(s, n, lenEnd);
|
||
if (str = PairEnd) then
|
||
begin
|
||
Dec(x);
|
||
if (x <= 0) then
|
||
Break;
|
||
end;
|
||
str := copy(s, n, lenBegin);
|
||
if (str = PairBegin) then
|
||
Inc(x);
|
||
Result := Result + s[n];
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function CountOfChar(const Value: string; Chr: char): integer;
|
||
var
|
||
n: integer;
|
||
begin
|
||
Result := 0;
|
||
for n := 1 to Length(Value) do
|
||
if Value[n] = chr then
|
||
Inc(Result);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
// ! do not use AnsiExtractQuotedStr, it's very buggy and can crash application!
|
||
function UnquoteStr(const Value: string; Quote: Char): string;
|
||
var
|
||
n: integer;
|
||
inq, dq: Boolean;
|
||
c, cn: char;
|
||
begin
|
||
Result := '';
|
||
if Value = '' then
|
||
Exit;
|
||
if Value = Quote + Quote then
|
||
Exit;
|
||
inq := False;
|
||
dq := False;
|
||
for n := 1 to Length(Value) do
|
||
begin
|
||
c := Value[n];
|
||
if n <> Length(Value) then
|
||
cn := Value[n + 1]
|
||
else
|
||
cn := #0;
|
||
if c = quote then
|
||
if dq then
|
||
dq := False
|
||
else
|
||
if not inq then
|
||
inq := True
|
||
else
|
||
if cn = quote then
|
||
begin
|
||
Result := Result + Quote;
|
||
dq := True;
|
||
end
|
||
else
|
||
inq := False
|
||
else
|
||
Result := Result + c;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function QuoteStr(const Value: string; Quote: Char): string;
|
||
var
|
||
n: integer;
|
||
begin
|
||
Result := '';
|
||
for n := 1 to length(value) do
|
||
begin
|
||
Result := result + Value[n];
|
||
if value[n] = Quote then
|
||
Result := Result + Quote;
|
||
end;
|
||
Result := Quote + Result + Quote;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
procedure HeadersToList(const Value: TStrings);
|
||
var
|
||
n, x, y: integer;
|
||
s: string;
|
||
begin
|
||
for n := 0 to Value.Count -1 do
|
||
begin
|
||
s := Value[n];
|
||
x := Pos(':', s);
|
||
if x > 0 then
|
||
begin
|
||
y:= Pos('=',s);
|
||
if not ((y > 0) and (y < x)) then
|
||
begin
|
||
s[x] := '=';
|
||
Value[n] := s;
|
||
end;
|
||
end;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
procedure ListToHeaders(const Value: TStrings);
|
||
var
|
||
n, x: integer;
|
||
s: string;
|
||
begin
|
||
for n := 0 to Value.Count -1 do
|
||
begin
|
||
s := Value[n];
|
||
x := Pos('=', s);
|
||
if x > 0 then
|
||
begin
|
||
s[x] := ':';
|
||
Value[n] := s;
|
||
end;
|
||
end;
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function SwapBytes(Value: integer): integer;
|
||
var
|
||
s: AnsiString;
|
||
x, y, xl, yl: Byte;
|
||
begin
|
||
s := CodeLongInt(Value);
|
||
x := Ord(s[4]);
|
||
y := Ord(s[3]);
|
||
xl := Ord(s[2]);
|
||
yl := Ord(s[1]);
|
||
Result := ((x * 256 + y) * 65536) + (xl * 256 + yl);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function ReadStrFromStream(const Stream: TStream; len: integer): AnsiString;
|
||
var
|
||
x: integer;
|
||
{$IFDEF CIL}
|
||
buf: Array of Byte;
|
||
{$ENDIF}
|
||
begin
|
||
{$IFDEF CIL}
|
||
Setlength(buf, Len);
|
||
x := Stream.read(buf, Len);
|
||
SetLength(buf, x);
|
||
Result := StringOf(Buf);
|
||
{$ELSE}
|
||
Setlength(Result, Len);
|
||
x := Stream.read(PAnsiChar(Result)^, Len);
|
||
SetLength(Result, x);
|
||
{$ENDIF}
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
procedure WriteStrToStream(const Stream: TStream; Value: AnsiString);
|
||
{$IFDEF CIL}
|
||
var
|
||
buf: Array of Byte;
|
||
{$ENDIF}
|
||
begin
|
||
{$IFDEF CIL}
|
||
buf := BytesOf(Value);
|
||
Stream.Write(buf,length(Value));
|
||
{$ELSE}
|
||
Stream.Write(PAnsiChar(Value)^, Length(Value));
|
||
{$ENDIF}
|
||
end;
|
||
|
||
{==============================================================================}
|
||
function GetTempFile(const Dir, prefix: AnsiString): AnsiString;
|
||
{$IFNDEF FPC}
|
||
{$IFDEF MSWINDOWS}
|
||
var
|
||
Path: AnsiString;
|
||
x: integer;
|
||
{$ENDIF}
|
||
{$ENDIF}
|
||
begin
|
||
{$IFDEF FPC}
|
||
Result := GetTempFileName(Dir, Prefix);
|
||
{$ELSE}
|
||
{$IFNDEF MSWINDOWS}
|
||
Result := tempnam(Pointer(Dir), Pointer(prefix));
|
||
{$ELSE}
|
||
{$IFDEF CIL}
|
||
Result := System.IO.Path.GetTempFileName;
|
||
{$ELSE}
|
||
if Dir = '' then
|
||
begin
|
||
SetLength(Path, MAX_PATH);
|
||
x := GetTempPath(Length(Path), PChar(Path));
|
||
SetLength(Path, x);
|
||
end
|
||
else
|
||
Path := Dir;
|
||
x := Length(Path);
|
||
if Path[x] <> '\' then
|
||
Path := Path + '\';
|
||
SetLength(Result, MAX_PATH + 1);
|
||
GetTempFileName(PChar(Path), PChar(Prefix), 0, PChar(Result));
|
||
Result := PChar(Result);
|
||
SetFileattributes(PChar(Result), GetFileAttributes(PChar(Result)) or FILE_ATTRIBUTE_TEMPORARY);
|
||
{$ENDIF}
|
||
{$ENDIF}
|
||
{$ENDIF}
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function PadString(const Value: AnsiString; len: integer; Pad: AnsiChar): AnsiString;
|
||
begin
|
||
if length(value) >= len then
|
||
Result := Copy(value, 1, len)
|
||
else
|
||
Result := Value + StringOfChar(Pad, len - length(value));
|
||
end;
|
||
|
||
{==============================================================================}
|
||
|
||
function NormalizeHeader(Value: TStrings; var Index: Integer): string;
|
||
var
|
||
s, t: string;
|
||
n: Integer;
|
||
begin
|
||
s := Value[Index];
|
||
Inc(Index);
|
||
if s <> '' then
|
||
while (Value.Count - 1) > Index do
|
||
begin
|
||
t := Value[Index];
|
||
if t = '' then
|
||
Break;
|
||
for n := 1 to Length(t) do
|
||
if t[n] = #9 then
|
||
t[n] := ' ';
|
||
if not(AnsiChar(t[1]) in [' ', '"', ':', '=']) then
|
||
Break
|
||
else
|
||
begin
|
||
s := s + ' ' + Trim(t);
|
||
Inc(Index);
|
||
end;
|
||
end;
|
||
Result := TrimRight(s);
|
||
end;
|
||
|
||
{==============================================================================}
|
||
var
|
||
n: integer;
|
||
begin
|
||
for n := 1 to 12 do
|
||
begin
|
||
CustomMonthNames[n] := ShortMonthNames[n];
|
||
MyMonthNames[0, n] := ShortMonthNames[n];
|
||
end;
|
||
end.
|