Files
hakorune/src/runner/json_v0_bridge.rs

1607 lines
60 KiB
Rust
Raw Normal View History

use crate::mir::{
BasicBlockId, BinaryOp, ConstValue, EffectMask, FunctionSignature, MirFunction, MirInstruction,
MirModule, MirPrinter, MirType, ValueId,
};
use serde::{Deserialize, Serialize};
#[derive(Debug, Deserialize, Serialize)]
struct ProgramV0 {
version: i32,
kind: String,
body: Vec<StmtV0>,
}
#[derive(Debug, Deserialize, Serialize, Clone)]
#[serde(tag = "type")]
enum StmtV0 {
Return {
expr: ExprV0,
},
Extern {
iface: String,
method: String,
args: Vec<ExprV0>,
},
// Optional: expression statement (side effects only)
Expr {
expr: ExprV0,
},
// Optional: local binding (Stage-2)
Local {
name: String,
expr: ExprV0,
},
// Optional: if/else (Stage-2)
If {
cond: ExprV0,
then: Vec<StmtV0>,
#[serde(rename = "else", default)]
r#else: Option<Vec<StmtV0>>,
},
// Optional: loop (Stage-2)
Loop {
cond: ExprV0,
body: Vec<StmtV0>,
},
Break,
Continue,
Try {
#[serde(rename = "try")]
try_body: Vec<StmtV0>,
#[serde(default)]
catches: Vec<CatchV0>,
#[serde(default)]
finally: Vec<StmtV0>,
},
}
#[derive(Debug, Deserialize, Serialize, Clone, Default)]
struct CatchV0 {
#[serde(rename = "param", default)]
param: Option<String>,
#[serde(rename = "typeHint", default)]
type_hint: Option<String>,
#[serde(default)]
body: Vec<StmtV0>,
}
#[derive(Debug, Deserialize, Serialize, Clone)]
#[serde(tag = "type")]
enum ExprV0 {
Int {
value: serde_json::Value,
},
Str {
value: String,
},
Bool {
value: bool,
},
Binary {
op: String,
lhs: Box<ExprV0>,
rhs: Box<ExprV0>,
},
Extern {
iface: String,
method: String,
args: Vec<ExprV0>,
},
Compare {
op: String,
lhs: Box<ExprV0>,
rhs: Box<ExprV0>,
},
Logical {
op: String,
lhs: Box<ExprV0>,
rhs: Box<ExprV0>,
}, // short-circuit: &&, || (or: "and"/"or")
// Stage-2 additions (optional):
Call {
name: String,
args: Vec<ExprV0>,
},
Method {
recv: Box<ExprV0>,
method: String,
args: Vec<ExprV0>,
},
New {
class: String,
args: Vec<ExprV0>,
},
Var {
name: String,
},
Throw {
expr: Box<ExprV0>,
},
}
#[derive(Clone, Copy)]
struct LoopContext {
cond_bb: BasicBlockId,
exit_bb: BasicBlockId,
}
fn lower_throw(
f: &mut MirFunction,
cur_bb: BasicBlockId,
exception_value: ValueId,
) -> (ValueId, BasicBlockId) {
if std::env::var("NYASH_BRIDGE_THROW_ENABLE").ok().as_deref() == Some("1") {
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.set_terminator(MirInstruction::Throw {
exception: exception_value,
effects: EffectMask::PANIC,
});
}
(exception_value, cur_bb)
} else {
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::Const {
dst,
value: ConstValue::Integer(0),
});
}
(dst, cur_bb)
}
}
pub fn parse_json_v0_to_module(json: &str) -> Result<MirModule, String> {
let prog: ProgramV0 =
serde_json::from_str(json).map_err(|e| format!("invalid JSON v0: {}", e))?;
if prog.version != 0 || prog.kind != "Program" {
return Err("unsupported IR: expected {version:0, kind:\"Program\"}".into());
}
// Create module and main function
let mut module = MirModule::new("ny_json_v0".into());
let sig = FunctionSignature {
name: "main".into(),
params: vec![],
return_type: MirType::Integer,
effects: EffectMask::PURE,
};
let entry = BasicBlockId::new(0);
let mut f = MirFunction::new(sig, entry);
if prog.body.is_empty() {
return Err("empty body".into());
}
// Variable map for simple locals (Stage-2; currently minimal)
let mut var_map: std::collections::HashMap<String, crate::mir::ValueId> =
std::collections::HashMap::new();
let mut loop_stack: Vec<LoopContext> = Vec::new();
let start_bb = f.entry_block;
let end_bb =
lower_stmt_list_with_vars(&mut f, start_bb, &prog.body, &mut var_map, &mut loop_stack)?;
// Ensure function terminates: add `ret 0` to last un-terminated block (prefer end_bb else entry)
let need_default_ret = f.blocks.iter().any(|(_k, b)| !b.is_terminated());
if need_default_ret {
let target_bb = end_bb;
let dst_id = f.next_value_id();
if let Some(bb) = f.get_block_mut(target_bb) {
if !bb.is_terminated() {
bb.add_instruction(MirInstruction::Const {
dst: dst_id,
value: ConstValue::Integer(0),
});
bb.set_terminator(MirInstruction::Return {
value: Some(dst_id),
});
}
}
}
// Keep return type unknown to allow dynamic display (VM/Interpreter)
f.signature.return_type = MirType::Unknown;
module.add_function(f);
Ok(module)
}
fn next_block_id(f: &MirFunction) -> BasicBlockId {
let mut mx = 0u32;
for k in f.blocks.keys() {
if k.0 >= mx {
mx = k.0 + 1;
}
}
BasicBlockId::new(mx)
}
fn lower_expr(
f: &mut MirFunction,
cur_bb: BasicBlockId,
e: &ExprV0,
) -> Result<(crate::mir::ValueId, BasicBlockId), String> {
match e {
ExprV0::Int { value } => {
// Accept number or stringified digits
let ival: i64 = if let Some(n) = value.as_i64() {
n
} else if let Some(s) = value.as_str() {
s.parse().map_err(|_| "invalid int literal")?
} else {
return Err("invalid int literal".into());
};
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::Const {
dst,
value: ConstValue::Integer(ival),
});
}
Ok((dst, cur_bb))
}
ExprV0::Str { value } => {
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::Const {
dst,
value: ConstValue::String(value.clone()),
});
}
Ok((dst, cur_bb))
}
ExprV0::Bool { value } => {
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::Const {
dst,
value: ConstValue::Bool(*value),
});
}
Ok((dst, cur_bb))
}
ExprV0::Binary { op, lhs, rhs } => {
let (l, cur_after_l) = lower_expr(f, cur_bb, lhs)?;
let (r, cur_after_r) = lower_expr(f, cur_after_l, rhs)?;
let bop = match op.as_str() {
"+" => BinaryOp::Add,
"-" => BinaryOp::Sub,
"*" => BinaryOp::Mul,
"/" => BinaryOp::Div,
_ => return Err("unsupported op".into()),
};
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_after_r) {
bb.add_instruction(MirInstruction::BinOp {
dst,
op: bop,
lhs: l,
rhs: r,
});
}
Ok((dst, cur_after_r))
}
ExprV0::Extern {
iface,
method,
args,
} => {
let (arg_ids, cur2) = lower_args(f, cur_bb, args)?;
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur2) {
bb.add_instruction(MirInstruction::ExternCall {
dst: Some(dst),
iface_name: iface.clone(),
method_name: method.clone(),
args: arg_ids,
effects: EffectMask::IO,
});
}
Ok((dst, cur2))
}
ExprV0::Compare { op, lhs, rhs } => {
let (l, cur_after_l) = lower_expr(f, cur_bb, lhs)?;
let (r, cur_after_r) = lower_expr(f, cur_after_l, rhs)?;
let cop = match op.as_str() {
"==" => crate::mir::CompareOp::Eq,
"!=" => crate::mir::CompareOp::Ne,
"<" => crate::mir::CompareOp::Lt,
"<=" => crate::mir::CompareOp::Le,
">" => crate::mir::CompareOp::Gt,
">=" => crate::mir::CompareOp::Ge,
_ => return Err("unsupported compare op".into()),
};
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_after_r) {
bb.add_instruction(MirInstruction::Compare {
dst,
op: cop,
lhs: l,
rhs: r,
});
}
Ok((dst, cur_after_r))
}
ExprV0::Logical { op, lhs, rhs } => {
// Short-circuit boolean logic with branches (+phi or edge-copy)
let (l, cur_after_l) = lower_expr(f, cur_bb, lhs)?;
let rhs_bb = next_block_id(f);
let fall_bb = BasicBlockId::new(rhs_bb.0 + 1);
let merge_bb = BasicBlockId::new(rhs_bb.0 + 2);
f.add_block(crate::mir::BasicBlock::new(rhs_bb));
f.add_block(crate::mir::BasicBlock::new(fall_bb));
f.add_block(crate::mir::BasicBlock::new(merge_bb));
// Branch depending on op
let is_and = matches!(op.as_str(), "&&" | "and");
if let Some(bb) = f.get_block_mut(cur_after_l) {
if is_and {
bb.set_terminator(MirInstruction::Branch {
condition: l,
then_bb: rhs_bb,
else_bb: fall_bb,
});
} else {
// OR: if lhs true, go to fall_bb (true path), else evaluate rhs
bb.set_terminator(MirInstruction::Branch {
condition: l,
then_bb: fall_bb,
else_bb: rhs_bb,
});
}
}
// Telemetry: note short-circuit lowering
crate::jit::events::emit_lower(
serde_json::json!({
"id": "shortcircuit",
"op": if is_and { "and" } else { "or" },
"rhs_bb": rhs_bb.0,
"fall_bb": fall_bb.0,
"merge_bb": merge_bb.0
}),
"shortcircuit",
"<json_v0>",
);
if std::env::var("NYASH_CLI_VERBOSE").ok().as_deref() == Some("1") {
eprintln!(
"[bridge/logical] op={} rhs_bb={} fall_bb={} merge_bb={}",
if is_and { "and" } else { "or" },
rhs_bb.0,
fall_bb.0,
merge_bb.0
);
}
// false/true constant in fall_bb depending on op
let cdst = f.next_value_id();
if let Some(bb) = f.get_block_mut(fall_bb) {
let cval = if is_and {
ConstValue::Bool(false)
} else {
ConstValue::Bool(true)
};
bb.add_instruction(MirInstruction::Const {
dst: cdst,
value: cval,
});
bb.set_terminator(MirInstruction::Jump { target: merge_bb });
}
// evaluate rhs starting at rhs_bb and ensure the terminal block jumps to merge
let (rval, rhs_end) = lower_expr(f, rhs_bb, rhs)?;
if let Some(bb) = f.get_block_mut(rhs_end) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump { target: merge_bb });
}
}
if std::env::var("NYASH_CLI_VERBOSE").ok().as_deref() == Some("1") {
eprintln!(
"[bridge/logical] rhs_end={} jump->merge_bb={}",
rhs_end.0, merge_bb.0
);
}
// Merge: PHI または edge-copy で合流値を定義
let no_phi = crate::config::env::mir_no_phi();
let out = f.next_value_id();
if no_phi {
// Edge copies in predecessors
if let Some(bb) = f.get_block_mut(fall_bb) {
bb.add_instruction(MirInstruction::Copy { dst: out, src: cdst });
}
if let Some(bb) = f.get_block_mut(rhs_end) {
bb.add_instruction(MirInstruction::Copy { dst: out, src: rval });
}
} else if let Some(bb) = f.get_block_mut(merge_bb) {
let mut inputs: Vec<(BasicBlockId, ValueId)> = vec![(fall_bb, cdst)];
if rhs_end != fall_bb {
inputs.push((rhs_end, rval));
} else {
// Degenerate case: RHS ended in fall_bb (e.g., constant expression).
// Reuse the constant to keep PHI well-formed.
inputs.push((fall_bb, rval));
}
inputs.sort_by_key(|(bbid, _)| bbid.0);
bb.insert_instruction_after_phis(MirInstruction::Phi { dst: out, inputs });
}
Ok((out, merge_bb))
}
ExprV0::Call { name, args } => {
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
// Special: array literal lowering — Call{name:"array.of", args:[...]} → new ArrayBox(); push(...); result=array
if name == "array.of" {
// Create array first
let arr = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::NewBox {
dst: arr,
box_type: "ArrayBox".into(),
args: vec![],
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
// For each element: eval then push
let mut cur = cur_bb;
for e in args {
let (v, c) = lower_expr(f, cur, e)?;
cur = c;
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
let tmp = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::BoxCall {
dst: Some(tmp),
box_val: arr,
method: "push".into(),
method_id: None,
args: vec![v],
effects: EffectMask::READ,
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
}
return Ok((arr, cur));
}
// Special: map literal lowering — Call{name:"map.of", args:[k1, v1, k2, v2, ...]} → new MapBox(); set(k,v)...; result=map
if name == "map.of" {
let mapv = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::NewBox {
dst: mapv,
box_type: "MapBox".into(),
args: vec![],
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
let mut cur = cur_bb;
let mut it = args.iter();
while let Some(k) = it.next() {
if let Some(v) = it.next() {
let (kv, cur2) = lower_expr(f, cur, k)?;
cur = cur2;
let (vv, cur3) = lower_expr(f, cur, v)?;
cur = cur3;
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
let tmp = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::BoxCall {
dst: Some(tmp),
box_val: mapv,
method: "set".into(),
method_id: None,
args: vec![kv, vv],
effects: EffectMask::READ,
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
} else {
break;
}
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
return Ok((mapv, cur));
}
// Fallback: treat as normal dynamic call
let (arg_ids, cur) = lower_args(f, cur_bb, args)?;
let fun_val = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::Const {
dst: fun_val,
value: ConstValue::String(name.clone()),
});
}
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::Call {
dst: Some(dst),
func: fun_val,
args: arg_ids,
effects: EffectMask::READ,
});
}
Ok((dst, cur))
}
ExprV0::Method { recv, method, args } => {
// Heuristic: new ConsoleBox().println(x) → externcall env.console.log(x)
let recv_is_console_new =
matches!(&**recv, ExprV0::New { class, .. } if class == "ConsoleBox");
if recv_is_console_new && (method == "println" || method == "print" || method == "log")
{
let (arg_ids, cur2) = lower_args(f, cur_bb, args)?;
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur2) {
bb.add_instruction(MirInstruction::ExternCall {
dst: Some(dst),
iface_name: "env.console".into(),
method_name: "log".into(),
args: arg_ids,
effects: EffectMask::READ,
});
}
return Ok((dst, cur2));
}
let (recv_v, cur) = lower_expr(f, cur_bb, recv)?;
let (arg_ids, cur2) = lower_args(f, cur, args)?;
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur2) {
bb.add_instruction(MirInstruction::BoxCall {
dst: Some(dst),
box_val: recv_v,
method: method.clone(),
method_id: None,
args: arg_ids,
effects: EffectMask::READ,
});
}
Ok((dst, cur2))
}
ExprV0::New { class, args } => {
let (arg_ids, cur) = lower_args(f, cur_bb, args)?;
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::NewBox {
dst,
box_type: class.clone(),
args: arg_ids,
});
}
Ok((dst, cur))
}
ExprV0::Var { name } => Err(format!("undefined variable in this context: {}", name)),
ExprV0::Throw { expr } => {
let (exc, cur) = lower_expr(f, cur_bb, expr)?;
let (dst, cur) = lower_throw(f, cur, exc);
Ok((dst, cur))
}
}
}
fn lower_expr_with_vars(
f: &mut MirFunction,
cur_bb: BasicBlockId,
e: &ExprV0,
vars: &mut std::collections::HashMap<String, crate::mir::ValueId>,
) -> Result<(crate::mir::ValueId, BasicBlockId), String> {
match e {
ExprV0::Var { name } => {
if let Some(&vid) = vars.get(name) {
return Ok((vid, cur_bb));
}
if name == "me" {
// Optional gate: allow a dummy 'me' instance for Stage-2 JSON smoke
if std::env::var("NYASH_BRIDGE_ME_DUMMY").ok().as_deref() == Some("1") {
let class = std::env::var("NYASH_BRIDGE_ME_CLASS")
.unwrap_or_else(|_| "Main".to_string());
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::NewBox {
dst,
box_type: class,
args: vec![],
});
}
vars.insert("me".to_string(), dst);
return Ok((dst, cur_bb));
} else {
return Err("undefined 'me' outside box context (set NYASH_BRIDGE_ME_DUMMY=1 to inject placeholder)".into());
}
}
Err(format!("undefined variable: {}", name))
}
ExprV0::Throw { expr } => {
let (exc, cur) = lower_expr_with_vars(f, cur_bb, expr, vars)?;
let (dst, cur) = lower_throw(f, cur, exc);
Ok((dst, cur))
}
ExprV0::Call { name, args } => {
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
// Special: array literal lowering in vars context
if name == "array.of" {
let arr = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::NewBox {
dst: arr,
box_type: "ArrayBox".into(),
args: vec![],
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
let mut cur = cur_bb;
for e in args {
let (v, c) = lower_expr_with_vars(f, cur, e, vars)?;
cur = c;
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
let tmp = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::BoxCall {
dst: Some(tmp),
box_val: arr,
method: "push".into(),
method_id: None,
args: vec![v],
effects: EffectMask::READ,
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
}
return Ok((arr, cur));
}
// Special: map literal lowering in vars context
if name == "map.of" {
let mapv = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::NewBox {
dst: mapv,
box_type: "MapBox".into(),
args: vec![],
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
let mut cur = cur_bb;
let mut it = args.iter();
while let Some(k) = it.next() {
if let Some(v) = it.next() {
let (kv, cur2) = lower_expr_with_vars(f, cur, k, vars)?;
cur = cur2;
let (vv, cur3) = lower_expr_with_vars(f, cur, v, vars)?;
cur = cur3;
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
let tmp = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::BoxCall {
dst: Some(tmp),
box_val: mapv,
method: "set".into(),
method_id: None,
args: vec![kv, vv],
effects: EffectMask::READ,
});
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
} else {
break;
}
🔍 Research: GPT-5-Codex capabilities and GitHub PR integration ## Summary Investigated OpenAI's new GPT-5-Codex model and Codex GitHub PR review integration capabilities. ## GPT-5-Codex Analysis ### Benchmark Performance (Good) - SWE-bench Verified: 74.5% (vs GPT-5's 72.8%) - Refactoring tasks: 51.3% (vs GPT-5's 33.9%) - Code review: Higher developer ratings ### Real-World Issues (Concerning) - Users report degraded coding performance - Scripts that previously worked now fail - Less consistent than GPT-4.5 - Longer response times (minutes vs instant) - "Creatively and emotionally flat" - Basic errors (e.g., counting letters incorrectly) ### Key Finding Classic case of "optimizing for benchmarks vs real usability" - scores well on tests but performs poorly in practice. ## Codex GitHub PR Integration ### Setup Process 1. Enable MFA and connect GitHub account 2. Authorize Codex GitHub app for repos 3. Enable "Code review" in repository settings ### Usage Methods - **Manual**: Comment '@codex review' in PR - **Automatic**: Triggers when PR moves from draft to ready ### Current Limitations - One-way communication (doesn't respond to review comments) - Prefers creating new PRs over updating existing ones - Better for single-pass reviews than iterative feedback ## 'codex resume' Feature New session management capability: - Resume previous codex exec sessions - Useful for continuing long tasks across days - Maintains context from interrupted work 🐱 The investigation reveals that while GPT-5-Codex shows benchmark improvements, practical developer experience has declined - a reminder that metrics don't always reflect real-world utility\!
2025-09-16 16:28:25 +09:00
}
return Ok((mapv, cur));
}
// Lower args
let (arg_ids, cur) = lower_args_with_vars(f, cur_bb, args, vars)?;
// Encode as: const fun_name; call
let fun_val = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::Const {
dst: fun_val,
value: ConstValue::String(name.clone()),
});
}
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::Call {
dst: Some(dst),
func: fun_val,
args: arg_ids,
effects: EffectMask::READ,
});
}
Ok((dst, cur))
}
ExprV0::Method { recv, method, args } => {
let recv_is_console_new =
matches!(&**recv, ExprV0::New { class, .. } if class == "ConsoleBox");
if recv_is_console_new && (method == "println" || method == "print" || method == "log")
{
let (arg_ids, cur2) = lower_args_with_vars(f, cur_bb, args, vars)?;
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur2) {
bb.add_instruction(MirInstruction::ExternCall {
dst: Some(dst),
iface_name: "env.console".into(),
method_name: "log".into(),
args: arg_ids,
effects: EffectMask::READ,
});
}
return Ok((dst, cur2));
}
let (recv_v, cur) = lower_expr_with_vars(f, cur_bb, recv, vars)?;
let (arg_ids, cur2) = lower_args_with_vars(f, cur, args, vars)?;
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur2) {
bb.add_instruction(MirInstruction::BoxCall {
dst: Some(dst),
box_val: recv_v,
method: method.clone(),
method_id: None,
args: arg_ids,
effects: EffectMask::READ,
});
}
Ok((dst, cur2))
}
ExprV0::New { class, args } => {
let (arg_ids, cur) = lower_args_with_vars(f, cur_bb, args, vars)?;
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::NewBox {
dst,
box_type: class.clone(),
args: arg_ids,
});
}
Ok((dst, cur))
}
ExprV0::Binary { op, lhs, rhs } => {
let (l, cur_after_l) = lower_expr_with_vars(f, cur_bb, lhs, vars)?;
let (r, cur_after_r) = lower_expr_with_vars(f, cur_after_l, rhs, vars)?;
let bop = match op.as_str() {
"+" => BinaryOp::Add,
"-" => BinaryOp::Sub,
"*" => BinaryOp::Mul,
"/" => BinaryOp::Div,
_ => return Err("unsupported op".into()),
};
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_after_r) {
bb.add_instruction(MirInstruction::BinOp {
dst,
op: bop,
lhs: l,
rhs: r,
});
}
Ok((dst, cur_after_r))
}
ExprV0::Compare { op, lhs, rhs } => {
let (l, cur_after_l) = lower_expr_with_vars(f, cur_bb, lhs, vars)?;
let (r, cur_after_r) = lower_expr_with_vars(f, cur_after_l, rhs, vars)?;
let cop = match op.as_str() {
"==" => crate::mir::CompareOp::Eq,
"!=" => crate::mir::CompareOp::Ne,
"<" => crate::mir::CompareOp::Lt,
"<=" => crate::mir::CompareOp::Le,
">" => crate::mir::CompareOp::Gt,
">=" => crate::mir::CompareOp::Ge,
_ => return Err("unsupported compare op".into()),
};
let dst = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_after_r) {
bb.add_instruction(MirInstruction::Compare {
dst,
op: cop,
lhs: l,
rhs: r,
});
}
Ok((dst, cur_after_r))
}
ExprV0::Logical { op, lhs, rhs } => {
let (l, cur_after_l) = lower_expr_with_vars(f, cur_bb, lhs, vars)?;
let rhs_bb = next_block_id(f);
let fall_bb = BasicBlockId::new(rhs_bb.0 + 1);
let merge_bb = BasicBlockId::new(rhs_bb.0 + 2);
f.add_block(crate::mir::BasicBlock::new(rhs_bb));
f.add_block(crate::mir::BasicBlock::new(fall_bb));
f.add_block(crate::mir::BasicBlock::new(merge_bb));
let is_and = matches!(op.as_str(), "&&" | "and");
if let Some(bb) = f.get_block_mut(cur_after_l) {
if is_and {
bb.set_terminator(MirInstruction::Branch {
condition: l,
then_bb: rhs_bb,
else_bb: fall_bb,
});
} else {
bb.set_terminator(MirInstruction::Branch {
condition: l,
then_bb: fall_bb,
else_bb: rhs_bb,
});
}
}
let cdst = f.next_value_id();
if let Some(bb) = f.get_block_mut(fall_bb) {
let cval = if is_and {
ConstValue::Bool(false)
} else {
ConstValue::Bool(true)
};
bb.add_instruction(MirInstruction::Const {
dst: cdst,
value: cval,
});
bb.set_terminator(MirInstruction::Jump { target: merge_bb });
}
let (rval, rhs_end) = lower_expr_with_vars(f, rhs_bb, rhs, vars)?;
if let Some(bb) = f.get_block_mut(rhs_end) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump { target: merge_bb });
}
}
let out = f.next_value_id();
if let Some(bb) = f.get_block_mut(merge_bb) {
bb.insert_instruction_after_phis(MirInstruction::Phi {
dst: out,
inputs: vec![(rhs_end, rval), (fall_bb, cdst)],
});
}
Ok((out, merge_bb))
}
_ => lower_expr(f, cur_bb, e),
}
}
fn lower_stmt_with_vars(
f: &mut MirFunction,
cur_bb: BasicBlockId,
s: &StmtV0,
vars: &mut std::collections::HashMap<String, crate::mir::ValueId>,
loop_stack: &mut Vec<LoopContext>,
) -> Result<BasicBlockId, String> {
match s {
StmtV0::Return { expr } => {
let (v, cur) = lower_expr_with_vars(f, cur_bb, expr, vars)?;
if let Some(bb) = f.get_block_mut(cur) {
bb.set_terminator(MirInstruction::Return { value: Some(v) });
}
Ok(cur)
}
StmtV0::Extern {
iface,
method,
args,
} => {
let (arg_ids, cur) = lower_args_with_vars(f, cur_bb, args, vars)?;
if let Some(bb) = f.get_block_mut(cur) {
bb.add_instruction(MirInstruction::ExternCall {
dst: None,
iface_name: iface.clone(),
method_name: method.clone(),
args: arg_ids,
effects: EffectMask::IO,
});
}
Ok(cur)
}
StmtV0::Expr { expr } => {
let (_v, cur) = lower_expr_with_vars(f, cur_bb, expr, vars)?;
Ok(cur)
}
StmtV0::Local { name, expr } => {
let (v, cur) = lower_expr_with_vars(f, cur_bb, expr, vars)?;
vars.insert(name.clone(), v);
Ok(cur)
}
StmtV0::Break => {
if let Some(ctx) = loop_stack.last().copied() {
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.set_terminator(MirInstruction::Jump {
target: ctx.exit_bb,
});
}
crate::jit::events::emit_lower(
serde_json::json!({
"id": "loop_break",
"exit_bb": ctx.exit_bb.0,
"decision": "lower",
}),
"loop",
"<json_v0>",
);
} else if std::env::var("NYASH_CLI_VERBOSE").ok().as_deref() == Some("1") {
eprintln!("[bridge/break] ignoring break outside loop context");
}
Ok(cur_bb)
}
StmtV0::Continue => {
if let Some(ctx) = loop_stack.last().copied() {
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.set_terminator(MirInstruction::Jump {
target: ctx.cond_bb,
});
}
crate::jit::events::emit_lower(
serde_json::json!({
"id": "loop_continue",
"cond_bb": ctx.cond_bb.0,
"decision": "lower",
}),
"loop",
"<json_v0>",
);
} else if std::env::var("NYASH_CLI_VERBOSE").ok().as_deref() == Some("1") {
eprintln!("[bridge/continue] ignoring continue outside loop context");
}
Ok(cur_bb)
}
StmtV0::Try {
try_body,
catches,
finally,
} => {
let try_enabled = std::env::var("NYASH_BRIDGE_TRY_ENABLE").ok().as_deref() == Some("1");
if !try_enabled || catches.is_empty() || catches.len() > 1 {
let mut tmp_vars = vars.clone();
let mut next_bb =
lower_stmt_list_with_vars(f, cur_bb, try_body, &mut tmp_vars, loop_stack)?;
if !finally.is_empty() {
next_bb =
lower_stmt_list_with_vars(f, next_bb, finally, &mut tmp_vars, loop_stack)?;
}
*vars = tmp_vars;
return Ok(next_bb);
}
let base_vars = vars.clone();
let try_bb = next_block_id(f);
f.add_block(crate::mir::BasicBlock::new(try_bb));
let catch_clause = &catches[0];
let catch_bb = next_block_id(f);
f.add_block(crate::mir::BasicBlock::new(catch_bb));
let finally_bb = if !finally.is_empty() {
let id = next_block_id(f);
f.add_block(crate::mir::BasicBlock::new(id));
Some(id)
} else {
None
};
let exit_bb = next_block_id(f);
f.add_block(crate::mir::BasicBlock::new(exit_bb));
let handler_target = finally_bb.unwrap_or(exit_bb);
let exception_value = f.next_value_id();
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::Catch {
exception_type: catch_clause.type_hint.clone(),
exception_value,
handler_bb: catch_bb,
});
bb.set_terminator(MirInstruction::Jump { target: try_bb });
}
let mut try_vars = vars.clone();
let try_end =
lower_stmt_list_with_vars(f, try_bb, try_body, &mut try_vars, loop_stack)?;
if let Some(bb) = f.get_block_mut(try_end) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump {
target: handler_target,
});
}
}
let try_branch_vars = try_vars.clone();
let mut catch_vars = base_vars.clone();
if let Some(param) = &catch_clause.param {
catch_vars.insert(param.clone(), exception_value);
}
let catch_end = lower_stmt_list_with_vars(
f,
catch_bb,
&catch_clause.body,
&mut catch_vars,
loop_stack,
)?;
if let Some(param) = &catch_clause.param {
catch_vars.remove(param);
}
if let Some(bb) = f.get_block_mut(catch_end) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump {
target: handler_target,
});
}
}
let catch_branch_vars = catch_vars.clone();
use std::collections::HashSet;
let mut branch_vars = vec![(try_end, try_branch_vars), (catch_end, catch_branch_vars)];
let merge_target = handler_target;
if let Some(finally_block) = finally_bb {
// ensure finally block exists before inserting phi
let names: HashSet<String> = {
let mut set: HashSet<String> = base_vars.keys().cloned().collect();
for (_, map) in &branch_vars {
set.extend(map.keys().cloned());
}
set
};
let mut merged_vars = base_vars.clone();
let mut phi_entries: Vec<(ValueId, Vec<(BasicBlockId, ValueId)>)> = Vec::new();
for name in names {
let mut inputs: Vec<(BasicBlockId, ValueId)> = Vec::new();
for (bbid, map) in &branch_vars {
if let Some(&val) = map.get(&name) {
inputs.push((*bbid, val));
}
}
if inputs.is_empty() {
if let Some(&base_val) = base_vars.get(&name) {
merged_vars.insert(name.clone(), base_val);
}
continue;
}
let unique: HashSet<ValueId> = inputs.iter().map(|(_, v)| *v).collect();
if unique.len() == 1 {
merged_vars.insert(name.clone(), inputs[0].1);
continue;
}
let dst = f.next_value_id();
inputs.sort_by_key(|(bbid, _)| bbid.0);
phi_entries.push((dst, inputs));
merged_vars.insert(name.clone(), dst);
}
if let Some(bb) = f.get_block_mut(finally_block) {
for (dst, inputs) in phi_entries {
bb.insert_instruction_after_phis(MirInstruction::Phi { dst, inputs });
}
}
let mut finally_vars = merged_vars.clone();
let final_end = lower_stmt_list_with_vars(
f,
finally_block,
finally,
&mut finally_vars,
loop_stack,
)?;
if let Some(bb) = f.get_block_mut(final_end) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump { target: exit_bb });
}
}
*vars = finally_vars;
Ok(exit_bb)
} else {
let names: HashSet<String> = {
let mut set: HashSet<String> = base_vars.keys().cloned().collect();
for (_, map) in &branch_vars {
set.extend(map.keys().cloned());
}
set
};
let mut merged_vars = base_vars.clone();
let mut phi_entries: Vec<(ValueId, Vec<(BasicBlockId, ValueId)>)> = Vec::new();
for name in names {
let mut inputs: Vec<(BasicBlockId, ValueId)> = Vec::new();
for (bbid, map) in &branch_vars {
if let Some(&val) = map.get(&name) {
inputs.push((*bbid, val));
}
}
if inputs.is_empty() {
if let Some(&base_val) = base_vars.get(&name) {
merged_vars.insert(name.clone(), base_val);
}
continue;
}
let unique: HashSet<ValueId> = inputs.iter().map(|(_, v)| *v).collect();
if unique.len() == 1 {
merged_vars.insert(name.clone(), inputs[0].1);
continue;
}
let dst = f.next_value_id();
inputs.sort_by_key(|(bbid, _)| bbid.0);
phi_entries.push((dst, inputs));
merged_vars.insert(name.clone(), dst);
}
if let Some(bb) = f.get_block_mut(exit_bb) {
for (dst, inputs) in phi_entries {
bb.insert_instruction_after_phis(MirInstruction::Phi { dst, inputs });
}
}
*vars = merged_vars;
Ok(exit_bb)
}
}
StmtV0::If { cond, then, r#else } => {
// Lower condition first
let (cval, cur) = lower_expr_with_vars(f, cur_bb, cond, vars)?;
// Create then/else/merge blocks
let then_bb = next_block_id(f);
let else_bb = BasicBlockId::new(then_bb.0 + 1);
let merge_bb = BasicBlockId::new(then_bb.0 + 2);
f.add_block(crate::mir::BasicBlock::new(then_bb));
f.add_block(crate::mir::BasicBlock::new(else_bb));
f.add_block(crate::mir::BasicBlock::new(merge_bb));
// Branch to then/else
if let Some(bb) = f.get_block_mut(cur) {
bb.set_terminator(MirInstruction::Branch {
condition: cval,
then_bb,
else_bb,
});
}
// Clone current vars as branch-local maps
let base_vars = vars.clone();
let mut then_vars = base_vars.clone();
let tend = lower_stmt_list_with_vars(f, then_bb, then, &mut then_vars, loop_stack)?;
if let Some(bb) = f.get_block_mut(tend) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump { target: merge_bb });
}
}
let (else_end_pred, else_vars) = if let Some(elses) = r#else {
let mut ev = base_vars.clone();
let eend = lower_stmt_list_with_vars(f, else_bb, elses, &mut ev, loop_stack)?;
if let Some(bb) = f.get_block_mut(eend) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump { target: merge_bb });
}
}
(eend, ev)
} else {
// No else: empty path falls through with base vars
if let Some(bb) = f.get_block_mut(else_bb) {
bb.set_terminator(MirInstruction::Jump { target: merge_bb });
}
(else_bb, base_vars.clone())
};
// Merge at then/else predecessorsPHI or edge-copy
use std::collections::HashSet;
let no_phi = crate::config::env::mir_no_phi();
let mut names: HashSet<String> = base_vars.keys().cloned().collect();
for k in then_vars.keys() { names.insert(k.clone()); }
for k in else_vars.keys() { names.insert(k.clone()); }
for name in names {
let tv = then_vars.get(&name).copied();
let ev = else_vars.get(&name).copied();
let exists_base = base_vars.contains_key(&name);
match (tv, ev, exists_base) {
(Some(tval), Some(eval), _) => {
let merged = if tval == eval { tval } else {
let dst = f.next_value_id();
if no_phi {
if let Some(bb) = f.get_block_mut(tend) { bb.add_instruction(MirInstruction::Copy { dst, src: tval }); }
if let Some(bb) = f.get_block_mut(else_end_pred) { bb.add_instruction(MirInstruction::Copy { dst, src: eval }); }
} else if let Some(bb) = f.get_block_mut(merge_bb) {
bb.insert_instruction_after_phis(MirInstruction::Phi { dst, inputs: vec![(tend, tval), (else_end_pred, eval)] });
}
dst
};
vars.insert(name, merged);
}
(Some(tval), None, true) => {
if let Some(&bval) = base_vars.get(&name) {
let merged = if tval == bval { tval } else {
let dst = f.next_value_id();
if no_phi {
if let Some(bb) = f.get_block_mut(tend) { bb.add_instruction(MirInstruction::Copy { dst, src: tval }); }
if let Some(bb) = f.get_block_mut(else_end_pred) { bb.add_instruction(MirInstruction::Copy { dst, src: bval }); }
} else if let Some(bb) = f.get_block_mut(merge_bb) {
bb.insert_instruction_after_phis(MirInstruction::Phi { dst, inputs: vec![(tend, tval), (else_end_pred, bval)] });
}
dst
};
vars.insert(name, merged);
}
}
(None, Some(eval), true) => {
if let Some(&bval) = base_vars.get(&name) {
let merged = if eval == bval { eval } else {
let dst = f.next_value_id();
if no_phi {
if let Some(bb) = f.get_block_mut(tend) { bb.add_instruction(MirInstruction::Copy { dst, src: bval }); }
if let Some(bb) = f.get_block_mut(else_end_pred) { bb.add_instruction(MirInstruction::Copy { dst, src: eval }); }
} else if let Some(bb) = f.get_block_mut(merge_bb) {
bb.insert_instruction_after_phis(MirInstruction::Phi { dst, inputs: vec![(tend, bval), (else_end_pred, eval)] });
}
dst
};
vars.insert(name, merged);
}
}
_ => {}
}
}
Ok(merge_bb)
}
StmtV0::Loop { cond, body } => {
// Create loop blocks
let cond_bb = next_block_id(f);
let body_bb = BasicBlockId::new(cond_bb.0 + 1);
let exit_bb = BasicBlockId::new(cond_bb.0 + 2);
f.add_block(crate::mir::BasicBlock::new(cond_bb));
f.add_block(crate::mir::BasicBlock::new(body_bb));
f.add_block(crate::mir::BasicBlock::new(exit_bb));
// Preheader jump into cond
if let Some(bb) = f.get_block_mut(cur_bb) {
if !bb.is_terminated() {
bb.add_instruction(MirInstruction::Jump { target: cond_bb });
}
}
// Snapshot base vars and set up merged ids for loop-carried vars
let no_phi = crate::config::env::mir_no_phi();
let base_vars = vars.clone();
let orig_names: Vec<String> = base_vars.keys().cloned().collect();
let mut phi_map: std::collections::HashMap<String, crate::mir::ValueId> =
std::collections::HashMap::new();
for name in &orig_names {
if let Some(&bval) = base_vars.get(name) {
let dst = f.next_value_id();
if no_phi {
// Preheader edge-copycur_bb -> cond
if let Some(bb) = f.get_block_mut(cur_bb) {
bb.add_instruction(MirInstruction::Copy { dst, src: bval });
}
} else if let Some(bb) = f.get_block_mut(cond_bb) {
// Initial incoming from preheader via PHI
bb.insert_instruction_after_phis(MirInstruction::Phi { dst, inputs: vec![(cur_bb, bval)] });
}
phi_map.insert(name.clone(), dst);
}
}
// Redirect current vars to PHIs for use in cond/body
for (name, &phi) in &phi_map {
vars.insert(name.clone(), phi);
}
// Lower condition using phi-backed vars
let (cval, _cend) = lower_expr_with_vars(f, cond_bb, cond, vars)?;
if let Some(bb) = f.get_block_mut(cond_bb) {
bb.set_terminator(MirInstruction::Branch {
condition: cval,
then_bb: body_bb,
else_bb: exit_bb,
});
}
// Lower body; record end block and body-out vars
let mut body_vars = vars.clone();
loop_stack.push(LoopContext { cond_bb, exit_bb });
let bend_res = lower_stmt_list_with_vars(f, body_bb, body, &mut body_vars, loop_stack);
loop_stack.pop();
let bend = bend_res?;
if let Some(bb) = f.get_block_mut(bend) {
if !bb.is_terminated() {
bb.set_terminator(MirInstruction::Jump { target: cond_bb });
}
}
// Wire second incoming from latch (body end)
let backedge_to_cond = matches!(
f.blocks.get(&bend).and_then(|bb| bb.terminator.as_ref()),
Some(MirInstruction::Jump { target, .. }) if *target == cond_bb
);
if backedge_to_cond {
if no_phi {
// Latch edge-copybend -> cond
for (name, &phi_dst) in &phi_map {
if let Some(&latch_val) = body_vars.get(name) {
if let Some(bb) = f.get_block_mut(bend) {
bb.add_instruction(MirInstruction::Copy { dst: phi_dst, src: latch_val });
}
}
}
} else if let Some(bb) = f.get_block_mut(cond_bb) {
for (name, &phi_dst) in &phi_map {
if let Some(&latch_val) = body_vars.get(name) {
for inst in &mut bb.instructions {
if let MirInstruction::Phi { dst, inputs } = inst {
if *dst == phi_dst {
inputs.push((bend, latch_val));
break;
}
}
}
}
}
}
} else if std::env::var("NYASH_CLI_VERBOSE").ok().as_deref() == Some("1") {
eprintln!(
"[bridge/loop] skipped latch bb{} -> cond bb{}",
bend.0, cond_bb.0
);
}
// After the loop, keep vars mapped to the PHI values (current loop state)
for (name, &phi) in &phi_map {
vars.insert(name.clone(), phi);
}
Ok(exit_bb)
}
}
}
fn lower_stmt_list_with_vars(
f: &mut MirFunction,
start_bb: BasicBlockId,
stmts: &[StmtV0],
vars: &mut std::collections::HashMap<String, crate::mir::ValueId>,
loop_stack: &mut Vec<LoopContext>,
) -> Result<BasicBlockId, String> {
let mut cur = start_bb;
for s in stmts {
cur = lower_stmt_with_vars(f, cur, s, vars, loop_stack)?;
if let Some(bb) = f.blocks.get(&cur) {
if bb.is_terminated() {
break;
}
}
}
Ok(cur)
}
fn lower_args_with_vars(
f: &mut MirFunction,
cur_bb: BasicBlockId,
args: &[ExprV0],
vars: &mut std::collections::HashMap<String, crate::mir::ValueId>,
) -> Result<(Vec<crate::mir::ValueId>, BasicBlockId), String> {
let mut out = Vec::with_capacity(args.len());
let mut cur = cur_bb;
for a in args {
let (v, c) = lower_expr_with_vars(f, cur, a, vars)?;
out.push(v);
cur = c;
}
Ok((out, cur))
}
fn lower_args(
f: &mut MirFunction,
cur_bb: BasicBlockId,
args: &[ExprV0],
) -> Result<(Vec<crate::mir::ValueId>, BasicBlockId), String> {
let mut out = Vec::with_capacity(args.len());
let mut cur = cur_bb;
for a in args {
let (v, c) = lower_expr(f, cur, a)?;
out.push(v);
cur = c;
}
Ok((out, cur))
}
pub fn maybe_dump_mir(module: &MirModule) {
if std::env::var("NYASH_CLI_VERBOSE").ok().as_deref() == Some("1") {
let p = MirPrinter::new();
println!("{}", p.print_module(module));
}
}
#[cfg(test)]
mod tests {
use super::parse_json_v0_to_module;
use crate::mir::{BasicBlockId, MirInstruction, MirModule};
fn block_terminator(module: &MirModule, block: BasicBlockId) -> MirInstruction {
module
.get_function("main")
.unwrap()
.get_block(block)
.and_then(|bb| bb.terminator.clone())
.expect("terminator")
}
#[test]
fn stage3_break_jumps_to_exit() {
let json = "{\"version\":0,\"kind\":\"Program\",\"body\":[{\"type\":\"Loop\",\"cond\":{\"type\":\"Bool\",\"value\":true},\"body\":[{\"type\":\"Break\"}]},{\"type\":\"Return\",\"expr\":{\"type\":\"Int\",\"value\":0}}]}";
let module = parse_json_v0_to_module(json).unwrap();
match block_terminator(&module, BasicBlockId::new(2)) {
MirInstruction::Jump { target, .. } => assert_eq!(target, BasicBlockId::new(3)),
other => panic!("expected jump, got {:?}", other),
}
module.verify().unwrap();
}
#[test]
fn stage3_continue_jumps_to_head() {
let json = "{\"version\":0,\"kind\":\"Program\",\"body\":[{\"type\":\"Loop\",\"cond\":{\"type\":\"Bool\",\"value\":true},\"body\":[{\"type\":\"Continue\"}]},{\"type\":\"Return\",\"expr\":{\"type\":\"Int\",\"value\":0}}]}";
let module = parse_json_v0_to_module(json).unwrap();
match block_terminator(&module, BasicBlockId::new(2)) {
MirInstruction::Jump { target, .. } => assert_eq!(target, BasicBlockId::new(1)),
other => panic!("expected jump, got {:?}", other),
}
module.verify().unwrap();
}
}
// ========== Direct bridge (source → JSON v0 → MIR) ==========
#[derive(Clone, Debug)]
enum Tok {
Return,
Int(i64),
Plus,
Minus,
Star,
Slash,
LParen,
RParen,
Eof,
}
fn lex(input: &str) -> Result<Vec<Tok>, String> {
let bytes = input.as_bytes();
let mut i = 0usize;
let n = bytes.len();
let mut toks = Vec::new();
while i < n {
let c = bytes[i] as char;
// Treat semicolon as whitespace (Stage-1 minimal ASI: optional ';')
if c.is_whitespace() || c == ';' {
i += 1;
continue;
}
match c {
'+' => {
toks.push(Tok::Plus);
i += 1;
}
'-' => {
toks.push(Tok::Minus);
i += 1;
}
'*' => {
toks.push(Tok::Star);
i += 1;
}
'/' => {
toks.push(Tok::Slash);
i += 1;
}
'(' => {
toks.push(Tok::LParen);
i += 1;
}
')' => {
toks.push(Tok::RParen);
i += 1;
}
'0'..='9' => {
let start = i;
while i < n {
let cc = bytes[i] as char;
if cc.is_ascii_digit() {
i += 1;
} else {
break;
}
}
let s = std::str::from_utf8(&bytes[start..i]).unwrap();
let v: i64 = s.parse().map_err(|_| "invalid int")?;
toks.push(Tok::Int(v));
}
'r' => {
// return
if i + 6 <= n && &input[i..i + 6] == "return" {
toks.push(Tok::Return);
i += 6;
} else {
return Err("unexpected 'r'".into());
}
}
_ => return Err(format!("unexpected char '{}'", c)),
}
}
toks.push(Tok::Eof);
Ok(toks)
}
struct P {
toks: Vec<Tok>,
pos: usize,
}
impl P {
fn new(toks: Vec<Tok>) -> Self {
Self { toks, pos: 0 }
}
fn peek(&self) -> &Tok {
self.toks.get(self.pos).unwrap()
}
fn next(&mut self) -> Tok {
let t = self.toks.get(self.pos).unwrap().clone();
self.pos += 1;
t
}
fn expect_return(&mut self) -> Result<(), String> {
match self.next() {
Tok::Return => Ok(()),
_ => Err("expected 'return'".into()),
}
}
fn parse_program(&mut self) -> Result<ExprV0, String> {
self.expect_return()?;
self.parse_expr()
}
fn parse_expr(&mut self) -> Result<ExprV0, String> {
let mut left = self.parse_term()?;
loop {
match self.peek() {
Tok::Plus => {
self.next();
let r = self.parse_term()?;
left = ExprV0::Binary {
op: "+".into(),
lhs: Box::new(left),
rhs: Box::new(r),
};
}
Tok::Minus => {
self.next();
let r = self.parse_term()?;
left = ExprV0::Binary {
op: "-".into(),
lhs: Box::new(left),
rhs: Box::new(r),
};
}
_ => break,
}
}
Ok(left)
}
fn parse_term(&mut self) -> Result<ExprV0, String> {
let mut left = self.parse_factor()?;
loop {
match self.peek() {
Tok::Star => {
self.next();
let r = self.parse_factor()?;
left = ExprV0::Binary {
op: "*".into(),
lhs: Box::new(left),
rhs: Box::new(r),
};
}
Tok::Slash => {
self.next();
let r = self.parse_factor()?;
left = ExprV0::Binary {
op: "/".into(),
lhs: Box::new(left),
rhs: Box::new(r),
};
}
_ => break,
}
}
Ok(left)
}
fn parse_factor(&mut self) -> Result<ExprV0, String> {
match self.next() {
Tok::Int(v) => Ok(ExprV0::Int {
value: serde_json::Value::from(v),
}),
Tok::LParen => {
let e = self.parse_expr()?;
match self.next() {
Tok::RParen => Ok(e),
_ => Err(") expected".into()),
}
}
_ => Err("factor expected".into()),
}
}
}
pub fn parse_source_v0_to_json(input: &str) -> Result<String, String> {
let toks = lex(input)?;
let mut p = P::new(toks);
let expr = p.parse_program()?;
let prog = ProgramV0 {
version: 0,
kind: "Program".into(),
body: vec![StmtV0::Return { expr }],
};
serde_json::to_string(&prog).map_err(|e| e.to_string())
}
pub fn parse_source_v0_to_module(input: &str) -> Result<MirModule, String> {
let json = parse_source_v0_to_json(input)?;
if std::env::var("NYASH_DUMP_JSON_IR").ok().as_deref() == Some("1") {
println!("{}", json);
}
parse_json_v0_to_module(&json)
}