Skip to main content

compiler-development

Build NeverC or other LLVM compilers across frontend, IR, optimization, code generation, ABI, and target support. Use when the compiler pipeline is the subject; use narrower LLVM skills for isolated tooling or analysis.

Quellinformationen

Repository
NeverSight/NeverC
Letzte Quellaktivität
13. September 2026 um 11:17
Erkannte Sprache von SKILL.md
Englisch
Sterne
92
Forks
11

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

Datei-Explorer
5 Dateien

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
compiler-development
description
Build NeverC or other LLVM compilers across frontend, IR, optimization, code generation, ABI, and target support. Use when the compiler pipeline is the subject; use narrower LLVM skills for isolated tooling or analysis.
# Compiler development Use this skill when a change crosses compiler stages or target boundaries. Keep source-language semantics, LLVM IR, ABI lowering, object format, linker behavior, and runtime assumptions owned by the narrowest authoritative layer. ## Topic routing Read only the reference that matches the work: - [Targets and ABI](references/targets-and-abi.md) for platform matrices, calling conventions, build-system target handling, and module metadata. - [DynCode and testing](references/dyncode-and-testing.md) for cross-platform DynCode invariants and target-specific verification. - [Compiler pipeline](references/compiler-pipeline.md) for frontend, AST, IR generation, optimization, JIT, and language implementation patterns. - [Workflow and resources](references/workflow-and-resources.md) for broader compiler-development workflow and upstream references. For an isolated pass use `llvm-optimization`; for Clang/LLDB/LSP developer tooling use `llvm-tooling`; for static or dynamic analyses use their narrower skills. Prefer current NeverC source and repository documentation over examples in these references.
Auf GitHub ansehen