问题
I understand that in order to represent unions in C# I need to use StructLayout[LayoutKind.Explicit)] and [FieldOffset(x)] attribut to specify the byte offset inside the union. However, I have a following union I want to represent and FieldOffset attrib only offset by size of a byte.
union _myUnion
{
unsigned int info;
struct
{
unsigned int flag1:1 // bit 0
unsigned int flag2:1 // bit 1
unsigned int flag3:1 // bit 2
unsigned int flag4:1 // bit 3
unsigned int flag5:1 // bit 4
unsigned int flag6:1 // bit 5
.
.
.
unsigned int flag31:1 // bit 31
}
}
As you can see for the inner struct in the union, I can't use FieldOffset since I need something that can offset by a bit.
Is there a solution to this? I am trying to call a DLL function and one of the data struct was defined as such and I ran out of ideas on how to best represent this union struct.
回答1:
No need for union there; one field+property for the data, 8 properties that do bitwise "shift" operations, for example:
public uint Value {get;set;}
public uint Flag2 {
get { return Value >> 2; }
}
etc. I would also have thought you want bool here?
Normally I'd say: don't make mutable structs. PInvoke may (I'm not sure) be a valid scenario for that, so I'll ignore it :)
If the value is genuinely using more than 32 bits, consider switching the backing field to ulong.
回答2:
Most elegant solution for this is using flags enum
Better use uint instead int
[Flags]
public enum MyEnum
: uint
{
None=0,
Flag1=1,
Flag2=1<<1,
Flag3=1<<2,
// etc
Flag32=1<<31
}
After it you can use int as enum and as uint
MyEnum value=MyEnum.Flag1|MyEnum.Flag2;
uint uintValue=(uint)value;
// uintValue=3
This will be normally marshal by PInvoke
回答3:
Yes, you can do this. You are on the right path, the answer lies in the use of BitVector32 along with the FieldOffset and StructLayout attributes. However, there are a number of things you need to keep in mind when doing this:
You need to explicitly specify the size of the variables that will contain the data in question. This first item is very important to pay attention to. In your question above, for example, you specify info as unsigned int. What size is unsigned int? 32 bits? 64 bits? That depends on the version of the OS where this particular version of .NET is running (this may be .NET Core, Mono, or Win32/Win64).
What 'endianness ' or bit order is this? Again, we may be running on any type of hardware (think Mobile/Xamarin, not just laptop or tablet) -- as such, you cannot assume Intel bit order.
We will want to avoid any memory management that is language dependent, or in C/C++ parlance POD (plain old data) types. That will mean sticking to value types only.
I'm going to make the assumption, based on your question and the specification of flags 0-31, that sizeof(int) == 32.
The trick then is to ensure the following:
- All data is byte aligned.
- The bitfields and info field align on the same byte boundary.
Here is how we accomplish that:
[StructLayout(LayoutKind.Explicit, Size = 1, CharSet = CharSet.Ansi)]
public struct MyUnion
{
#region Lifetime
/// <summary>
/// Ctor
/// </summary>
/// <param name="foo"></param>
public MyUnion(int foo)
{
// allocate the bitfield
info = new BitVector32(0);
// initialize bitfield sections
flag1 = BitVector32.CreateSection(1);
flag2 = BitVector32.CreateSection(1, flag1);
flag3 = BitVector32.CreateSection(1, flag2);
}
#endregion
#region Bifield
// Creates and initializes a BitVector32.
[FieldOffset(0)]
private BitVector32 info;
#endregion
#region Bitfield sections
/// <summary>
/// Section - Flag 1
/// </summary>
private static BitVector32.Section flag1;
/// <summary>
/// Section - Flag 2
/// </summary>
private static BitVector32.Section flag2;
/// <summary>
/// Section - Flag 3
/// </summary>
private static BitVector32.Section flag3;
#endregion
#region Properties
/// <summary>
/// Flag 1
/// </summary>
public bool Flag1
{
get { return info[flag1] != 0; }
set { info[flag1] = value ? 1 : 0; }
}
/// <summary>
/// Flag 2
/// </summary>
public bool Flag2
{
get { return info[flag2] != 0; }
set { info[flag2] = value ? 1 : 0; }
}
/// <summary>
/// Flag 1
/// </summary>
public bool Flag3
{
get { return info[flag3] != 0; }
set { info[flag3] = value ? 1 : 0; }
}
#endregion
#region ToString
/// <summary>
/// Allows us to represent this in human readable form
/// </summary>
/// <returns></returns>
public override string ToString()
{
return $"Name: {nameof(MyUnion)}{Environment.NewLine}Flag1: {Flag1}: Flag2: {Flag2} Flag3: {Flag3} {Environment.NewLine}BitVector32: {info}{Environment.NewLine}";
}
#endregion
}
Pay particular attention to the constructor. By definition, you cannot define default constructor for structs in C#. However, we need some way to ensure that the BitVector32 object and its sections are initialized properly before use. We accomplish that by requiring a constructor that takes a dummy integer parameter, and initialize the object like this:
/// <summary>
/// Main entry point
/// </summary>
/// <param name="args"></param>
static void Main(string[] args)
{
// brew up one of these...
var myUnion = new MyUnion(0)
{
Flag2 = true
};
By the way, you aren't by any means limited to single bitfields -- you can define any size bitfield you like. For example, if I were to change your example to be:
union _myUnion
{
unsigned int info;
struct
{
unsigned int flag1 : 3 // bit 0-2
unsigned int flag2 : 1 // bit 3
unsigned int flag3 : 4 // bit 4-7
.
.
.
unsigned int flag31 : 1 // bit 31
}
}
I would just change my class to be:
[StructLayout(LayoutKind.Explicit, Size = 1, CharSet = CharSet.Ansi)]
public struct MyUnion2
{
#region Lifetime
/// <summary>
/// Ctor
/// </summary>
/// <param name="foo"></param>
public MyUnion2(int foo)
{
// allocate the bitfield
info = new BitVector32(0);
// initialize bitfield sections
flag1 = BitVector32.CreateSection(0x07);
flag2 = BitVector32.CreateSection(1, flag1);
flag3 = BitVector32.CreateSection(0x0f, flag2);
}
#endregion
#region Bifield
// Creates and initializes a BitVector32.
[FieldOffset(0)]
private BitVector32 info;
#endregion
#region Bitfield sections
/// <summary>
/// Section - Flag1
/// </summary>
private static BitVector32.Section flag1;
/// <summary>
/// Section - Flag2
/// </summary>
private static BitVector32.Section flag2;
/// <summary>
/// Section - Flag3
/// </summary>
private static BitVector32.Section flag3;
#endregion
#region Properties
/// <summary>
/// Flag 1
/// </summary>
public int Flag1
{
get { return info[flag1]; }
set { info[flag1] = value; }
}
/// <summary>
/// Flag 2
/// </summary>
public bool Flag2
{
get { return info[flag2] != 0; }
set { info[flag2] = value ? 1 : 0; }
}
/// <summary>
/// Flag 1
/// </summary>
public int Flag3
{
get { return info[flag3]; }
set { info[flag3] = value; }
}
#endregion
#region ToString
/// <summary>
/// Allows us to represent this in human readable form
/// </summary>
/// <returns></returns>
public override string ToString()
{
return $"Name: {nameof(MyUnion2)}{Environment.NewLine}Flag1: {Flag1}: Flag2: {Flag2} Flag3: {Flag3} {Environment.NewLine}BitVector32: {info}{Environment.NewLine}";
}
#endregion
}
One last word about this topic... it should be obvious that this should only be done if you absolutely MUST do this. Clearly, this requires specialized knowledge of your OS environment, the language you are running in, your calling convention, and a host of other brittle requirements.
In any other context there are so many code smells here that this clearly screams of non-portability. But from the context of your question, I would surmise the whole point is that you need to run close to the hardware and need this kind of precision.
Caveat emptor!
来源：https://stackoverflow.com/questions/4830978/representing-union-bitfields-using-cs-struclayout-and-fieldoffset