[jogibear9988]

Binary ArrayIndex emission leaves the array reference on the stack while evaluating its index. Save and reload the receiver when an exception region can occur during emission, so a try/finally index retains the array and evaluation order. Share the aggressively inlined EmitLoadLeftAndRight helper with arithmetic emission: restore a saved left operand beneath a right operand, saving the right first only when it is still on the stack. Existing local types, allocation order and operand order are preserved.

Vector and non-vector arrays with lower bounds -2, 0 and 1; receiver/index/finally side effects. Arithmetic regressions also check subtraction results and left/right/finally effects for int, long, decimal and their nullable forms, including null left operands, against System compilation. Regressions use strict FEC compilation for both Standard and LightExpression.

Validation

Release solution build: zero warnings/errors.

Runtime	TestX groups	Legacy tests
.NET 10	467 passed	1,756 passed
.NET 8	467 passed	1,755 passed
.NET Framework 4.7.2	446 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.

net8.0 results (master → change)

API / workload	Compile ns	Invoke ns	Compile B/op	Invoke B/op
Standard / add_control	1186.4 → 1187.4	2.0 → 2.0	872.0 → 872.0	0.0 → 0.0
Standard / binary_index_unrelated_try	1683.7 → 1925.1	2.0 → 2.0	1472.0 → 1472.0	0.0 → 0.0
Standard / subtract_try	1474.2 → 1574.5	1.9 → 1.9	1472.0 → 1472.0	0.0 → 0.0
Standard / nullable_subtract_try	3397.7 → 3413.5	2.2 → 2.2	2656.0 → 2656.0	0.0 → 0.0
Standard / nullable_subtract_try_null	4512.0 → 3996.7	2.2 → 2.2	2656.0 → 2656.0	0.0 → 0.0
Standard / nullable_add	3204.1 → 3270.5	2.5 → 2.5	1928.1 → 1928.1	0.0 → 0.0
Standard / nullable_add_null	3194.9 → 3130.9	2.4 → 2.4	1928.1 → 1928.1	0.0 → 0.0
Standard / vector_array_read	1244.8 → 1295.6	1.8 → 1.8	872.0 → 872.0	0.0 → 0.0
Light / add_control	1053.7 → 1103.3	1.9 → 2.0	872.0 → 872.0	0.0 → 0.0
Light / binary_index_unrelated_try	1401.6 → 1533.0	2.0 → 2.1	1472.0 → 1472.0	0.0 → 0.0
Light / subtract_try	1396.2 → 1561.1	2.0 → 2.0	1472.0 → 1472.0	0.0 → 0.0
Light / nullable_subtract_try	3129.2 → 2850.4	2.2 → 2.2	2656.0 → 2656.0	0.0 → 0.0
Light / nullable_subtract_try_null	2979.7 → 3038.2	2.3 → 2.3	2656.0 → 2656.0	0.0 → 0.0
Light / nullable_add	3135.4 → 3256.1	2.7 → 2.8	1928.0 → 1928.0	0.0 → 0.0
Light / nullable_add_null	2817.3 → 2881.6	2.5 → 2.4	1928.0 → 1928.0	0.0 → 0.0
Light / vector_array_read	1109.6 → 990.8	1.8 → 1.8	872.0 → 872.0	0.0 → 0.0

net10.0 results (master → change)

API / workload	Compile ns	Invoke ns	Compile B/op	Invoke B/op
Standard / add_control	4181.2 → 4208.7	2.3 → 2.3	872.1 → 872.1	0.0 → 0.0
Standard / binary_index_unrelated_try	4252.3 → 4872.7	2.3 → 2.3	1472.1 → 1472.1	0.0 → 0.0
Standard / subtract_try	9554.6 → 8988.9	2.3 → 2.3	1472.1 → 1472.1	0.0 → 0.0
Standard / nullable_subtract_try	9779.6 → 9953.6	2.2 → 2.4	2656.1 → 2656.1	0.0 → 0.0
Standard / nullable_subtract_try_null	10685.2 → 10190.9	2.0 → 2.0	2656.1 → 2656.1	0.0 → 0.0
Standard / nullable_add	9611.3 → 9574.6	2.3 → 2.3	1928.0 → 1928.0	0.0 → 0.0
Standard / nullable_add_null	9480.3 → 9161.8	2.3 → 2.2	1928.0 → 1928.0	0.0 → 0.0
Standard / vector_array_read	3151.1 → 3396.0	2.0 → 1.8	872.0 → 872.0	0.0 → 0.0
Light / add_control	3629.0 → 3082.5	2.0 → 2.0	872.0 → 872.0	0.0 → 0.0
Light / binary_index_unrelated_try	3494.4 → 3616.5	2.0 → 2.0	1472.0 → 1472.0	0.0 → 0.0
Light / subtract_try	3591.1 → 3922.6	2.0 → 2.0	1472.0 → 1472.0	0.0 → 0.0
Light / nullable_subtract_try	8789.2 → 8837.9	2.2 → 2.1	2656.1 → 2656.1	0.0 → 0.0
Light / nullable_subtract_try_null	9406.8 → 9029.0	2.0 → 2.0	2656.1 → 2656.1	0.0 → 0.0
Light / nullable_add	8932.2 → 9130.1	2.3 → 2.4	1928.0 → 1928.0	0.0 → 0.0
Light / nullable_add_null	11487.0 → 9502.8	2.3 → 2.3	1928.0 → 1928.0	0.0 → 0.0
Light / vector_array_read	3115.6 → 3208.4	2.1 → 1.9	872.0 → 872.0	0.0 → 0.0

Array-index spilling remains conservative: a try elsewhere in the lambda can also activate the save/reload path. This accounts for the IL difference in binary_index_unrelated_try.

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