In Turbopascal it was possible with offs()

type
 rectype = record
  var1 : longint;
  var2 : longint;
  var3 : boolean;
 end;
var
 o : longint;
 recvar : rectype;
 o := offs( recvar.var1 );  // --> 0
 o := offs( recvar.var2 );  // --> 4
 o := offs( recvar.var3 );  // --> 8

o := PtrUInt( @recvar.var1 ) - PtrUInt( @recvar );
o := PtrIUnt( @recvar.var2 ) - PtrUInt( @recvar );
o := PtrInUt( @recvar.var3 ) - PtrUInt( @recvar );

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

o := PtrUInt( @recvar.var1 ) - PtrUInt( @recvar ); o := PtrIUnt( @recvar.var2 ) - PtrUInt( @recvar ); o := PtrInUt( @recvar.var3 ) - PtrUInt( @recvar );

Actually even
o := PtrUInt(@(rectype(nil^).var1)));
is possible.

var addrx : longint;
type
 rectype = record
  var1 : longint;
  var2 : longint;
  var3 : boolean;
 end;
 rectype_2 = record
  ofs1 : longint;
  ofs2 : longint;
  ofs3 : longint;
 end;
var
 recvar : rectype;
procedure t;
var o1 , o2 , o3 : longint;
begin
 // works
 o1 := ptruint( @recvar.var1 ) - ptruint( @recvar );
 o2 := ptruint( @recvar.var2 ) - ptruint( @recvar );
 o3 := ptruint( @recvar.var3 ) - ptruint( @recvar );
end;
const rec : rectype_2 = (
 // does not work! (illegal expression)
 ofs1 : ptruint( @recvar.var1 ) - ptruint( @recvar );
 ofs2 : ptruint( @recvar.var2 ) - ptruint( @recvar );
 ofs3 : ptruint( @recvar.var3 ) - ptruint( @recvar );
 );

That's incorrect. In TP, Ofs(recvar.var1) (with one "f") returns the offset in the segment that contains the "recvar" variable. It does not return the relative offset of the field inside the record type.

Jonas

That are 2 different things.

There was a function which gives the relative offset of the field I am very sure of that. And I am very sure that it was offs (wich 2 "f").

Jonas_______________________________________________ fpc-pascal maillist - http://lists.freepascal.org/mailman/listinfo/fpc-pascal

I'm quite sure it's not, see attachment. The only thing that existed was the "offset" operator, but it only worked in assembler blocks (and afaik it also works in FPC assembler blocks).

Jonas

Hello Jonas,

then it was the offset() method or somehow else... But it was possible.

This is ages ago, but IIRC getmem would allocate memory on segment boundaries (offset 0). In a dynamically allocated record ofs(record.member) (one f) would give you the offset to the start of record. This was obviously a special case. I'm not sure if large/huge memory model was part of it. Otherwise, conceptually, I can't imagine a function that would return an offset to the start of a structure by only giving a reference to a member of that structure. Even Florians solution contains the starting point: nil.

Ludo

But
const o : longint = PtrUInt(@(rectype(nil^).var1))); unfortunately is not possible.

But in at least some cases Ofs() was normalised to always contain 0..15, so it would be easy to (mis)interpret it as an offset into an allocated block.