Repository navigation
codegen: the emitter's one reflection site (IsClrTypeGenericInNamespace) spells a CLR type by resolving it at emit time against FrameworkOnly while semantic analysis resolved it against the registry — Rule-2 cell, no materialized fact #2320
Description
Class
Contract (CLAUDE.md Rule 2): the RoslynEmitter makes no type decision and performs no reflection; every fact it reads is materialized in semantic analysis (Symbol.CodeGenInfo or a node-keyed SemanticInfo carrier). RoslynEmitter.IsClrTypeGenericInNamespace (src/Sharpy.Compiler/CodeGen/RoslynEmitter.CompilationUnit.cs:588 @ 045ce6b) is the emitter's one remaining reflection site: it decides how to spell a type name (Sharpy.IList vs System.Collections.Generic.IList<T>) by resolving {namespace}.{name}`` {1..4}` through reflection at emit time.
Cells
| # | cell | observed @ 045ce6b | expected |
|---|---|---|---|
| 1 | the site itself | reflection in CodeGen; its comment says "Rule-2 cell (#2250 findings ledger; no tracker yet)" — a known defect comment with no issue (Rule 8) | a materialized fact; no reflection in CodeGen |
| 2 | which world it asks (P54 seam divergence, code-read by the class-cure auditor) | since 333de07 the site asks AssemblyVisibility.FrameworkOnly (the shared framework + Sharpy.Core — a process constant, so the P54 contract holds); semantic analysis answers the same question from the compilation's registry visibility ({framework} ∪ {this registry's references}). Its caller IsNetFrameworkNamespace admits any System.*/Microsoft.* namespace, so for a referenced non-framework package in such a namespace (e.g. -r System.Reactive.dll, System.Reactive.Linq) the semantic side sees the generic type and the emitter does not — the two halves of one decision read two different sets |
one decision, made once, in semantic analysis |
Lane ice (a mismatched spelling surfaces as SPY0908 behind the C# compiler, if it surfaces at all), reach probe (needs a referenced package in a System.*/Microsoft.* namespace whose generic type is spelled through this arm — no program-level repro recorded; cell 2 is code-read).
Acceptance
- The spelling decision is recorded on the type symbol's
CodeGenInfo(or a node-keyed carrier merged inSemanticInfo.MergeFrom) during semantic analysis, computed from the registry's visibility;IsClrTypeGenericInNamespaceand theFrameworkOnlyread in CodeGen are deleted. EmitterCarrierOnlyConformanceTests/ProcessWideResolutionScanTestsratchet the site out (the scan'ssrc/Sharpy.Compilercensus drops by one; the emitter names noDiscovery.AssemblyVisibility).- A program with a referenced package generic in a
System.*namespace prints the same underrunbefore and after (direction record), and a mutation that re-readsFrameworkOnlyin the emitter is red.
Found by the P54 (#2250) plan's own trackers list ("Rule-2 emitter reflection: no open tracker — the lead files ONE class issue") and confirmed by the /verify-implementation class-cure auditor @ 045ce6b.
Activity
test(conformance): the P54 ratchets cite an issue by number, not the c…
antonsynd commentedon Oct 15, 2026
Batching 2026-10-15 @ f8c37f9 — charter
.claude/plans/remediation-round-2026-09-20-batching.md§3.0, chore (same band as #2261 part 2, #2311, #2313 cell 0; rides P21e's gate). Ruling R-HM (owner, 2026-10-15): (a) a chore — the emitter's one remaining reflection site becomes a materialized fact.Internals (read @ f8c37f9):
RoslynEmitter.CompilationUnit.cs:588IsClrTypeGenericInNamespaceresolves{ns}.{name}`1..4 againstAssemblyVisibility.FrameworkOnlyto skip ausingalias of an open generic (CS0305); its one caller (:510) sits underIsNetFrameworkNamespace(:546), which admits anySystem.*/Microsoft.*module name, so a referenced package in such a namespace is visible to the checker (registry visibility) and invisible here. Mechanism: one symbol-keyed fact on the imported type symbol'sCodeGenInfo("is an open generic definition"), computed from the registry's visibility in semantic analysis and frozen atMaterializeCodeGenInfo; the method and theFrameworkOnlyread are deleted. Guard:EmitterCarrierOnlyConformanceTests/ProcessWideResolutionScanTestsdrop one census entry (the emitter names noDiscovery.AssemblyVisibility); direction record on a program importing a referencedSystem.*-namespace generic by name (prints the same before and after); mutation (re-readFrameworkOnlyin the emitter → red). Semantic + CodeGen → a whole-solution gate; seam-free against P21e. Rejected: (b) a P26 cell (~20 groups out; no layout change rides on this fact). Two-way. Lane / reach: ice · probe. Class tracker: this issue.refactor(codegen): the emitter performs no reflection — whether an im…
antonsynd commentedon Oct 22, 2026
Landed (chores band 2026-10-22) on
dev@ 4d22212, PR #2346 to mainline. Whole-solution gate 81669/0/7 @ 4d22212 (measured), against 81093/0/7 @ be4664e.Commit f6cf5c9 (R-HM a).
CodeGenInfo.IsOpenGenericDefinitionis recorded byCodeGenInfoComputer.ProcessFromImport. It is computed against the registry's own visibility, framework plus references, and freezes atMaterializeCodeGenInfo. The emitter'sIsClrTypeGenericInNamespace, itsAssemblyVisibility.FrameworkOnlyread and the TODO are deleted.SemanticBinding.MergeFromcarries the fact ontoas-aliased imports.Premise correction. Neither
EmitterCarrierOnlyConformanceTestsnorProcessWideResolutionScanTestshad a census entry for this site, so there was nothing to ratchet out. The new guard isEmitterCarrierOnlyConformanceTests.CodeGenSources_NameNoClrVisibilitySeam. It denies CodeGen every Discovery type that is or carriesAssemblyVisibility, derived by reflection, with a synthetic positive and negative control.Matrix:
OpenGenericImportAliasTests, visibility {framework, referenced} × spelling {bare,as} × name {generic arity group, non-generic}.Mutation: the emitter re-reading
FrameworkOnlyturned 1 row red.Direction: framework imports print the same and the generated C# is byte-identical. A referenced
System.Widgetsprints the same, and the only C# change is thatusing Box = …Box;is no longer emitted, because the framework-only read missedBox<T>.Every acceptance bullet is met. Closing.
Merge pull request #2346 from antonsynd/dev