🔧 Fix peek expression terminator issues and add ternary operator support

- Fix LLVM IR terminator missing in peek expression entry blocks
- Add proper jump instructions between peek blocks
- Implement ternary operator (? :) as syntactic sugar for peek
- Update Python LLVM externcall handling for improved compatibility
- Add comprehensive test cases for peek and ternary expressions
- Update language guide with ternary operator documentation

ChatGPTが頑張って修正してくれたにゃ!🐱

Co-Authored-By: ChatGPT <noreply@openai.com>
This commit is contained in:
Selfhosting Dev
2025-09-14 20:30:38 +09:00
parent 8f1b2ffa12
commit 5cad0ab20c
17 changed files with 241 additions and 38 deletions

View File

@ -1,5 +1,23 @@
# Current Task (2025-09-14 改定) — Phase 15 llvmlite既定+ PyVM新規
Context Snapshot — Open After Reset
- Status: A6 受入PyVM↔llvmlite parity + LLVM verify→.o→EXE完了。
- Lang: peek ブロック式最後の式が値OK、式文フォールバックOK、三項(?:)パーサ導入済みVM E2E 緑)。
- Docs: 言語/アーキの入口を整備guides/language-guide.md, reference/language/**, reference/architecture/**)。
- Parser MVP: Python Stage1 実装 + roundtrip スモーク緑。Nyash 実装スケルトン配置済み。
Next (short)
1) 三項(?:)の PyVM/llvmlite パリティE2E`tools/parity.sh`
2) peek サンプル/テスト拡充(式文・ブロック・戻り値)
3) Docs 追補最小例、must_use 注意)
4) Parser MVP Stage2 設計local/if/loop/call/method/new/me/substring/length/lastIndexOf
Quick Verify
- Ternary (VM): create file with `return (1 < 2) ? 10 : 20` and run
`./target/release/nyash --backend vm /tmp/tern4.nyash`
- Peek block (VM): `./target/release/nyash --backend vm apps/tests/peek_expr_block.nyash`
- Python parser RT: `NYASH_CLI_VERBOSE=1 tools/ny_parser_mvp_roundtrip.sh`
Summary
- JIT/Cranelift は一時停止。Rust/inkwell LLVM は参照のみ。
- 既定の実行/ビルド経路は Python/llvmlite ハーネスMIR JSON→.o→NyRT link

Binary file not shown.

Binary file not shown.

BIN
app_parity_peek_expr_block Normal file

Binary file not shown.

Binary file not shown.

BIN
app_parity_ternary_basic Normal file

Binary file not shown.

BIN
app_parity_ternary_nested Normal file

Binary file not shown.

BIN
app_peek_expr_block Normal file

Binary file not shown.

View File

@ -0,0 +1,15 @@
static box Main {
main(args) {
// Use peek as an expression to produce a value
local d = "1"
local v = peek d {
"0" => { 0 }
"1" => { 1 }
else => { 0 }
}
// Show the result and finish
local console = new ConsoleBox()
console.println(v)
return 0
}
}

View File

@ -0,0 +1,10 @@
static box Main {
main(args) {
local a = 3
local b = 5
// Nested ternary: should evaluate to 50
local v = (a < b) ? ((b < 0) ? 40 : 50) : 60
return v
}
}

View File

@ -15,6 +15,39 @@ Common Constructs
- Null-coalesce: `x ?? y``peek x { null => y, else => x }`
- Safe access: `a?.b``peek a { null => null, else => a.b }`
Minimal Examples
- Ternary
```nyash
static box Main {
main(args) {
local a = 3
local b = 5
// Nested ternary is supported
local v = (a < b) ? ((b < 0) ? 40 : 50) : 60
return v
}
}
```
- Peek as expression block (last expression is the value)
```nyash
static box Main {
main(args) {
local d = "1"
// Each arm can be a block; the last expression becomes the value
local dv = peek d {
"0" => { print("found zero") 0 }
"1" => { print("found one") 1 }
else => { print("other") 0 }
}
return dv
}
}
```
must_use Notes
- Peek arms are expressions. When using a block arm `{ ... }`, the last expression is the resulting value; statements without a final expression yield no usable value.
- Ternary is an expression; ensure both branches are type-compatible at MIR level (e.g., both yield integer or both yield string handle in current phase).
When you need the implementation details
- Tokenizer: src/tokenizer.rs
- Parser: src/parser/expressions.rs, src/parser/statements.rs

View File

@ -0,0 +1,59 @@
# Parser MVP Stage-2 Design (Phase 15)
Scope
- Expand Stage-1 (const/binop/return) to cover the minimal everyday surface:
- local bindings: `local x = expr`
- if/else: `if cond { ... } else { ... }`
- loop: `loop(cond) { ... }` and `loop cond { ... }`
- call: `f(a, b)` (free function)
- method: `obj.method(a, b)`
- constructor: `new BoxType(args)`
- implicit receiver `me` in box methods
- String methods: `substring`, `length`, `lastIndexOf`
Grammar Sketch (incremental)
- Expression
- precedence keeps Phase 12.7: ternary inside pipe; ternary lowers to if/branch
- postfix: method call, field access
- primary: literals, identifiers, parenthesized
- Statement
- `local ident = expr` (no rebind in Stage-2)
- expression statement (for side effects)
- return
AST Nodes (delta)
- Add nodes for Local, If, Loop, Call, MethodCall, New, Identifier, Return.
- Keep Peek and Ternary in the AST; lower to MIR via the same builder hooks.
Lowering to MIR JSON v0
- Local → assign to fresh value id, store into vmap
- If/Ternary → compare + branch blocks; PHI for join values when used as expression
- Loop → cond block → body block → back-edge; loop-carried values explicitly phied
- Call → `call { func: "name", args: [...] }`
- Method → `boxcall { box: vid, method: "name", args: [...] }`
- New → `newbox { type: "BoxType", args: [...] }`
- String ops
- `length()` → i64 length on string handle (VM: Python len; LLVM: handle→ptr at call sites only)
- `substring(a,b)``nyash.string.substring_hh` equivalent
- `lastIndexOf(x)` → returns i64 index or -1
Type Meta (emitted with JSON where needed)
- Compare: when both sides are string → `cmp_kind: "string"`
- BinOp `+`: if `dst_type = { kind: "handle", box_type: "StringBox" }`, force concat_hh
- Known APIs: annotate `dst_type` for Console.println (i64), dirname/join/read as `StringBox(handle)`
Policy & Invariants
- Resolver-only: do not wire PHIs in-lowering; snapshots + `finalize_phis` bind incomings
- Strings: inter-block are handle(i64) only; i8* materialized at call sites
- PHI placeholders: created at block heads from JSON-declared phi
Acceptance
- Parity on:
- `apps/tests/min_str_cat_loop/main.nyash`
- `apps/tests/esc_dirname_smoke.nyash`
- Stage-2 new smokes (ternary nested, peek return)
- `tools/parity.sh --lhs pyvm --rhs llvmlite --show-diff <app>` → green
Notes
- Keep Stage-2 additive and test-driven; avoid broad refactors.
- Future: numeric methods, map/array minimal, and using/namespace gate.

View File

@ -82,6 +82,7 @@ def lower_externcall(
# Prepare/coerce arguments
call_args: List[ir.Value] = []
for i, arg_id in enumerate(args):
orig_arg_id = arg_id
# Prefer resolver
if resolver is not None and preds is not None and block_end_values is not None and bb_map is not None:
if len(func.args) > i and isinstance(func.args[i].type, ir.PointerType):
@ -99,19 +100,47 @@ def lower_externcall(
expected_ty = func.args[i].type
if isinstance(expected_ty, ir.PointerType):
# Need pointer
if hasattr(aval, 'type'):
if isinstance(aval.type, ir.IntType):
aval = builder.inttoptr(aval, expected_ty, name=f"ext_i2p_arg{i}")
elif not aval.type.is_pointer:
aval = ir.Constant(expected_ty, None)
else:
# Pointer but wrong element type: if pointer-to-array -> GEP to i8*
try:
if isinstance(aval.type.pointee, ir.ArrayType) and isinstance(expected_ty.pointee, ir.IntType) and expected_ty.pointee.width == 8:
c0 = ir.Constant(ir.IntType(32), 0)
aval = builder.gep(aval, [c0, c0], name=f"ext_gep_arg{i}")
except Exception:
pass
# Prefer string literal pointer or handle->i8* bridge when argument is string-ish
used_string_h2p = False
try:
if resolver is not None and hasattr(resolver, 'string_ptrs'):
sp = resolver.string_ptrs.get(orig_arg_id)
if sp is not None:
aval = sp
used_string_h2p = True
if not used_string_h2p and resolver is not None and hasattr(resolver, 'is_stringish') and resolver.is_stringish(orig_arg_id):
# Declare nyash.string.to_i8p_h(i64) and call with handle
i64 = ir.IntType(64)
i8p = ir.IntType(8).as_pointer()
to_i8p = None
for f in module.functions:
if f.name == 'nyash.string.to_i8p_h':
to_i8p = f; break
if to_i8p is None:
to_i8p = ir.Function(module, ir.FunctionType(i8p, [i64]), name='nyash.string.to_i8p_h')
# Ensure we have an i64 handle to pass
if hasattr(aval, 'type') and isinstance(aval.type, ir.PointerType):
aval = builder.ptrtoint(aval, i64, name=f"ext_p2h_{i}")
elif hasattr(aval, 'type') and isinstance(aval.type, ir.IntType) and aval.type.width != 64:
aval = builder.zext(aval, i64, name=f"ext_zext_h_{i}")
aval = builder.call(to_i8p, [aval], name=f"ext_h2p_arg{i}")
used_string_h2p = True
except Exception:
used_string_h2p = used_string_h2p or False
if not used_string_h2p:
if hasattr(aval, 'type'):
if isinstance(aval.type, ir.IntType):
aval = builder.inttoptr(aval, expected_ty, name=f"ext_i2p_arg{i}")
elif not aval.type.is_pointer:
aval = ir.Constant(expected_ty, None)
else:
# Pointer but wrong element type: if pointer-to-array -> GEP to i8*
try:
if isinstance(aval.type.pointee, ir.ArrayType) and isinstance(expected_ty.pointee, ir.IntType) and expected_ty.pointee.width == 8:
c0 = ir.Constant(ir.IntType(32), 0)
aval = builder.gep(aval, [c0, c0], name=f"ext_gep_arg{i}")
except Exception:
pass
else:
aval = ir.Constant(expected_ty, None)
elif isinstance(expected_ty, ir.IntType) and expected_ty.width == 64:

View File

@ -9,42 +9,81 @@ impl super::MirBuilder {
arms: Vec<(LiteralValue, ASTNode)>,
else_expr: ASTNode,
) -> Result<ValueId, String> {
// Evaluate scrutinee in the current block
let scr_val = self.build_expression_impl(scrutinee)?;
// Prepare merge and result
let merge_block: BasicBlockId = self.block_gen.next();
let result_val = self.value_gen.next();
let mut phi_inputs: Vec<(BasicBlockId, ValueId)> = Vec::new();
let mut next_block = self.block_gen.next();
self.start_new_block(next_block)?;
// Create dispatch block where we start comparing arms
let dispatch_block = self.block_gen.next();
// Jump from current block to dispatch (ensure terminator exists)
let need_jump = {
let cur = self.current_block;
if let (Some(cb), Some(ref func)) = (cur, &self.current_function) {
if let Some(bb) = func.blocks.get(&cb) { !bb.is_terminated() } else { true }
} else { true }
};
if need_jump {
self.emit_instruction(super::MirInstruction::Jump { target: dispatch_block })?;
}
self.start_new_block(dispatch_block)?;
// If there are no arms, fall through to else directly
if arms.is_empty() {
let else_block = self.block_gen.next();
self.emit_instruction(super::MirInstruction::Jump { target: else_block })?;
self.start_new_block(else_block)?;
let else_val = self.build_expression_impl(else_expr)?;
phi_inputs.push((else_block, else_val));
self.emit_instruction(super::MirInstruction::Jump { target: merge_block })?;
self.start_new_block(merge_block)?;
self.emit_instruction(super::MirInstruction::Phi { dst: result_val, inputs: phi_inputs })?;
return Ok(result_val);
}
// Else block to handle default case
let else_block = self.block_gen.next();
// Chain dispatch blocks for each arm
let mut cur_dispatch = dispatch_block;
for (i, (label, arm_expr)) in arms.iter().cloned().enumerate() {
let then_block = self.block_gen.next();
// Only string labels handled here (behavior unchanged)
// Next dispatch (only for non-last arm)
let next_dispatch = if i + 1 < arms.len() { Some(self.block_gen.next()) } else { None };
let else_target = next_dispatch.unwrap_or(else_block);
// In current dispatch block, compare and branch
self.start_new_block(cur_dispatch)?;
if let LiteralValue::String(s) = label {
let lit_id = self.value_gen.next();
self.emit_instruction(super::MirInstruction::Const { dst: lit_id, value: super::ConstValue::String(s) })?;
let cond_id = self.value_gen.next();
self.emit_instruction(super::MirInstruction::Compare { dst: cond_id, op: super::CompareOp::Eq, lhs: scr_val, rhs: lit_id })?;
self.emit_instruction(super::MirInstruction::Branch { condition: cond_id, then_bb: then_block, else_bb: next_block })?;
self.start_new_block(then_block)?;
let then_val = self.build_expression_impl(arm_expr)?;
phi_inputs.push((then_block, then_val));
self.emit_instruction(super::MirInstruction::Jump { target: merge_block })?;
if i < arms.len() - 1 {
let b = self.block_gen.next();
self.start_new_block(b)?;
next_block = b;
}
self.emit_instruction(super::MirInstruction::Branch { condition: cond_id, then_bb: then_block, else_bb: else_target })?;
}
// then arm
self.start_new_block(then_block)?;
let then_val = self.build_expression_impl(arm_expr)?;
phi_inputs.push((then_block, then_val));
self.emit_instruction(super::MirInstruction::Jump { target: merge_block })?;
// Move to next dispatch or else block
cur_dispatch = else_target;
}
let else_block_id = next_block;
self.start_new_block(else_block_id)?;
// Lower else expression in else_block
self.start_new_block(else_block)?;
let else_val = self.build_expression_impl(else_expr)?;
phi_inputs.push((else_block_id, else_val));
phi_inputs.push((else_block, else_val));
self.emit_instruction(super::MirInstruction::Jump { target: merge_block })?;
// Merge and yield result
self.start_new_block(merge_block)?;
self.emit_instruction(super::MirInstruction::Phi { dst: result_val, inputs: phi_inputs })?;
Ok(result_val)
}
}

View File

@ -98,7 +98,7 @@ impl NyashParser {
condition: Box::new(cond),
then_body: vec![then_expr],
else_body: Some(vec![else_expr]),
span: Span::Unknown,
span: Span::unknown(),
});
}
Ok(cond)

View File

@ -151,15 +151,14 @@ impl NyashRunner {
.status()
.map_err(|e| format!("spawn pyvm: {}", e))
.unwrap();
// Always propagate PyVM exit code to match llvmlite semantics
let code = status.code().unwrap_or(1);
if !status.success() {
eprintln!("❌ PyVM failed (status={})", status.code().unwrap_or(-1));
process::exit(1);
if std::env::var("NYASH_CLI_VERBOSE").ok().as_deref() == Some("1") {
eprintln!("❌ PyVM failed (status={})", code);
}
}
// Propagate exit code if set
if let Some(code) = status.code() {
process::exit(code);
}
process::exit(0);
process::exit(code);
} else {
eprintln!("❌ PyVM runner not found: {}", runner.display());
process::exit(1);

View File

@ -67,6 +67,7 @@ normalize() {
-e '/^🔌/d' \
-e '/^✅/d' \
-e '/^🚀/d' \
-e '/^❌/d' \
-e '/^⚡/d' \
-e '/^🦀/d' \
-e '/^🧠/d' \