[fpc-pascal] Extended float to single and pchar types

[Urbansound]

Please ignore this.  It is an FPC mailing list error, repeated from August


----- Original Message ----- 
From: "Urbansound" <urbansound at sbcglobal.net>
To: <fpc-pascal at lists.freepascal.org>
Cc: "Matt Emson" <memsom at interalpha.co.uk>
Sent: Tuesday, August 02, 2005 1:41 PM
Subject: Re: [fpc-pascal] Extended float to single and pchar types


> Thanks Matt.
>
> See replies within the body please.
>
> ----- Original Message ----- 
> From: "Matt Emson" <memsom at interalpha.co.uk>
> To: "Urbansound" <urbansound at sbcglobal.net>; "FPC-Pascal users 
> discussions"
> <fpc-pascal at lists.freepascal.org>
> Sent: Tuesday, August 02, 2005 11:26 AM
> Subject: Re: [fpc-pascal] Extended float to single and pchar types
>
>
>>> Hi,
>>>
>>> I have a stdcall reference to a c based dll. called target.dll
>>> In c the setup is ...
>>
>> Are you sure it is stdcall? Quite often it's the calling convention that
>> messes up the calling of a C DLL. Could it be safecall or cdecl? Which
>> compiler made the DLL?
>
> Tried all three and results are the same always, (which I find odd,
> considering mangling)
>
>>
>>> float Get_l1_BidPriceSingle(void* p);
>>>
>>> The Pascal reference is...
>>>
>>> function Get_l1_BidPrice(p : pointer) : single;
>>> stdcall; external 'target.dll' name 'Get_l1_BidPrice';
>>
>> Which all looks okay.
>>
>>> The end result is to allow a usage in win32 as ...
>>> SendMessage(listbox, LB_ADDSTRING, 0, longint(Pchar(s)));
>>
>> Hmmm.. this is not a good idea if 's' is of type string. If 's' is a
>> string
>> (AnsiString or whatever) you should probably call SetLength on it, and 
>> I'd
>> also advise passing a pointer to the first char too, rather relying on 
>> the
>> cast.
>>
>> E.g.
>>
>>  s := 'Test data';
>>  SetLength(s, Length(s));
>>  SendMessage(listbox, LB_ADDSTRING, 0, longint(@(s[1])));
>
> Working only by virtue of the non error checking method using
> Str(bid_price:5:3,s);
> See example at bottom.
>
>> That probably has more chance of working, but still assumes you are not
>> crossing process boundaries. If threads are involved, it's not going to
>> work.
>
> Single thread, confirmed
>
>>> The c reference is  a 'Single' float, but each time I make refernece to
>> data
>>> conversion, errors tell me that the reference to the call is returning a
>> 10
>>> bit "Extended" float which fails, appearing to be a default autotype
>>> conversions to an extended 10 byte float, in fpc.
>>
>> If it is a "reference", it should probably be a pointer. Is the C 
>> function
>> really returning the resultant by value, or by reference?
>>
>>> Converted data as an "Ansistring" seems to fail any type of conversion,
>>> as
>> a
>>> runtime 207 crash.
>>
>> Convert how? FloatToStr should handle any floating point type.
>
> Apparently that is a problem when the float is in a callback.
>
>>
>> e.g.
>>
>> var
>>  f: single;
>>  d: double;
>>  e: extended;
>> begin
>>  f := 10.01;
>>  d := 10.01;
>>  e := 10.01;
>>
>>  messagebox(0, PChar(FloatToStr( .. ), 0); // replace '..' with f, d or e
>> end;
>
> Crashes when incorporated into the callback.
>
>>> If I store Str conversions to a "shortstring", I can format the fp and
>>> convert it to a fixed decimal rounded output, xxx.yyy, but using 
>>> length()
>>> direct or pointer access, still pulls the 0 byte and extra characters of
>> the
>>> string into any other converted output, requiring array managment and
>>> parsing.
>>
>> var
>>   s: string;
>>   f: single;
>> begin
>>  f := 123.45678; //this will round
>>  s := format('%3.3f', [f]); //think sprintf
>> end;
>
> Crashes same when f holds the value returned in the callback.
>
>>> Is there a way to take an "extended" FP and covert it ultimately to a
>> Pchar
>>> that would allow messaging to win32 list boxes or similar to a litteral
>>> equivelent behaviour, without having to run through all the symantics of
>>> array management?
>>>
>>> OR
>>>
>>> Can one circumvent the auto-typing to extended, so that single or double
>>> would behave to conversion methods like FloatToStr() that I could then
>> type
>>> over to pchar and serve the same purpose?
>>
>> Surely you could just use an extended and then cast it? So long as it is
>> possible to store the value in the variable you wish to, it'll work.
>>
>>> Clearly I'm not yet comfortable with string manipulation in Pascal, so 
>>> an
>>> example would go a long way toward getting me to where I can do the work
>>> myself.
>>
>> Well, it's an awful lot simpler that C until you have to interact with C
>> API
>> calls!!
>>
>> M
>
> Any ansistring causes a hard crash.
>
> Below will run without crashing, but again the string length is even cast 
> at
> 9, the max expected and equal white space being seed to the string first.
> You can see the image "output.jpg" attached, showing the residual string
> data, right of the values.  These are stock market symbols and their 
> related
> prices being quoted, which are the content coming back from the live API.
>
> Type shortstring = string[9];
> var
>    bid_price : extended; //or double, or single = same result
>    s : shortstring;
>
> begin
>     SendMessage(listbox, LB_ADDSTRING, 0,
>        longint(Get_secsym(quote)));
>    // quote data may range nn.nnn to nnnnn.nnn
>     bid_price := Get_l1_BidPrice(quote);
>     //s:=FloatToStr(bid_price);           //crashes 207
>     s:='         ';                                    // 9 bytes white
> space
>     Str(bid_price:5:3,s);                      //convert+ 5 places, 3
> decimal  = 6
>     SetLength(s, Length(s));
>     SendMessage(listbox, LB_ADDSTRING, 0, longint(@(s[1])));
> end;
>


--------------------------------------------------------------------------------


> _______________________________________________
> fpc-pascal maillist  -  fpc-pascal at lists.freepascal.org
> http://lists.freepascal.org/mailman/listinfo/fpc-pascal
>