perf: type-bound member accessors and no bool boxing (~10-33% faster object rendering)

[rexm]

Summary

Two targeted optimizations to the reflection-based member access path (the path used when rendering plain .NET objects, including anonymous types):

1. Type-bound member accessors. ObjectDescriptors are already cached per type, but the shared ReflectionMemberAccessor re-resolved its per-type accessor table on every member access via a type-keyed LookupSlim lookup. The descriptor's member accessor is now pre-bound to the described type at descriptor creation, so the common case (instance type matches the descriptor) goes straight to the member-name lookup. If the instance type doesn't match the bound type, it falls back to the original shared path — behavior is unchanged.

2. No boxing for bool property reads. The cached Func<object, object> getters box value-type property values on every read. bool properties — ubiquitous in {{#if}}/{{#unless}}-heavy templates — now return the cached BoxedValues.True/False instances instead of allocating a fresh box per read.

Dictionary-based data paths are untouched. All 1837 tests pass.

Benchmarks

MediumRun (LaunchCount=1, 15 iterations), Apple M4, .NET 10. Baseline is master @ 37ca7b9.

Object-path suites (the target of this change):

Benchmark	Case	master	This PR	Δ time	master alloc	PR alloc
RenderSimple	object	856.85 ns	575.79 ns	−32.8%	48 B	0 B
RenderNested	object, rows=5	6,764.33 ns	5,650.00 ns	−16.5%	1192 B	592 B
RenderNested	object, rows=20	24.52 us	18.94 us	−22.7%	4624 B	2224 B
RenderList	N=10, object	2,916.26 ns	2,585.00 ns	−11.4%	480 B	0 B
RenderList	N=100, object	27.48 us	22.77 us	−17.1%	6720 B	1920 B
RenderList	N=1000, object	259.14 us	236.41 us	−8.8%	71520 B	23520 B
RenderToString	clean	13.95 us	12.93 us	−7.3%	32.09 KB	30.91 KB
RenderToString	html	14.75 us	13.83 us	−6.2%	35.31 KB	34.14 KB

Full suite (including neutral paths):

All 30 benchmark cases

Benchmark	Case	master	This PR	Δ time	master alloc	PR alloc
Execution.CallHelperWithoutParameters	—	44.35 ns	44.08 ns	-0.6%	NA	NA
EndToEnd	N=5, dictionary	28.78 us	29.52 us	+2.6%	NA	NA
EndToEnd	N=5, object	29.49 us	28.38 us	-3.8%	NA	NA
RenderNested	dictionary, rows=5	5,413.42 ns	5,438.00 ns	+0.5%	544 B	544 B
RenderNested	object, rows=5	6,764.33 ns	5,650.00 ns	-16.5%	1192 B	592 B
RenderNested	dictionary, rows=20	18.54 us	18.14 us	-2.2%	2176 B	2176 B
RenderNested	object, rows=20	24.52 us	18.94 us	-22.7%	4624 B	2224 B
RenderList	N=10, dictionary	2,420.21 ns	2,435.00 ns	+0.6%	0 B	0 B
RenderList	N=10, object	2,916.26 ns	2,585.00 ns	-11.4%	480 B	0 B
RenderList	N=100, dictionary	23.42 us	23.20 us	-0.9%	1920 B	1920 B
RenderList	N=100, object	27.48 us	22.77 us	-17.1%	6720 B	1920 B
RenderList	N=1000, dictionary	231.70 us	235.50 us	+1.6%	23520 B	23520 B
RenderList	N=1000, object	259.14 us	236.41 us	-8.8%	71520 B	23520 B
LargeArray	N=20000	389.13 us	386.69 us	-0.6%	NA	NA
LargeArray	N=40000	801.86 us	777.20 us	-3.1%	NA	NA
LargeArray	N=80000	1.571 ms	1.561 ms	-0.7%	NA	NA
RenderSimple	dictionary	509.96 ns	501.56 ns	-1.6%	0 B	0 B
RenderSimple	expando	414.36 ns	410.34 ns	-1.0%	0 B	0 B
RenderSimple	object	856.85 ns	575.79 ns	-32.8%	48 B	0 B
Execution.CallHelperWithOneParameter	—	46.16 ns	45.68 ns	-1.0%	NA	NA
Execution.CallHelperWithTwoParameter	—	46.45 ns	47.37 ns	+2.0%	NA	NA
Execution.LateCallHelperWithoutParameters	—	44.11 ns	43.66 ns	-1.0%	NA	NA
Execution.LateCallHelperWithOneParameter	—	46.17 ns	45.70 ns	-1.0%	NA	NA
Execution.LateCallHelperWithTwoParameter	—	46.32 ns	47.71 ns	+3.0%	NA	NA
Execution.CallBlockHelperWithoutParameters	—	47.04 ns	47.44 ns	+0.9%	NA	NA
Execution.CallBlockHelperWithOneParameter	—	47.65 ns	47.47 ns	-0.4%	NA	NA
Execution.CallBlockHelperWithTwoParameter	—	48.15 ns	49.04 ns	+1.8%	NA	NA
RenderToString	clean	13.95 us	12.93 us	-7.3%	32.09 KB	31656 B
RenderToString	html	14.75 us	13.83 us	-6.2%	35.31 KB	34960 B

Dictionary/expando cases and helper-dispatch (Execution.*) benchmarks are neutral (within ±3% run-to-run noise), as expected — this change only affects the reflection member-access path.

Remaining RenderList allocations at larger N are iterator index boxing beyond the BoxedValues int cache (0–19) — identical for object and dictionary data, and a possible follow-up.

Note: includes the RenderToString benchmark suite commit (also submitted separately as #650) so the numbers above are reproducible from this branch.

🤖 Generated with Claude Code

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
84.5% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud