Fix parse-time pipeline type checking to support multiple output types for same input type (#16111)

# Description
Fixes #15485

This PR changes pipeline checking to keep track of all possible output
types instead of only first type matching input type which appears in
the input/output types. For example, in this command:
```nushell
def foo []: [int -> string, int -> record] {
  # ...
}
```
An `int` input to the command may result in a string or a record to be
output. Before this PR, Nushell would always assume that an `int` input
would cause a `string` output because it's the first matching
input/output type pair. This would cause issues during type checking
where the parser would incorrectly determine the output type. After this
PR, Nushell considers the command to output either a string or a record.

# User-Facing Changes
* Parse-time pipeline type checking now properly supports commands with
multiple pipeline output types for the same pipeline input type

# Tests + Formatting
Added a couple tests

# After Submitting
N/A

---------

Co-authored-by: Bahex <Bahex@users.noreply.github.com>
This commit is contained in:
132ikl
2025-08-01 21:35:25 -04:00
committed by GitHub
parent eb8d2d3206
commit da9615f971
7 changed files with 164 additions and 83 deletions

View File

@ -1,6 +1,7 @@
use nu_protocol::{
ParseError, Span, Type,
ast::{Assignment, Block, Comparison, Expr, Expression, Math, Operator, Pipeline, Range},
combined_type_string,
engine::StateWorkingSet,
};
@ -757,65 +758,71 @@ pub fn math_result_type(
}
}
/// Determine the possible output types of a pipeline.
///
/// Output is union of types in the `Vec`.
pub fn check_pipeline_type(
working_set: &StateWorkingSet,
pipeline: &Pipeline,
input_type: Type,
) -> (Type, Option<Vec<ParseError>>) {
let mut current_type = input_type;
) -> (Vec<Type>, Option<Vec<ParseError>>) {
let mut current_types: Vec<Type>;
let mut new_types: Vec<Type> = vec![input_type];
let mut output_errors: Option<Vec<ParseError>> = None;
'elem: for elem in &pipeline.elements {
for elem in &pipeline.elements {
current_types = std::mem::take(&mut new_types);
if elem.redirection.is_some() {
current_type = Type::Any;
} else if let Expr::Call(call) = &elem.expr.expr {
new_types = vec![Type::Any];
continue;
}
if let Expr::Call(call) = &elem.expr.expr {
let decl = working_set.get_decl(call.decl_id);
if current_type == Type::Any {
let mut new_current_type = None;
for (_, call_output) in decl.signature().input_output_types {
if let Some(inner_current_type) = &new_current_type {
if inner_current_type == &Type::Any {
break;
} else if inner_current_type != &call_output {
// Union unequal types to Any for now
new_current_type = Some(Type::Any)
}
} else {
new_current_type = Some(call_output.clone())
}
}
if let Some(new_current_type) = new_current_type {
current_type = new_current_type
} else {
current_type = Type::Any;
}
continue 'elem;
let io_types = decl.signature().input_output_types;
if new_types.contains(&Type::Any) {
// if input type is any, then output type could be any of the valid output types
new_types = io_types.into_iter().map(|(_, out_type)| out_type).collect();
} else {
for (call_input, call_output) in decl.signature().input_output_types {
if type_compatible(&call_input, &current_type) {
current_type = call_output.clone();
continue 'elem;
}
}
// any current type which matches an input type is a possible output type
new_types = io_types
.into_iter()
.filter(|(in_type, _)| {
current_types.iter().any(|ty| type_compatible(in_type, ty))
})
.map(|(_, out_type)| out_type)
.collect();
}
if !decl.signature().input_output_types.is_empty() {
if let Some(output_errors) = &mut output_errors {
output_errors.push(ParseError::InputMismatch(current_type, call.head))
} else {
output_errors = Some(vec![ParseError::InputMismatch(current_type, call.head)]);
}
if !new_types.is_empty() {
continue;
}
current_type = Type::Any;
if decl.signature().input_output_types.is_empty() {
new_types = vec![Type::Any];
continue;
}
let Some(types_string) = combined_type_string(&current_types, "or") else {
output_errors
.get_or_insert_default()
.push(ParseError::InternalError(
"Pipeline has no type at this point".to_string(),
elem.expr.span,
));
continue;
};
output_errors
.get_or_insert_default()
.push(ParseError::InputMismatch(types_string, call.head));
} else {
current_type = elem.expr.ty.clone();
new_types = vec![elem.expr.ty.clone()];
}
}
(current_type, output_errors)
(new_types, output_errors)
}
pub fn check_block_input_output(working_set: &StateWorkingSet, block: &Block) -> Vec<ParseError> {
@ -824,21 +831,29 @@ pub fn check_block_input_output(working_set: &StateWorkingSet, block: &Block) ->
for (input_type, output_type) in &block.signature.input_output_types {
let mut current_type = input_type.clone();
let mut current_output_type = Type::Nothing;
let mut current_output_types = vec![];
for pipeline in &block.pipelines {
let (checked_output_type, err) =
let (checked_output_types, err) =
check_pipeline_type(working_set, pipeline, current_type);
current_output_type = checked_output_type;
current_output_types = checked_output_types;
current_type = Type::Nothing;
if let Some(err) = err {
output_errors.extend_from_slice(&err);
}
}
if !type_compatible(output_type, &current_output_type)
&& output_type != &Type::Any
&& current_output_type != Type::Any
if block.pipelines.is_empty() {
current_output_types = vec![Type::Nothing];
}
if output_type == &Type::Any || current_output_types.contains(&Type::Any) {
continue;
}
if !current_output_types
.iter()
.any(|ty| type_compatible(output_type, ty))
{
let span = if block.pipelines.is_empty() {
if let Some(span) = block.span {
@ -858,9 +873,17 @@ pub fn check_block_input_output(working_set: &StateWorkingSet, block: &Block) ->
.span
};
let Some(current_ty_string) = combined_type_string(&current_output_types, "or") else {
output_errors.push(ParseError::InternalError(
"Block has no type at this point".to_string(),
span,
));
continue;
};
output_errors.push(ParseError::OutputMismatch(
output_type.clone(),
current_output_type.clone(),
current_ty_string,
span,
))
}