How to simulate bit-fields in Delphi records?

Question

I would like to declare a record in Delphi that contains the same layout as it has in C.

For those interested : This record is part of a union in the Windows OS's LDT_ENTRY record. (I need to use this record in Delphi because I'm working on an Xbox emulator in Delphi - see project Dxbx on sourceforge).

Anyway, the record in question is defined as:

    struct
    {
        DWORD   BaseMid : 8;
        DWORD   Type : 5;
        DWORD   Dpl : 2;
        DWORD   Pres : 1;
        DWORD   LimitHi : 4;
        DWORD   Sys : 1;
        DWORD   Reserved_0 : 1;
        DWORD   Default_Big : 1;
        DWORD   Granularity : 1;
        DWORD   BaseHi : 8;
    }
    Bits;

As far as I know, there are no bit-fields possible in Delphi. I did try this:

 Bits = record
      BaseMid: Byte; // 8 bits
      _Type: 0..31; // 5 bits
      Dpl: 0..3; // 2 bits
      Pres: Boolean; // 1 bit
      LimitHi: 0..15; // 4 bits
      Sys: Boolean; // 1 bit
      Reserved_0: Boolean; // 1 bit
      Default_Big: Boolean; // 1 bit
      Granularity: Boolean; // 1 bit
      BaseHi: Byte; // 8 bits
  end;

But alas: it's size becomes 10 bytes, instead of the expected 4. I would like to know how I should declare the record, so that I get a record with the same layout, the same size, and the same members. Preferrably without loads of getter/setters.

TIA.

Answer 1

Well, you basically need to get down to the dirty with bit-manipulation.

Why, specifically, do you need to retain that structure?

If you only need to talk to a legacy program that either talks in this dialect (TCP/IP or similar), or stores data in this manner (files, etc.), then I would map a normal Delphi structure to a bit-version compatible. In other words, I would use a normally structured Delphi structure in memory, and write code to write and read that structure in a compatible manner.

If you need to save memory, I would make getters and setters that manipulate bits of internal integers or similar. This will have a performance impact, but not much more than what the original C program would have, the only difference is that the bit-manipulation would be added by compiler magic in the C version, whereas you will have to write it yourself.

If you don't have many records in memory, and don't need to talk to another program, I'd use a natural Delphi structure. Trade-off for higher performance will be more memory used.

But it all depends on your criteria.

In any case, you won't be able to talk the Delphi compiler into doing the same job for you as the C compiler.

PACKED RECORD, suggested by another here, doesn't do that, and was never meant to. It will only remove alignment padding to put integers on 32-bit boundaries and similar, but won't pack multiple fields into one byte.

Note that a common way to do this is through Delphi SETS, which are implementing internally using bit-fields. Again, you will have different code than the C variant.

Answer 2

Rudy's Delphi Corner is the best resource I know of regarding Delphi and C/C++ interoperability. His Pitfalls of conversion is pretty much a must read when using C/C++ APIs in Delphi. The chapter you'll be most interested in is Records and alignment -> Bitfields, but I urge you to read the entire thing top to bottom, twice. The other articles are definitely worth the time investment, too.

Answer 3

Ok, my bit manipulation is a bit rusty, so I could have reversed the bytes. But the code below gives the general idea:

type
  TBits = record
  private
    FBaseMid     : Byte;
    FTypeDplPres :  Byte;
    FLimitHiSysEa: Byte;
    FBaseHi      : Byte;

    function GetType: Byte;
    procedure SetType(const AType: Byte);
    function GetDpl: Byte;
    procedure SetDbl(const ADpl: Byte);
    function GetBit1(const AIndex: Integer): Boolean;
    procedure SetBit1(const AIndex: Integer; const AValue: Boolean);
    function GetLimitHi: Byte;
    procedure SetLimitHi(const AValue: Byte);
    function GetBit2(const AIndex: Integer): Boolean;
    procedure SetBit2(const AIndex: Integer; const AValue: Boolean);

  public
    property BaseMid: Byte read FBaseMid write FBaseMid;
    property &Type: Byte read GetType write SetType; // 0..31
    property Dpl: Byte read GetDpl write SetDbl; // 0..3
    property Pres: Boolean index 128 read GetBit1 write SetBit1; 
    property LimitHi: Byte read GetLimitHi write SetLimitHi; // 0..15

    property Sys: Boolean index 16 read GetBit2 write SetBit2; 
    property Reserved0: Boolean index 32 read GetBit2 write SetBit2; 
    property DefaultBig: Boolean index 64 read GetBit2 write SetBit2; 
    property Granularity: Boolean index 128 read GetBit2 write SetBit2; 
    property BaseHi: Byte read FBaseHi write FBaseHi;
  end;

  function TBits.GetType: Byte;
  begin
    Result := (FTypeDplPres shr 3) and $1F;
  end;

  procedure TBits.SetType(const AType: Byte);
  begin
    FTypeDplPres := (FTypeDplPres and $07) + ((AType and $1F) shr 3);
  end;

  function TBits.GetDpl: Byte;
  begin
    Result := (FTypeDplPres and $06) shr 1;
  end;

  procedure TBits.SetDbl(const ADpl: Byte);
  begin
    FTypeDblPres := (FTypeDblPres and $F9) + ((ADpl and $3) shl 1);
  end;

  function TBits.GetBit1(const AIndex: Integer): Boolean;
  begin
    Result := FTypeDplPres and AIndex = AIndex;
  end;

  procedure TBits.SetBit1(const AIndex: Integer; const AValue: Boolean);
  begin
    if AValue then
      FTypeDblPres := FTypeDblPres or AIndex
    else
      FTypeDblPres := FTypeDblPres and not AIndex;
  end;

  function TBits.GetLimitHi: Byte;
  begin
    Result := (FLimitHiSysEa shr 4) and $0F;
  end;

  procedure TBits.SetLimitHi(const AValue: Byte);
  begin
    FLimitHiSysEa := (FLimitHiSysEa and $0F) + ((AValue and $0F) shr 4);
  end;

  function TBits.GetBit2(const AIndex: Integer): Boolean;
  begin
    Result := FLimitHiSysEa and AIndex = AIndex;
  end;

  procedure TBits.SetBit2(const AIndex: Integer; const AValue: Boolean);
  begin
    if AValue then
      FLimitHiSysEa := FLimitHiSysEa or AIndex
    else
      FLimitHiSysEa := FLimitHiSysEa and not AIndex;
  end;

Answer 4

Thanks everyone!

Based on this information, I reduced this to :

RBits = record
public
  BaseMid: BYTE;
private
  Flags: WORD;
  function GetBits(const aIndex: Integer): Integer;
  procedure SetBits(const aIndex: Integer; const aValue: Integer);
public
  BaseHi: BYTE;
  property _Type: Integer index $0005 read GetBits write SetBits; // 5 bits at offset 0
  property Dpl: Integer index $0502 read GetBits write SetBits; // 2 bits at offset 5
  property Pres: Integer index $0701 read GetBits write SetBits; // 1 bit at offset 7
  property LimitHi: Integer index $0804 read GetBits write SetBits; // 4 bits at offset 8
  property Sys: Integer index $0C01 read GetBits write SetBits; // 1 bit at offset 12
  property Reserved_0: Integer index $0D01 read GetBits write SetBits; // 1 bit at offset 13
  property Default_Big: Integer index $0E01 read GetBits write SetBits; // 1 bit at offset 14
  property Granularity: Integer index $0F01 read GetBits write SetBits; // 1 bit at offset 15
end;

The index is encoded as follows : (BitOffset shl 8) + NrBits. Where 1<=NrBits<=32 and 0<=BitOffset<=31

Now, I can get and set these bits as follows :

{$OPTIMIZATION ON}
{$OVERFLOWCHECKS OFF}
function RBits.GetBits(const aIndex: Integer): Integer;
var
  Offset: Integer;
  NrBits: Integer;
  Mask: Integer;
begin
  NrBits := aIndex and $FF;
  Offset := aIndex shr 8;

  Mask := ((1 shl NrBits) - 1);

  Result := (Flags shr Offset) and Mask;
end;

procedure RBits.SetBits(const aIndex: Integer; const aValue: Integer);
var
  Offset: Integer;
  NrBits: Integer;
  Mask: Integer;
begin
  NrBits := aIndex and $FF;
  Offset := aIndex shr 8;

  Mask := ((1 shl NrBits) - 1);
  Assert(aValue <= Mask);

  Flags := (Flags and (not (Mask shl Offset))) or (aValue shl Offset);
end;

Pretty nifty, don't you think?!?!

PS: Rudy Velthuis now included a revised version of this in his excellent "Pitfalls of converting"-article.