メニュー
Compare our compiler's bytecode output against tclsh reference disassembly (8.4, 8.5, 8.6, 9.0). Use when verifying bytecode correctness, investigating instruction sequence differences, or checking that our codegen matches C Tcl.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Debug the Tcl compiler pipeline (lexer → green tree → segmenter → IR → CFG → SSA → codegen) running from source — never a built pyz/zipapp. Wraps the tcl-explorer CLI and adds source-slice and raw-token views that make off-by-one range bugs obvious. Use when investigating a wrong diagnostic position, a stray-brace / unmatched-delimiter false positive, a mis-tokenised word, an IR statement whose range looks wrong, or any time you need to see what the compiler produces for a snippet of Tcl.
Download and extract Tcl source trees (8.4, 8.5, 8.6, 9.0) to tmp/ for test suites and bytecode reference. Idempotent — skips versions already present.
Manage differential-fuzzer findings: query fixed/unfixed status, list by category, verify findings against the current VM, mark findings as fixed/unfixed, and batch-update categories. Use when working on fuzz findings, checking what's fixed, or triaging new results.
Use when verifying Tcl LSP server behavior: semantic tokens, diagnostics, formatting, hover, completion, definition, references, code actions, optimizations, document symbols, diagram extraction, event/command registry lookups, or benchmarking server performance and collecting timing logs. Tests the server directly over JSON-RPC without VS Code.
Query reference test results captured from real tclsh (8.4–9.0). Look up individual tests, search groups, view suite summaries, and compare results across Tcl versions.
Run the full release workflow: validate branch, test, generate changelog, tag, push, and publish to all editor marketplaces. Asks for patch/minor/major if not specified. Tag-only — no source-file edits, no commit on main (main is protected).
| name | bytecode-compare |
| description | Compare our compiler's bytecode output against tclsh reference disassembly (8.4, 8.5, 8.6, 9.0). Use when verifying bytecode correctness, investigating instruction sequence differences, or checking that our codegen matches C Tcl. |
| allowed-tools | Bash, Read |
Compares the bytecode produced by our compiler pipeline (lexer → lowering → IR → CFG → codegen) against tclsh reference disassembly (8.4, 8.5, 8.6, 9.0) for a set of 219 test snippets.
Run from the project root:
python3 .claude/skills/bytecode-compare/bytecode_compare.py [-v VERSION] <subcommand> [args...]
| Flag | Description |
|---|---|
-v, --version | Tcl version to compare against: 8.4, 8.5, 8.6, 9.0 (default: 9.0) |
| Subcommand | Arguments | What it does |
|---|---|---|
all | Compare all 219 snippets, show summary table | |
diff | <snippet> | Detailed instruction-by-instruction diff for one snippet (e.g. 08_while) |
summary | One-line per snippet: match/mismatch with instruction counts | |
instructions | <snippet> | Side-by-side normalised instruction listing |
refresh | Regenerate our bytecode reference and re-compare | |
categories | Group differences by category (variable access, jumps, opcodes, etc.) | |
versions | <snippet> | Compare one snippet against all Tcl versions (8.4, 8.5, 8.6, 9.0) |
01_set_simple MISMATCH ours: 3 instrs 9.0: 4 instrs
03_expr_braced MATCH ours: 2 instrs 9.0: 2 instrs
22_puts_invoke MATCH ours: 4 instrs 9.0: 4 instrs
MATCH means identical normalised instruction sequences. MISMATCH means differences exist.
=== 08_while ===
9.0: (0) push1 0 # "i"
9.0: (2) push1 1 # "0"
9.0: (4) storeStk
ours: (0) push1 0 # "0"
ours: (2) storeScalar1 %v0 # var "i"
--- variable access: storeStk (name-based) vs storeScalar1 (LVT-indexed)
Groups all divergences into categories:
server/compiler/codegen.py — verify instruction sequencessummary as a gate check$ARGUMENTS