[jogibear9988]

Interface property and indexer getters on struct receivers can produce invalid IL. Reference getters and collection Add methods also require callvirt to check null receivers even when the target method is nonvirtual. Use one receiver-aware dispatch helper for ordinary method calls, property reads, indexer reads and collection initialization. It emits direct calls for methods declared on the struct, constrained virtual calls for interface/base methods on structs, and virtual calls for reference receivers. Pop discarded property/indexer getter results while preserving their side effects.

Tests compare results and exceptions with System compilation: interface getters, ordinary interface/base calls, struct mutation, indexer reads, discarded results, null reference receivers, interface Add calls on structs and collection initialization through a null member. LightExpression collection trees are constructed directly because conversion from System ListInit is unsupported. Getter opcode expectations in three legacy issue fixtures are updated for the required callvirt instruction; their behavior assertions remain. Regressions use strict FEC compilation for both Standard and LightExpression.

Validation

Release solution build: zero warnings/errors.

Runtime	TestX groups	Legacy tests
.NET 10	475 passed	1,756 passed
.NET 8	475 passed	1,755 passed
.NET Framework 4.7.2	454 passed	1,639 passed

Measurements

Current master ce28e2b0 → this change, Release, Windows ARM64, .NET 8.0.28 / 10.0.9. Compilation uses prebuilt trees; runtime measures invocation through the same wrapper. Two fresh process rounds with reversed load order, seven samples per round, about 30 ms calibration, one logical CPU and tiered compilation disabled. Tables use the median of the two per-round medians.

The controls below produce correct results on master for their measured inputs. Newly supported trees have no valid before/after execution comparison because master rejects them, produces incorrect results or emits invalid IL. Timings include machine and JIT/code-placement variation; the unchanged add control helps show that variation. These observations do not prove universal performance neutrality.

The reference getter control uses a non-null receiver. Its new callvirt also supplies the previously missing null check, so any runtime cost of that correctness change is included.

net8.0 results (master → change)

API / workload	Compile ns	Invoke ns	Compile B/op	Invoke B/op
Standard / add_control	1649.8 → 1755.3	2.6 → 2.7	872.0 → 872.0	0.0 → 0.0
Standard / struct_read	1604.6 → 1562.8	2.9 → 2.6	904.1 → 904.1	0.0 → 0.0
Standard / struct_interface_call	2279.1 → 1432.0	3.6 → 3.3	936.1 → 936.1	0.0 → 0.0
Standard / nonvirtual_list_initializer	1412.0 → 1771.0	4.8 → 4.2	960.1 → 960.1	24.0 → 24.0
Standard / list_initializer	3854.1 → 2882.3	16.3 → 21.7	1008.0 → 1008.0	72.0 → 72.0
Standard / list_read	2884.6 → 3242.8	3.7 → 3.3	952.0 → 952.0	0.0 → 0.0
Standard / property_read	2189.9 → 2230.6	3.4 → 3.4	904.1 → 904.1	0.0 → 0.0
Standard / nullable_hasvalue	2294.1 → 2532.7	3.4 → 3.4	952.0 → 952.0	0.0 → 0.0
Light / add_control	1911.2 → 1712.2	2.6 → 2.6	872.0 → 872.0	0.0 → 0.0
Light / struct_read	1377.0 → 1446.2	2.5 → 2.6	904.0 → 904.0	0.0 → 0.0
Light / struct_interface_call	1960.4 → 1600.6	3.7 → 4.3	936.0 → 936.0	0.0 → 0.0
Light / nonvirtual_list_initializer	1371.3 → 1709.6	5.4 → 9.3	992.0 → 992.0	24.0 → 24.0
Light / list_initializer	4149.8 → 1779.1	91.8 → 45.6	1040.0 → 1040.0	72.0 → 72.0
Light / list_read	8908.4 → 4283.1	4.3 → 4.0	952.1 → 952.1	0.0 → 0.0
Light / property_read	15768.0 → 9652.4	14.8 → 19.2	904.0 → 904.0	0.0 → 0.0
Light / nullable_hasvalue	3209.7 → 4328.2	6.2 → 5.9	952.0 → 952.0	0.0 → 0.0

net10.0 results (master → change)

API / workload	Compile ns	Invoke ns	Compile B/op	Invoke B/op
Standard / add_control	5923.9 → 5652.0	7.3 → 12.8	872.1 → 872.1	0.0 → 0.0
Standard / struct_read	5313.3 → 10639.2	2.9 → 3.6	904.2 → 904.2	0.0 → 0.0
Standard / struct_interface_call	5653.4 → 5858.0	8.3 → 3.7	936.8 → 936.8	0.0 → 0.0
Standard / nonvirtual_list_initializer	5999.4 → 5691.1	7.7 → 6.9	960.3 → 960.3	24.0 → 24.0
Standard / list_initializer	5360.9 → 5028.9	76.6 → 66.5	1008.1 → 1008.1	72.0 → 72.0
Standard / list_read	5861.9 → 5856.4	2.6 → 2.8	952.1 → 952.1	0.0 → 0.0
Standard / property_read	5239.6 → 4768.8	2.7 → 2.7	904.2 → 904.2	0.0 → 0.0
Standard / nullable_hasvalue	5194.4 → 5126.3	2.7 → 2.8	952.0 → 952.0	0.0 → 0.0
Light / add_control	3998.7 → 3769.7	3.1 → 2.9	872.1 → 872.1	0.0 → 0.0
Light / struct_read	4647.2 → 4145.4	2.5 → 2.6	904.0 → 904.0	0.0 → 0.0
Light / struct_interface_call	3908.0 → 3832.1	2.4 → 2.6	936.0 → 936.0	0.0 → 0.0
Light / nonvirtual_list_initializer	3619.7 → 3436.2	5.0 → 4.7	992.0 → 992.0	24.0 → 24.0
Light / list_initializer	3361.6 → 3532.8	61.7 → 64.7	1040.0 → 1040.0	72.0 → 72.0
Light / list_read	3346.9 → 5806.3	3.5 → 3.4	952.0 → 952.0	0.0 → 0.0
Light / property_read	5553.0 → 5567.6	3.9 → 4.1	904.0 → 904.0	0.0 → 0.0
Light / nullable_hasvalue	4140.1 → 4360.7	3.6 → 4.0	952.0 → 952.0	0.0 → 0.0

Emitted IL differs in these measured workloads: nonvirtual_list_initializer, property_read.
No measured allocation increase above 0.1 B/op in these controls.