Files
hakorune/src/mir/builder/control_flow/joinir/routing.rs
nyash-codex 632e495e51 docs(mir): Phase 137-6-S3 - Router委譲の準備コメント追加
## 目的
将来の Router → Canonicalizer 委譲に備えた TODO コメント拡充

## 変更内容

### routing.rs の TODO コメント拡充
- 有効化条件を明記(新フラグ or 既存フラグ)
- 注意事項追加(全 Pattern の parity green 必須)
- コード例を追加(将来実装時の参考)

### Phase 137 README 更新
- Phase 137-6(完了)セクション追加
- S1/S2/S3 の実装内容を記録
- 効果と受け入れ基準達成を記録

## 効果
-  将来の委譲に備えた明確なガイド
-  Phase 137-6 完了記録
-  既定挙動不変(フラグOFF時)

## Phase 137-6 完了サマリー

### 実装完了内容
- **S1**: choose_pattern_kind SSOT 入口(61行追加)
- **S2**: dev-only parity check 統合(52行追加)
- **S3**: Router 委譲準備コメント(ドキュメント拡充)

### 受け入れ基準
-  strict parity green(skip_whitespace)
-  既定挙動不変(フラグOFF時)
-  新 env 追加なし
-  choose_pattern_kind が SSOT 入口として機能
-  全テスト PASS(退行なし)

### テスト結果
-  `cargo build --release`: 成功
-  スモークテスト(simple_*): 5/5 PASS
-  parity check 動作確認:
  ```
  NYASH_JOINIR_DEV=1 HAKO_JOINIR_STRICT=1 ./target/release/hakorune \
    tools/selfhost/test_pattern3_skip_whitespace.hako
  → [choose_pattern_kind/PARITY] OK
  ```

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2025-12-16 07:44:57 +09:00

327 lines
14 KiB
Rust
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

//! JoinIR routing logic for loop lowering
use super::trace;
use crate::ast::ASTNode;
use crate::mir::builder::MirBuilder;
use crate::mir::ValueId;
/// Pattern 選択の SSOT 入口
///
/// 既存の分散した選択ロジックをここに集約する。
/// 将来的には Canonicalizer decision に委譲する。
///
/// Phase 137-6-S1: 現時点では既存の router ロジックLoopFeatures ベース)を使用
/// Phase 137-6-S2: dev-only で canonicalizer decision を提案として受け取る
pub(in crate::mir::builder) fn choose_pattern_kind(
condition: &ASTNode,
body: &[ASTNode],
) -> crate::mir::loop_pattern_detection::LoopPatternKind {
use crate::mir::builder::control_flow::joinir::patterns::ast_feature_extractor as ast_features;
use crate::mir::loop_pattern_detection;
// Phase 193: Use AST Feature Extractor Box for break/continue detection
let has_continue = ast_features::detect_continue_in_body(body);
let has_break = ast_features::detect_break_in_body(body);
// Phase 193: Extract features using modularized extractor
let features = ast_features::extract_features(condition, body, has_continue, has_break);
// Phase 192: Classify pattern based on features (既存の router 結果)
let router_choice = loop_pattern_detection::classify(&features);
// Phase 137-6-S2: dev-only で Canonicalizer の提案を取得
if crate::config::env::joinir_dev_enabled() {
use crate::ast::Span;
use crate::mir::loop_canonicalizer::canonicalize_loop_expr;
let loop_ast = ASTNode::Loop {
condition: Box::new(condition.clone()),
body: body.to_vec(),
span: Span::unknown(),
};
if let Ok((_skeleton, decision)) = canonicalize_loop_expr(&loop_ast) {
if let Some(canonical_choice) = decision.chosen {
// parity check
if canonical_choice != router_choice {
let msg = format!(
"[choose_pattern_kind/PARITY] router={:?}, canonicalizer={:?}",
router_choice, canonical_choice
);
if crate::config::env::joinir_dev::strict_enabled() {
// strict mode: 不一致は Fail-Fast
panic!("{}", msg);
} else {
// debug mode: ログのみ
eprintln!("{}", msg);
}
} else {
// Patterns match - success!
eprintln!(
"[choose_pattern_kind/PARITY] OK: canonical and actual agree on {:?}",
canonical_choice
);
}
// TODO (Phase 137-6-S3): ここで canonical_choice を返す
// 現時点では router_choice を維持(既定挙動不変)
//
// 有効化条件(将来実装):
// 1. joinir_dev_enabled() && 新フラグ(例: canonicalizer_preferred()
// 2. または joinir_dev_enabled() をそのまま使用
//
// 注意: 有効化時は全 Pattern の parity が green であること
//
// 有効化後のコード例:
// ```rust
// if crate::config::env::canonicalizer_preferred() {
// return canonical_choice;
// }
// ```
}
}
}
router_choice
}
impl MirBuilder {
/// Phase 49: Try JoinIR Frontend for mainline integration
///
/// Returns `Ok(Some(value))` if the loop is successfully lowered via JoinIR,
/// `Ok(None)` if no JoinIR pattern matched (unsupported loop structure).
/// Phase 187-2: Legacy LoopBuilder removed - all loops must use JoinIR.
///
/// # Phase 49-4: Multi-target support
///
/// Targets are enabled via separate dev flags:
/// - `HAKO_JOINIR_PRINT_TOKENS_MAIN=1`: JsonTokenizer.print_tokens/0
/// - `HAKO_JOINIR_ARRAY_FILTER_MAIN=1`: ArrayExtBox.filter/2
///
/// Note: Arity in function names does NOT include implicit `me` receiver.
/// - Instance method `print_tokens()` → `/0` (no explicit params)
/// - Static method `filter(arr, pred)` → `/2` (two params)
pub(in crate::mir::builder) fn try_cf_loop_joinir(
&mut self,
condition: &ASTNode,
body: &[ASTNode],
) -> Result<Option<ValueId>, String> {
// Get current function name
let func_name = self
.scope_ctx
.current_function
.as_ref()
.map(|f| f.signature.name.clone())
.unwrap_or_default();
// Phase 195: Use unified trace
trace::trace().routing("router", &func_name, "try_cf_loop_joinir called");
// Phase 170-4: Structure-based routing option
// When NYASH_JOINIR_STRUCTURE_ONLY=1, skip function name whitelist
// and route purely based on loop structure analysis
// Phase 196: Default to structure-first routing now that LoopBuilder is removed.
// - Default: ON (structure_only = true) to allow JoinIR patterns to run for all loops.
// - To revert to the previous whitelist-only behavior, set NYASH_JOINIR_STRUCTURE_ONLY=0.
let structure_only = match std::env::var("NYASH_JOINIR_STRUCTURE_ONLY").ok().as_deref() {
Some("0") | Some("off") => false,
_ => true,
};
if structure_only {
trace::trace().routing(
"router",
&func_name,
"Structure-only mode enabled, skipping whitelist",
);
} else {
// Phase 49-4 + Phase 80: Multi-target routing (legacy whitelist)
// - JoinIR は常時 ON。legacy LoopBuilder は削除済み。
// - 代表2本print_tokens / ArrayExt.filterも常に JoinIR で試行する。
// Note: Arity does NOT include implicit `me` receiver
// Phase 188: Add "main" routing for loop pattern expansion
// Phase 170: Add JsonParserBox methods for selfhost validation
let is_target = match func_name.as_str() {
"main" => true, // Phase 188-Impl-1: Enable JoinIR for main function (Pattern 1)
"JoinIrMin.main/0" => true, // Phase 188-Impl-2: Enable JoinIR for JoinIrMin.main/0 (Pattern 2)
"JsonTokenizer.print_tokens/0" => true,
"ArrayExtBox.filter/2" => true,
// Phase 170-A-1: Enable JsonParserBox methods for JoinIR routing
"JsonParserBox._trim/1" => true,
"JsonParserBox._skip_whitespace/2" => true,
"JsonParserBox._match_literal/3" => true, // Phase 182: Fixed arity (s, pos, literal)
"JsonParserBox._parse_string/2" => true,
"JsonParserBox._parse_array/2" => true,
"JsonParserBox._parse_object/2" => true,
// Phase 182: Add simple loop methods
"JsonParserBox._parse_number/2" => true, // P2 Break (s, pos)
"JsonParserBox._atoi/1" => true, // P2 Break (s)
// Phase 170-A-1: Test methods (simplified versions)
"TrimTest.trim/1" => true,
"Main.trim/1" => true, // Phase 171-fix: Main box variant
"Main.trim_string_simple/1" => true, // Phase 33-13: Simple trim variant
"TrimTest.main/0" => true, // Phase 170: TrimTest.main for loop pattern test
// Phase 173: JsonParser P5 expansion test
"JsonParserTest._skip_whitespace/3" => true,
"JsonParserTest.main/0" => true,
// Phase 174: JsonParser complex loop P5B extension test
"JsonParserStringTest.parse_string_min/0" => true,
"JsonParserStringTest.main/0" => true,
// Phase 175: P5 multi-carrier support (2 carriers: pos + result)
"JsonParserStringTest2.parse_string_min2/0" => true,
"JsonParserStringTest2.main/0" => true,
_ => false,
};
if !is_target {
return Ok(None);
}
}
// Debug log when routing through JoinIR Frontend
// Phase 195: Check trace flags directly from JoinLoopTrace
let debug = trace::trace().is_loopform_enabled() || trace::trace().is_mainline_enabled();
trace::trace().routing(
"router",
&func_name,
"Routing through JoinIR Frontend mainline",
);
// Phase 49-3: Implement JoinIR Frontend integration
self.cf_loop_joinir_impl(condition, body, &func_name, debug)
}
/// Phase 49-3: JoinIR Frontend integration implementation
///
/// Routes loop compilation through either:
/// 1. Pattern-based router (Phase 194+) - preferred for new patterns
/// 2. Legacy binding path (Phase 49-3) - for whitelisted functions only
pub(in crate::mir::builder) fn cf_loop_joinir_impl(
&mut self,
condition: &ASTNode,
body: &[ASTNode],
func_name: &str,
debug: bool,
) -> Result<Option<ValueId>, String> {
// Phase 137-2/137-4: Dev-only observation via Loop Canonicalizer
if crate::config::env::joinir_dev_enabled() {
use crate::ast::Span;
use crate::mir::loop_canonicalizer::canonicalize_loop_expr;
// Reconstruct loop AST for canonicalizer
let loop_ast = ASTNode::Loop {
condition: Box::new(condition.clone()),
body: body.to_vec(),
span: Span::unknown(),
};
match canonicalize_loop_expr(&loop_ast) {
Ok((skeleton, decision)) => {
eprintln!("[loop_canonicalizer] Function: {}", func_name);
eprintln!(
"[loop_canonicalizer] Skeleton steps: {}",
skeleton.steps.len()
);
eprintln!(
"[loop_canonicalizer] Carriers: {}",
skeleton.carriers.len()
);
eprintln!(
"[loop_canonicalizer] Has exits: {}",
skeleton.exits.has_any_exit()
);
eprintln!(
"[loop_canonicalizer] Decision: {}",
if decision.is_success() {
"SUCCESS"
} else {
"FAIL_FAST"
}
);
if let Some(pattern) = decision.chosen {
eprintln!("[loop_canonicalizer] Chosen pattern: {:?}", pattern);
}
eprintln!(
"[loop_canonicalizer] Missing caps: {:?}",
decision.missing_caps
);
if decision.is_fail_fast() {
eprintln!(
"[loop_canonicalizer] Reason: {}",
decision.notes.join("; ")
);
}
// Phase 137-4: Router parity verification
if let Some(canonical_pattern) = decision.chosen {
// Get actual pattern from router (will be determined by LoopPatternContext)
// We need to defer this check until after ctx is created
// Store decision for later parity check
trace::trace().debug(
"canonicalizer",
&format!(
"Phase 137-4: Canonical pattern chosen: {:?} (parity check pending)",
canonical_pattern
),
);
}
}
Err(e) => {
eprintln!("[loop_canonicalizer] Function: {}", func_name);
eprintln!("[loop_canonicalizer] Error: {}", e);
}
}
}
// Phase 194: Use table-driven router instead of if/else chain
use super::patterns::{route_loop_pattern, LoopPatternContext};
// Phase 200-C: Pass fn_body_ast to LoopPatternContext if available
// Clone fn_body_ast to avoid borrow checker issues
let fn_body_clone = self.comp_ctx.fn_body_ast.clone();
trace::trace().routing(
"router",
func_name,
&format!(
"fn_body_ast is {}",
if fn_body_clone.is_some() {
"SOME"
} else {
"NONE"
}
),
);
let ctx = if let Some(ref fn_body) = fn_body_clone {
trace::trace().routing(
"router",
func_name,
&format!("Creating ctx with fn_body ({} nodes)", fn_body.len()),
);
LoopPatternContext::with_fn_body(condition, body, &func_name, debug, fn_body)
} else {
LoopPatternContext::new(condition, body, &func_name, debug)
};
// Phase 137-4: Router parity verification (after ctx is created)
if crate::config::env::joinir_dev_enabled() {
self.verify_router_parity(condition, body, func_name, &ctx)?;
}
if let Some(result) = route_loop_pattern(self, &ctx)? {
trace::trace().routing("router", func_name, "Pattern router succeeded");
return Ok(Some(result));
}
// Phase 187-2: Pattern router failed, try legacy whitelist
trace::trace().routing(
"router",
func_name,
"Pattern router found no match, trying legacy whitelist",
);
// Delegate to legacy binding path (routing_legacy_binding.rs)
self.cf_loop_joinir_legacy_binding(condition, body, func_name, debug)
}
}