core/codegen: refactor gen_{for,comprehension} to match on iter type

This commit is contained in:
lyken 2024-08-01 12:25:10 +08:00
parent 318a675ea6
commit b20779a7f4
3 changed files with 235 additions and 191 deletions

View File

@ -995,8 +995,10 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
ctx.builder.position_at_end(init_bb);
let Comprehension { target, iter, ifs, .. } = &generators[0];
let iter_ty = iter.custom.unwrap();
let iter_val = if let Some(v) = generator.gen_expr(ctx, iter)? {
v.to_basic_value_enum(ctx, generator, iter.custom.unwrap())?
v.to_basic_value_enum(ctx, generator, iter_ty)?
} else {
for bb in [test_bb, body_bb, cont_bb] {
ctx.builder.position_at_end(bb);
@ -1014,10 +1016,12 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
ctx.builder.build_store(index, zero_size_t).unwrap();
let elem_ty = ctx.get_llvm_type(generator, elt.custom.unwrap());
let is_range = ctx.unifier.unioned(iter.custom.unwrap(), ctx.primitives.range);
let list;
if is_range {
match &*ctx.unifier.get_ty(iter_ty) {
TypeEnum::TObj { obj_id, .. }
if *obj_id == ctx.primitives.range.obj_id(&ctx.unifier).unwrap() =>
{
let iter_val = RangeValue::from_ptr_val(iter_val.into_pointer_value(), Some("range"));
let (start, stop, step) = destructure_range(ctx, iter_val);
let diff = ctx.builder.build_int_sub(stop, start, "diff").unwrap();
@ -1025,7 +1029,8 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
// the length may be 1 more than the actual length if the division is exact, but the
// length is a upper bound only anyway so it does not matter.
let length = ctx.builder.build_int_signed_div(diff, step, "div").unwrap();
let length = ctx.builder.build_int_add(length, int32.const_int(1, false), "add1").unwrap();
let length =
ctx.builder.build_int_add(length, int32.const_int(1, false), "add1").unwrap();
// in case length is non-positive
let is_valid =
ctx.builder.build_int_compare(IntPredicate::SGT, length, zero_32, "check").unwrap();
@ -1034,7 +1039,9 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
.builder
.build_select(
is_valid,
ctx.builder.build_int_z_extend_or_bit_cast(length, size_t, "z_ext_len").unwrap(),
ctx.builder
.build_int_z_extend_or_bit_cast(length, size_t, "z_ext_len")
.unwrap(),
zero_size_t,
"listcomp.alloc_size",
)
@ -1053,7 +1060,11 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
.unwrap();
ctx.builder
.build_conditional_branch(gen_in_range_check(ctx, start, stop, step), test_bb, cont_bb)
.build_conditional_branch(
gen_in_range_check(ctx, start, stop, step),
test_bb,
cont_bb,
)
.unwrap();
ctx.builder.position_at_end(test_bb);
@ -1068,11 +1079,18 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
.unwrap();
ctx.builder.build_store(i, tmp).unwrap();
ctx.builder
.build_conditional_branch(gen_in_range_check(ctx, tmp, stop, step), body_bb, cont_bb)
.build_conditional_branch(
gen_in_range_check(ctx, tmp, stop, step),
body_bb,
cont_bb,
)
.unwrap();
ctx.builder.position_at_end(body_bb);
} else {
}
TypeEnum::TObj { obj_id, .. }
if *obj_id == ctx.primitives.list.obj_id(&ctx.unifier).unwrap() =>
{
let length = ctx
.build_gep_and_load(
iter_val.into_pointer_value(),
@ -1088,7 +1106,8 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
ctx.builder.build_unconditional_branch(test_bb).unwrap();
ctx.builder.position_at_end(test_bb);
let tmp = ctx.builder.build_load(counter, "i").map(BasicValueEnum::into_int_value).unwrap();
let tmp =
ctx.builder.build_load(counter, "i").map(BasicValueEnum::into_int_value).unwrap();
let tmp = ctx.builder.build_int_add(tmp, size_t.const_int(1, false), "inc").unwrap();
ctx.builder.build_store(counter, tmp).unwrap();
let cmp = ctx.builder.build_int_compare(IntPredicate::SLT, tmp, length, "cmp").unwrap();
@ -1105,6 +1124,13 @@ pub fn gen_comprehension<'ctx, G: CodeGenerator>(
let val = ctx.build_gep_and_load(arr_ptr, &[tmp], Some("val"));
generator.gen_assign(ctx, target, val.into())?;
}
_ => {
panic!(
"unsupported list comprehension iterator type: {}",
ctx.unifier.stringify(iter_ty)
);
}
}
// Emits the content of `cont_bb`
let emit_cont_bb =

View File

@ -315,9 +315,6 @@ pub fn gen_for<G: CodeGenerator>(
let orelse_bb =
if orelse.is_empty() { cont_bb } else { ctx.ctx.append_basic_block(current, "for.orelse") };
// Whether the iterable is a range() expression
let is_iterable_range_expr = ctx.unifier.unioned(iter.custom.unwrap(), ctx.primitives.range);
// The BB containing the increment expression
let incr_bb = ctx.ctx.append_basic_block(current, "for.incr");
// The BB containing the loop condition check
@ -326,17 +323,23 @@ pub fn gen_for<G: CodeGenerator>(
// store loop bb information and restore it later
let loop_bb = ctx.loop_target.replace((incr_bb, cont_bb));
let iter_ty = iter.custom.unwrap();
let iter_val = if let Some(v) = generator.gen_expr(ctx, iter)? {
v.to_basic_value_enum(ctx, generator, iter.custom.unwrap())?
v.to_basic_value_enum(ctx, generator, iter_ty)?
} else {
return Ok(());
};
if is_iterable_range_expr {
match &*ctx.unifier.get_ty(iter_ty) {
TypeEnum::TObj { obj_id, .. }
if *obj_id == ctx.primitives.range.obj_id(&ctx.unifier).unwrap() =>
{
let iter_val = RangeValue::from_ptr_val(iter_val.into_pointer_value(), Some("range"));
// Internal variable for loop; Cannot be assigned
let i = generator.gen_var_alloc(ctx, int32.into(), Some("for.i.addr"))?;
// Variable declared in "target" expression of the loop; Can be reassigned *or* shadowed
let Some(target_i) = generator.gen_store_target(ctx, target, Some("for.target.addr"))?
let Some(target_i) =
generator.gen_store_target(ctx, target, Some("for.target.addr"))?
else {
unreachable!()
};
@ -345,8 +348,10 @@ pub fn gen_for<G: CodeGenerator>(
ctx.builder.build_store(i, start).unwrap();
// Check "If step is zero, ValueError is raised."
let rangenez =
ctx.builder.build_int_compare(IntPredicate::NE, step, int32.const_zero(), "").unwrap();
let rangenez = ctx
.builder
.build_int_compare(IntPredicate::NE, step, int32.const_zero(), "")
.unwrap();
ctx.make_assert(
generator,
rangenez,
@ -363,7 +368,10 @@ pub fn gen_for<G: CodeGenerator>(
.build_conditional_branch(
gen_in_range_check(
ctx,
ctx.builder.build_load(i, "").map(BasicValueEnum::into_int_value).unwrap(),
ctx.builder
.build_load(i, "")
.map(BasicValueEnum::into_int_value)
.unwrap(),
stop,
step,
),
@ -393,7 +401,10 @@ pub fn gen_for<G: CodeGenerator>(
)
.unwrap();
generator.gen_block(ctx, body.iter())?;
} else {
}
TypeEnum::TObj { obj_id, .. }
if *obj_id == ctx.primitives.list.obj_id(&ctx.unifier).unwrap() =>
{
let index_addr = generator.gen_var_alloc(ctx, size_t.into(), Some("for.index.addr"))?;
ctx.builder.build_store(index_addr, size_t.const_zero()).unwrap();
let len = ctx
@ -431,9 +442,14 @@ pub fn gen_for<G: CodeGenerator>(
.map(BasicValueEnum::into_int_value)
.unwrap();
let val = ctx.build_gep_and_load(arr_ptr, &[index], Some("val"));
generator.gen_assign(ctx, target, val.into())?;
generator.gen_block(ctx, body.iter())?;
}
_ => {
panic!("unsupported for loop iterator type: {}", ctx.unifier.stringify(iter_ty));
}
}
for (k, (_, _, counter)) in &var_assignment {
let (_, static_val, counter2) = ctx.var_assignment.get_mut(k).unwrap();

View File

@ -100,16 +100,18 @@ pub struct Inferencer<'a> {
pub in_handler: bool,
}
type InferenceError = HashSet<String>;
struct NaiveFolder();
impl Fold<()> for NaiveFolder {
type TargetU = Option<Type>;
type Error = HashSet<String>;
type Error = InferenceError;
fn map_user(&mut self, (): ()) -> Result<Self::TargetU, Self::Error> {
Ok(None)
}
}
fn report_error<T>(msg: &str, location: Location) -> Result<T, HashSet<String>> {
fn report_error<T>(msg: &str, location: Location) -> Result<T, InferenceError> {
Err(HashSet::from([format!("{msg} at {location}")]))
}
@ -117,13 +119,13 @@ fn report_type_error<T>(
kind: TypeErrorKind,
loc: Option<Location>,
unifier: &Unifier,
) -> Result<T, HashSet<String>> {
) -> Result<T, InferenceError> {
Err(HashSet::from([TypeError::new(kind, loc).to_display(unifier).to_string()]))
}
impl<'a> Fold<()> for Inferencer<'a> {
type TargetU = Option<Type>;
type Error = HashSet<String>;
type Error = InferenceError;
fn map_user(&mut self, (): ()) -> Result<Self::TargetU, Self::Error> {
Ok(None)
@ -612,22 +614,22 @@ impl<'a> Fold<()> for Inferencer<'a> {
}
}
type InferenceResult = Result<Type, HashSet<String>>;
type InferenceResult = Result<Type, InferenceError>;
impl<'a> Inferencer<'a> {
/// Constrain a <: b
/// Currently implemented as unification
fn constrain(&mut self, a: Type, b: Type, location: &Location) -> Result<(), HashSet<String>> {
fn constrain(&mut self, a: Type, b: Type, location: &Location) -> Result<(), InferenceError> {
self.unify(a, b, location)
}
fn unify(&mut self, a: Type, b: Type, location: &Location) -> Result<(), HashSet<String>> {
fn unify(&mut self, a: Type, b: Type, location: &Location) -> Result<(), InferenceError> {
self.unifier.unify(a, b).map_err(|e| {
HashSet::from([e.at(Some(*location)).to_display(self.unifier).to_string()])
})
}
fn infer_pattern(&mut self, pattern: &ast::Expr<()>) -> Result<(), HashSet<String>> {
fn infer_pattern(&mut self, pattern: &ast::Expr<()>) -> Result<(), InferenceError> {
match &pattern.node {
ExprKind::Name { id, .. } => {
if !self.defined_identifiers.contains(id) {
@ -716,7 +718,7 @@ impl<'a> Inferencer<'a> {
location: Location,
args: Arguments,
body: ast::Expr<()>,
) -> Result<ast::Expr<Option<Type>>, HashSet<String>> {
) -> Result<ast::Expr<Option<Type>>, InferenceError> {
if !args.posonlyargs.is_empty()
|| args.vararg.is_some()
|| !args.kwonlyargs.is_empty()
@ -787,7 +789,7 @@ impl<'a> Inferencer<'a> {
location: Location,
elt: ast::Expr<()>,
mut generators: Vec<Comprehension>,
) -> Result<ast::Expr<Option<Type>>, HashSet<String>> {
) -> Result<ast::Expr<Option<Type>>, InferenceError> {
if generators.len() != 1 {
return report_error(
"Only 1 generator statement for list comprehension is supported",
@ -893,7 +895,7 @@ impl<'a> Inferencer<'a> {
id: StrRef,
arg_index: usize,
shape_expr: Located<ExprKind>,
) -> Result<(u64, ast::Expr<Option<Type>>), HashSet<String>> {
) -> Result<(u64, ast::Expr<Option<Type>>), InferenceError> {
/*
### Further explanation
@ -1030,7 +1032,7 @@ impl<'a> Inferencer<'a> {
func: &ast::Expr<()>,
args: &mut Vec<ast::Expr<()>>,
keywords: &[Located<ast::KeywordData>],
) -> Result<Option<ast::Expr<Option<Type>>>, HashSet<String>> {
) -> Result<Option<ast::Expr<Option<Type>>>, InferenceError> {
let Located { location: func_location, node: ExprKind::Name { id, ctx }, .. } = func else {
return Ok(None);
};
@ -1512,7 +1514,7 @@ impl<'a> Inferencer<'a> {
func: ast::Expr<()>,
mut args: Vec<ast::Expr<()>>,
keywords: Vec<Located<ast::KeywordData>>,
) -> Result<ast::Expr<Option<Type>>, HashSet<String>> {
) -> Result<ast::Expr<Option<Type>>, InferenceError> {
if let Some(spec_call_func) =
self.try_fold_special_call(location, &func, &mut args, &keywords)?
{