fix: honor InternalsVisibleTo when checking constructor accessibility

[vbreuss]

Awaiten used two different accessibility tests and only one of them was correct. Module member resolution asks Roslyn (compilation.IsSymbolAccessibleWithin(candidate, containerSymbol)), which honors InternalsVisibleTo. Constructor selection, setter injection and the container's inherited-member lookup instead hand-rolled a SymbolEqualityComparer.Default.Equals(a.ContainingAssembly, b.ContainingAssembly) "same assembly" comparison, which does not.

The two disagreed within a single scan. A referenced assembly that grants [assembly: InternalsVisibleTo("MyApp")] and exposes a public sealed class Roaster with an internal constructor passed IsScanCandidate (that check already asked Roslyn), and was then rejected by IsAccessibleConstructor — failing the build with AWT104 on a new the generated container could legally have emitted.

Replaces the hand-rolled comparison with Roslyn's own check at all three sites:

• SelectConstructor / IsAccessibleConstructor (AWT104): an internal constructor across an InternalsVisibleTo boundary is now selected, and without the grant still reports AWT104.
• IsAccessibleSetter (AWT136): property injection into an internal setter across the boundary now works; without the grant it still surfaces as AWT136 rather than an inaccessible-setter error in generated code.
• AccessibleMembers / IsAccessibleFromDerived: the base-type walk for factory / instance / hook lookup. The container's own members keep their deliberate permissiveness — they are yielded by the first loop untouched, and IsSymbolAccessibleWithin(privateMember, containerSymbol) would return true for them anyway, since a generated partial can use its container's private members.

Threads Compilation to the sites that lacked it (SelectConstructor and its five callers, InjectionContext, ExpandOpenGenerics / RequiredServiceTypes, and the scan's FirstUnconstructableReason / UnsatisfiableInjectedMemberReason).

Beyond InternalsVisibleTo, Roslyn's check also walks containing types, so a constructor that is public on a type nested privately in some other class is no longer treated as reachable. That code never compiled anyway.

AwaitenGenerator.OpenGenerics.cs:593 looks like the same pattern but is not: it checks DeclaredAccessibility == Accessibility.Public for a new() constraint, and the constraint genuinely requires a public parameterless constructor. where T : new() over a type whose only parameterless constructor is internal is CS0310 even within one assembly, so relaxing it would make expansion synthesize closed generics that do not compile. Left as is.

All 591 generator tests pass unchanged, including CrossAssemblyModuleTests.CrossAssemblyModule_InternalFactory_ReportsAwt108NamingTheModule, which pins the opposite boundary: a cross-assembly internal member without InternalsVisibleTo is not imported into the referencing compilation's symbol tables at all, so it stays AWT108 (invisibility) rather than becoming AWT153 (inaccessibility).

[github-actions]

Test Results

   18 files  ± 0     18 suites  ±0   8m 56s ⏱️ +19s
1 135 tests + 7  1 134 ✅ + 7  1 💤 ±0  0 ❌ ±0 
5 522 runs  +21  5 521 ✅ +21  1 💤 ±0  0 ❌ ±0 

Results for commit 678b7e9. ± Comparison against base commit bc305b8.

♻️ This comment has been updated with latest results.

[github-actions]

🚀 Benchmark Results

Details

BenchmarkDotNet v0.15.8, Linux Ubuntu 24.04.4 LTS (Noble Numbat)

AMD EPYC 9V74 2.60GHz, 1 CPU, 4 logical and 2 physical cores

.NET SDK 10.0.302

[Host] : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v3
Job=InProcess  Toolchain=InProcessEmitToolchain  IterationCount=15

LaunchCount=1  WarmupCount=10

Build	Size	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
baseline*	8	16.37 ns	0.108 ns	0.090 ns	0.97	144 B	1.00
Awaiten	8	16.936 ns	0.2115 ns	0.1875 ns	1.00	144 B	1.00
MsDI	8	1,409.918 ns	7.5525 ns	6.6951 ns	83.26	5688 B	39.50
Autofac	8	34,119.371 ns	164.8775 ns	154.2265 ns	2,014.78	33098 B	229.85
Jab	8	9.840 ns	0.0281 ns	0.0220 ns	0.58	96 B	0.67
PureDI	8	14.789 ns	0.1489 ns	0.1320 ns	0.87	128 B	0.89
DryIoc	8	744.705 ns	10.0350 ns	9.3868 ns	43.98	1528 B	10.61
SimpleInjector	8	11,082.052 ns	82.3888 ns	73.0355 ns	654.40	24760 B	171.94
baseline*	256	86.97 ns	0.352 ns	0.329 ns	1.06	2128 B	1.00
Awaiten	256	81.692 ns	0.3295 ns	0.2921 ns	1.00	2128 B	1.00
MsDI	256	13,721.151 ns	66.3990 ns	62.1097 ns	167.96	61016 B	28.67
Autofac	256	826,597.188 ns	2,661.0898 ns	2,077.6039 ns	10,118.63	720917 B	338.78
Jab	256	65.700 ns	0.3695 ns	0.3276 ns	0.80	2080 B	0.98
PureDI	256	75.709 ns	0.0893 ns	0.0746 ns	0.93	2112 B	0.99
DryIoc	256	44,695.060 ns	333.7634 ns	312.2025 ns	547.13	80099 B	37.64
SimpleInjector	256	336,052.200 ns	1,674.9132 ns	1,484.7669 ns	4,113.72	573088 B	269.31

Details

BenchmarkDotNet v0.15.8, Linux Ubuntu 24.04.4 LTS (Noble Numbat)

AMD EPYC 7763 2.45GHz, 1 CPU, 4 logical and 2 physical cores

.NET SDK 10.0.302

[Host] : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v3
Job=InProcess  Toolchain=InProcessEmitToolchain  IterationCount=15

LaunchCount=1  WarmupCount=10

Realistic	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
baseline*	245.6 ns	1.71 ns	1.51 ns	1.06	568 B	1.00
Awaiten	232.7 ns	3.39 ns	3.17 ns	1.00	568 B	1.00
MsDI	592.5 ns	11.57 ns	10.26 ns	2.55	1104 B	1.94
Autofac	6,623.8 ns	28.00 ns	26.19 ns	28.47	10784 B	18.99
Jab	173.0 ns	0.19 ns	0.16 ns	0.74	432 B	0.76
DryIoc	404.0 ns	4.68 ns	4.15 ns	1.74	944 B	1.66
SimpleInjector	711.1 ns	4.06 ns	3.80 ns	3.06	1096 B	1.93
PureDI	174.6 ns	0.93 ns	0.78 ns	0.75	632 B	1.11

Details

BenchmarkDotNet v0.15.8, Linux Ubuntu 24.04.4 LTS (Noble Numbat)

AMD EPYC 7763 2.45GHz, 1 CPU, 4 logical and 2 physical cores

.NET SDK 10.0.302

[Host] : .NET 10.0.10 (10.0.10, 10.0.1026.32716), X64 RyuJIT x86-64-v3
Job=InProcess  Toolchain=InProcessEmitToolchain  IterationCount=15

LaunchCount=1  WarmupCount=10

Resolve	Size	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
baseline*	8	7.022 ns	0.0083 ns	0.0065 ns	1.00	-	NA
Awaiten	8	7.040 ns	0.0755 ns	0.0630 ns	1.00	-	NA
MsDI	8	7.497 ns	0.0049 ns	0.0044 ns	1.06	-	NA
Autofac	8	144.620 ns	4.5940 ns	4.2972 ns	20.54	656 B	NA
Jab	8	2.653 ns	0.0050 ns	0.0042 ns	0.38	-	NA
PureDI	8	5.436 ns	0.0210 ns	0.0186 ns	0.77	-	NA
DryIoc	8	9.075 ns	0.0054 ns	0.0045 ns	1.29	-	NA
SimpleInjector	8	10.897 ns	0.0183 ns	0.0162 ns	1.55	-	NA
baseline*	256	8.092 ns	0.0144 ns	0.0121 ns	1.03	-	NA
Awaiten	256	7.836 ns	0.0179 ns	0.0159 ns	1.00	-	NA
MsDI	256	7.357 ns	0.0133 ns	0.0111 ns	0.94	-	NA
Autofac	256	140.319 ns	2.5967 ns	2.4290 ns	17.91	656 B	NA
Jab	256	43.134 ns	0.0487 ns	0.0407 ns	5.50	-	NA
PureDI	256	8.267 ns	0.0075 ns	0.0067 ns	1.06	-	NA
DryIoc	256	9.030 ns	0.0082 ns	0.0069 ns	1.15	-	NA
SimpleInjector	256	14.493 ns	0.1070 ns	0.1001 ns	1.85	-	NA

baseline* rows show the corresponding Awaiten benchmark from the most recent successful main branch build with results, for regression comparison.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
85.7% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud