Windows – Win32：如何将字符串转换为日期？

class function TDateTimeUtils.TryStrToDateExact(const S,DateFormat: string; out Value: TDateTime): Boolean;
begin
{
    Assume Microsoft's de-facto standard for y2k fixup: 2029
        1930-2029
}
    Result := TDateTimeUtils.TryStrToDateExact(S,DateFormat,2029,{out}Value);
end;

class function TDateTimeUtils.TryStrToDateExact(const S,DateFormat: string; PivotYear: Integer;
        out Value: TDateTime): Boolean;
var
    Month,Day,Year: Integer;
    Tokens: TStringDynArray;
    CurrentToken: string;
    i,n: Integer;
    Partial: string;
    MaxValue: Integer;
    nCurrentYear: Integer;

    function GetCurrentYear: Word;
    var
        y,m,d: Word;
    begin
        DecodeDate(Now,y,d);
        Result := y;
    end;
begin
    Result := False;
{
    M/dd/yy

    Valid pictures codes are

        d       Day of the month as digits without leading zeros for single-digit days.
        dd      Day of the month as digits with leading zeros for single-digit days.
        ddd Abbreviated day of the week as specified by a LOCALE_SABBREVDAYNAME* value,for example,"Mon" in English (United States).
                Windows Vista and later: If a short version of the day of the week is required,your application should use the LOCALE_SSHORTESTDAYNAME* constants.
        dddd    Day of the week as specified by a LOCALE_SDAYNAME* value.

        M       Month as digits without leading zeros for single-digit months.
        MM      Month as digits with leading zeros for single-digit months.
        MMM Abbreviated month as specified by a LOCALE_SABBREVMONTHNAME* value,"Nov" in English (United States).
        MMMM    Month as specified by a LOCALE_SMONTHNAME* value,"November" for English (United States),and "Noviembre" for Spanish (Spain).

        y       Year represented only by the last digit.
        yy      Year represented only by the last two digits. A leading zero is added for single-digit years.
        yyyy    Year represented by a full four or five digits,depending on the calendar used. Thai Buddhist and Korean calendars have five-digit years. The "yyyy" pattern shows five digits for these two calendars,and four digits for all other supported calendars. Calendars that have single-digit or two-digit years,such as for the Japanese Emperor era,are represented differently. A single-digit year is represented with a leading zero,"03". A two-digit year is represented with two digits,"13". No additional leading zeros are displayed.
        yyyyy   Behaves identically to "yyyy".

        g,gg   Period/era string formatted as specified by the CAL_SERASTRING value.
                The "g" and "gg" format pictures in a date string are ignored if there is no associated era or period string.


        PivotYear
                The maximum year that a 1 or 2 digit year is assumed to be.
                The Microsoft de-factor standard for y2k is 2029. Any value greater
                than 29 is assumed to be 1930 or higher.

                e.g. 2029:
                    1930,...,2000,2001,2029

                If the PivotYear is between 0 and 99,then PivotYear is assumed to be
                a date range in the future. e.g. (assuming this is currently 2010):

                    Pivot   Range
                    0       1911..2010  (no future years)
                    1       1912..2011
                    ...
                    98      2009..2108
                    99      2010..2099  (no past years)

                0 ==> no years in the future
                99 ==> no years in the past
}
    if Length(S) = 0 then
        Exit;
    if Length(DateFormat) = 0 then
        Exit;

    Month := -1;
    Day := -1;
    Year := -1;

    Tokens := TDateTimeUtils.TokenizeFormat(DateFormat);
    n := 1; //input string index
    for i := Low(Tokens) to High(Tokens) do
    begin
        CurrentToken := Tokens[i];
        if CurrentToken = 'MMMM' then
        begin
            //Long month names,we don't support yet (you're free to write it)
            Exit;
        end
        else if CurrentToken = 'MMM' then
        begin
            //Short month names,we don't support yet (you're free to write it)
            Exit;
        end
        else if CurrentToken = 'MM' then
        begin
            //Month,with leading zero if needed
            if not ReadDigitString(S,n,2{MinDigits},2{MaxDigits},1{MinValue},12{MaxValue},{var}Month) then Exit;
        end
        else if CurrentToken = 'M' then
        begin
            //months
            if not ReadDigitString(S,1{MinDigits},{var}Month) then Exit;
        end
        else if CurrentToken = 'dddd' then
        begin
            Exit; //Long day names,we don't support yet (you're free to write it)
        end
        else if CurrentToken = 'ddd' then
        begin
            Exit; //Short day names,we don't support yet (you're free to write it);
        end
        else if CurrentToken = 'dd' then
        begin
            //If we know what month it is,and even better if we know what year it is,limit the number of valid days to that
            if (Month >= 1) and (Month <= 12) then
            begin
                if Year > 0 then
                    MaxValue := MonthDays[IsLeapYear(Year),Month]
                else
                    MaxValue := MonthDays[True,Month]; //we don't know the year,assume it's a leap year to be more generous
            end
            else
                MaxValue := 31; //we don't know the month,so assume it's the largest

            if not ReadDigitString(S,MaxValue{MaxValue},{var}Day) then Exit;
        end
        else if CurrentToken = 'd' then
        begin
            //days
            //If we know what month it is,{var}Day) then Exit;
        end
        else if (CurrentToken = 'yyyy') or (CurrentToken = 'yyyyy') then
        begin
            //Year represented by a full four or five digits,depending on the calendar used.
            {
                Thai Buddhist and Korean calendars have five-digit years.
                The "yyyy" pattern shows five digits for these two calendars,and four digits for all other supported calendars.
                Calendars that have single-digit or two-digit years,such as for
                    the Japanese Emperor era,are represented differently.
                    A single-digit year is represented with a leading zero,for
                    example,"13". No additional leading zeros are displayed.
            }
            if not ReadDigitString(S,4{MinDigits},4{MaxDigits},0{MinValue},9999{MaxValue},{var}Year) then Exit;
        end
        else if CurrentToken = 'yyy' then
        begin
            //i'm not sure what this would look like,so i'll ignore it
            Exit;
        end
        else if CurrentToken = 'yy' then
        begin
            //Year represented only by the last two digits. A leading zero is added for single-digit years.
            if not ReadDigitString(S,99{MaxValue},{var}Year) then Exit;

            nCurrentYear := GetCurrentYear;
            Year := (nCurrentYear div 100 * 100)+Year;

            if (PivotYear < 100) and (PivotYear >= 0) then
            begin
                //assume pivotyear is a delta from this year,not an absolute value
                PivotYear := nCurrentYear+PivotYear;
            end;

            //Check the pivot year value
            if Year > PivotYear then
                Year := Year - 100;
        end
        else if CurrentToken = 'y' then
        begin
            //Year represented only by the last digit.
            if not ReadDigitString(S,1{MaxDigits},9{MaxValue},{var}Year) then Exit;

            nCurrentYear := GetCurrentYear;
            Year := (nCurrentYear div 10 * 10)+Year;

            if (PivotYear < 100) and (PivotYear >= 0) then
            begin
                //assume pivotyear is a delta from this year,not an absolute value
                PivotYear := nCurrentYear+PivotYear;
            end;

            //Check the pivot year value
            if Year > PivotYear then
                Year := Year - 100;
        end
        else
        begin
            //The input string should contains CurrentToken starting at n
            Partial := Copy(S,Length(CurrentToken));
            Inc(n,Length(CurrentToken));
            if Partial <> CurrentToken then
                Exit;
        end;
    end;

    //If there's still stuff left over in the string,then it's not valid
    if n <> Length(s)+1 then
    begin
        Result := False;
        Exit;
    end;

    if Day > MonthDays[IsLeapYear(Year),Month] then
    begin
        Result := False;
        Exit;
    end;

    try
        Value := EncodeDate(Year,Day);
    except
        Result := False;
        Exit;
    end;
    Result := True;
end;

class function TDateTimeUtils.TokenizeFormat(fmt: string): TStringDynArray;
var
    i: Integer;
    partial: string;

    function IsDateFormatPicture(ch: AnsiChar): Boolean;
    begin
        case ch of
        'M','d','y': Result := True;
        else Result := False;
        end;
    end;
begin
    SetLength(Result,0);

    if Length(fmt) = 0 then
        Exit;

    //format is only one character long? If so then that's the tokenized entry
    if Length(fmt)=1 then
    begin
        SetLength(Result,1);
        Result[0] := fmt;
    end;

    partial := fmt[1];
    i := 2;
    while i <= Length(fmt) do
    begin
        //If the characters in partial are a format picture,and the character in fmt is not the same picture code then write partial to result,and reset partial
        if IsDateFormatPicture(partial[1]) then
        begin
            //if the current fmt character is different than the running partial picture
            if (partial[1] <> fmt[i]) then
            begin
                //Move the current partial to the output
                //and start a new partial
                SetLength(Result,Length(Result)+1);
                Result[High(Result)] := partial;
                Partial := fmt[i];
            end
            else
            begin
                //the current fmt character is more of the same format picture in partial
                //Add it to the partial
                Partial := Partial + fmt[i];
            end;
        end
        else
        begin
            //The running partial is not a format picture.
            //If the current fmt character is a picture code,then write out the partial and start a new partial
            if IsDateFormatPicture(fmt[i]) then
            begin
                //Move the current partial to the output
                //and start a new partial
                SetLength(Result,Length(Result)+1);
                Result[High(Result)] := partial;
                Partial := fmt[i];
            end
            else
            begin
                //The current fmt character is another non-picture code. Add it to the running partial
                Partial := Partial + fmt[i];
            end;
        end;

        Inc(i);
        Continue;
    end;

    //If we have a running partial,then add it to the output
    if partial <> '' then
    begin
        SetLength(Result,Length(Result)+1);
        Result[High(Result)] := partial;
    end;
end;

class function TDateTimeUtils.ReadDigitString(const S: string; var Pos: Integer;
            MinDigits,MaxDigits: Integer; MinValue,MaxValue: Integer;
            var Number: Integer): Boolean;
var
    Digits: Integer;
    Value: Integer;
    Partial: string;
    CandidateNumber: Integer;
    CandidateDigits: Integer;
begin
    Result := False;
    CandidateNumber := -1;
    CandidateDigits := 0;

    Digits := MinDigits;
    while Digits <= MaxDigits do
    begin
        Partial := Copy(S,Pos,Digits);
        if Length(Partial) < Digits then
        begin
            //we couldn't get all we wanted. We're done; use whatever we've gotten already
            Break;
        end;

        //Check that it's still a number
        if not TryStrToInt(Partial,Value) then
            Break;

        //Check that it's not too big - meaning that getting anymore wouldn't work
        if (Value > MaxValue) then
            Break;

        if (Value >= MinValue) then
        begin
            //Hmm,looks good. Keep it as our best possibility
            CandidateNumber := Value;
            CandidateDigits := Digits;
        end;

        Inc(Digits); //try to be greedy,grabbing even *MORE* digits
    end;

    if (CandidateNumber >= 0) or (CandidateDigits > 0) then
    begin
        Inc(Pos,CandidateDigits);
        Number := CandidateNumber;
        Result := True;
    end;
end;