C# Property pattern vs. type pattern

[deep-outcome]

Let discuss

private record IntRecord ( int IntProperty ) { }
static IntRecord rec = new IntRecord(1);

private static void IL1 ()
{
   _ = rec switch
  {
    { IntProperty: 1 } => 'A', 	 	
  };
}

private static void IL2 ()
{
  _ = rec switch
  { 	 	
    IntRecord r when r.IntProperty == 1 => 'B', 	 	 
  };
}

and their corresponding ILs

.method /* 06000009 */ private hidebysig static 
	void IL1 () cil managed 
{
	// Method begins at RVA 0x257c
	// Code size 48 (0x30)
	.maxstack 2
	.locals /* 11000004 */ init (
		[0] class   …✖✖✖/IntRecord,
		[1] int32
	)

	// sequence point: (line 170, col 5) to (line 170, col 6) in .cs
	IL_0000: nop
	// sequence point: (line 171, col 7) to (line 174, col 9) in   …❔❔❔.cs
	IL_0001: ldsfld class   …✖✖✖/IntRecord   …✖✖✖::rec /* 04000003 */
	IL_0006: stloc.0
	IL_0007: ldc.i4.1
	IL_0008: brtrue.s IL_000b

	// sequence point: (line 171, col 15) to (line 174, col 8) in   …❔❔❔.cs
	IL_000a: nop

	// sequence point: hidden
	IL_000b: ldloc.0
	IL_000c: brfalse.s IL_0020

	IL_000e: ldloc.0
	IL_000f: callvirt instance int32   …✖✖✖/IntRecord::get_IntProperty() /* 0600007B */
	IL_0014: stloc.1
	// sequence point: hidden
	IL_0015: ldloc.1
	IL_0016: ldc.i4.1
	IL_0017: beq.s IL_001b

	IL_0019: br.s IL_0020

	// sequence point: (line 173, col 31) to (line 173, col 34) in   …❔❔❔.cs
	IL_001b: ldc.i4.s 65
	IL_001d: pop
	IL_001e: br.s IL_002b

	IL_0020: ldc.i4.1
	IL_0021: brtrue.s IL_0024

	// sequence point: (line 171, col 15) to (line 174, col 8) in   …❔❔❔.cs
	IL_0023: nop

	// sequence point: hidden
	IL_0024: ldloc.0
	IL_0025: newobj instance void [System.Runtime]System.Runtime.CompilerServices.SwitchExpressionException::.ctor(object) /* 0A000025 */
	IL_002a: throw

	// sequence point: hidden
	IL_002b: ldc.i4.1
	IL_002c: brtrue.s IL_002f

	// sequence point: (line 171, col 7) to (line 174, col 9) in   …❔❔❔.cs
	IL_002e: nop

	// sequence point: (line 175, col 5) to (line 175, col 6) in   …❔❔❔.cs
	IL_002f: ret
} // end of method EnhancedPatternMatching::IL1

.method /* 0600000A */ private hidebysig static 
	void IL2 () cil managed 
{
	// Method begins at RVA 0x25b8
	// Code size 48 (0x30)
	.maxstack 2
	.locals /* 11000005 */ init (
		[0] class   …✖✖✖/IntRecord r
	)

	// sequence point: (line 178, col 5) to (line 178, col 6) in …❔❔❔.cs
	IL_0000: nop
	// sequence point: (line 179, col 7) to (line 182, col 9) in …❔❔❔.cs
	IL_0001: ldsfld class   …✖✖✖/IntRecord   …✖✖✖::rec /* 04000003 */
	IL_0006: stloc.0
	IL_0007: ldc.i4.1
	IL_0008: brtrue.s IL_000b

	// sequence point: (line 179, col 15) to (line 182, col 8) in …❔❔❔.cs
	IL_000a: nop

	// sequence point: hidden
	IL_000b: ldloc.0
	IL_000c: brtrue.s IL_0010

	IL_000e: br.s IL_0020

	// sequence point: (line 181, col 21) to (line 181, col 44) in …❔❔❔.cs
	IL_0010: ldloc.0
	IL_0011: callvirt instance int32   …✖✖✖/IntRecord::get_IntProperty() /* 0600007B */
	IL_0016: ldc.i4.1
	IL_0017: beq.s IL_001b

	// sequence point: hidden
	IL_0019: br.s IL_0020

	// sequence point: (line 181, col 48) to (line 181, col 51) in …❔❔❔.cs
	IL_001b: ldc.i4.s 66
	IL_001d: pop
	IL_001e: br.s IL_002b

	IL_0020: ldc.i4.1
	IL_0021: brtrue.s IL_0024

	// sequence point: (line 179, col 15) to (line 182, col 8) in …❔❔❔.cs
	IL_0023: nop

	// sequence point: hidden
	IL_0024: ldloc.0
	IL_0025: newobj instance void [System.Runtime]System.Runtime.CompilerServices.SwitchExpressionException::.ctor(object) /* 0A000025 */
	IL_002a: throw

	// sequence point: hidden
	IL_002b: ldc.i4.1
	IL_002c: brtrue.s IL_002f

	// sequence point: (line 179, col 7) to (line 182, col 9) in …❔❔❔.cs
	IL_002e: nop

	// sequence point: (line 183, col 5) to (line 183, col 6) in …❔❔❔.cs
	IL_002f: ret
} // end of method EnhancedPatternMatching::IL2

which differ singificantly only in

Does diff (namely no-use pop/load to/from local var) result out of unoptimized implementation of Property Pattern or it follows some rules arising purely from the syntax – i.e. the IL2 explicitly uses the == operator.

[huoyaoyuan]

unnecessary stloc/ldloc pair should usually be suboptimal codegen.

[orthoxerox]

Should this have been posted to dotnet/roslyn instead? Codegen is the compiler's task.

[deep-outcome]

Likely.

[yaakov-h]

The IL you've posted doesn't match latest SharpLab, so I suspect it is already out of date.

On SharpLab, both functions produce identical IL patterns in Release configuration.