Add lazy closure evaluation to default (#14160) (#15654)

# Description

This PR adds lazy closure evaluation to the `default` command (closes
#14160).

- For non-closure values and without providing a column name, `default`
acts the same as before
- The user can now provide multiple column names to populate if empty
- If the user provides a column name, the input must be a record or
list, otherwise an error is created.
- The user can now provide a closure as a default value
  - This closure is run without any arguments or input
  - The closure is never evaluated if the value isn't needed
- Even when column names are supplied, the closure is only run once (and
cached to prevent re-calling it)

For example:

```nushell
> default { 1 + 2 } # => 3
> null | default 3 a   # => previously `null`, now errors
> 1 | default { sleep 5sec; 3 } # => `1`, without waiting 5 seconds

> let optional_var = null; $optional_var | default { input 'Enter value: ' } # => Returns user input
> 5 | default { input 'Enter value: ' } # => `5`, without prompting user

> ls | default { sleep 5sec; 'N/A' } name # => No-op since `name` column is never empty
> ls | default { sleep 5sec; 'N/A' } foo bar # => creates columns `foo` and `bar`; only takes 5 seconds since closure result is cached

# Old behavior is the same
> [] | default 'foo' # => []
> [] | default --empty 'foo' # => 'foo'
> default 5 # => 5
```

# User-Facing Changes

- Users can add default values to multiple columns now.
- Users can now use closures as the default value passed to `default`.
- To return a closure, the user must wrap the closure they want to
return inside another closure, which will be run (`default { $my_closure
}`).

# Tests + Formatting

All tests pass.

# After Submitting

---------

Co-authored-by: 132ikl <132@ikl.sh>
This commit is contained in:
Firegem 2025-05-15 10:10:56 -04:00 committed by GitHub
parent 505cc014ac
commit bb37306d07
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 241 additions and 33 deletions

View File

@ -1,4 +1,4 @@
use nu_engine::command_prelude::*;
use nu_engine::{ClosureEval, command_prelude::*};
use nu_protocol::{ListStream, Signals};
#[derive(Clone)]
@ -19,7 +19,7 @@ impl Command for Default {
SyntaxShape::Any,
"The value to use as a default.",
)
.optional(
.rest(
"column name",
SyntaxShape::String,
"The name of the column.",
@ -43,8 +43,18 @@ impl Command for Default {
call: &Call,
input: PipelineData,
) -> Result<PipelineData, ShellError> {
let default_value: Value = call.req(engine_state, stack, 0)?;
let columns: Vec<String> = call.rest(engine_state, stack, 1)?;
let empty = call.has_flag(engine_state, stack, "empty")?;
default(engine_state, stack, call, input, empty)
default(
engine_state,
stack,
call,
input,
default_value,
empty,
columns,
)
}
fn examples(&self) -> Vec<Example> {
@ -100,6 +110,25 @@ impl Command for Default {
}),
])),
},
Example {
description: r#"Generate a default value from a closure"#,
example: "null | default { 1 + 2 }",
result: Some(Value::test_int(3)),
},
Example {
description: r#"Fill missing column values based on other columns"#,
example: r#"[{a:1 b:2} {b:1}] | upsert a {|rc| default { $rc.b + 1 } }"#,
result: Some(Value::test_list(vec![
Value::test_record(record! {
"a" => Value::test_int(1),
"b" => Value::test_int(2),
}),
Value::test_record(record! {
"a" => Value::test_int(2),
"b" => Value::test_int(1),
}),
])),
},
]
}
}
@ -109,43 +138,68 @@ fn default(
stack: &mut Stack,
call: &Call,
input: PipelineData,
default_value: Value,
default_when_empty: bool,
columns: Vec<String>,
) -> Result<PipelineData, ShellError> {
let input_span = input.span().unwrap_or(call.head);
let mut default_value = DefaultValue::new(engine_state, stack, default_value);
let metadata = input.metadata();
let value: Value = call.req(engine_state, stack, 0)?;
let column: Option<Spanned<String>> = call.opt(engine_state, stack, 1)?;
if let Some(column) = column {
input
.map(
move |mut item| match item {
Value::Record {
val: ref mut record,
..
} => {
let record = record.to_mut();
if let Some(val) = record.get_mut(&column.item) {
if matches!(val, Value::Nothing { .. })
|| (default_when_empty && val.is_empty())
{
*val = value.clone();
}
} else {
record.push(column.item.clone(), value.clone());
}
// If user supplies columns, check if input is a record or list of records
// and set the default value for the specified record columns
if !columns.is_empty() {
if matches!(input, PipelineData::Value(Value::Record { .. }, _)) {
let record = input.into_value(input_span)?.into_record()?;
fill_record(
record,
input_span,
&mut default_value,
columns.as_slice(),
default_when_empty,
)
.map(|x| x.into_pipeline_data_with_metadata(metadata))
} else if matches!(
input,
PipelineData::ListStream(..) | PipelineData::Value(Value::List { .. }, _)
) {
// Potential enhancement: add another branch for Value::List,
// and collect the iterator into a Result<Value::List, ShellError>
// so we can preemptively return an error for collected lists
let head = call.head;
Ok(input
.into_iter()
.map(move |item| {
let span = item.span();
if let Value::Record { val, .. } = item {
fill_record(
val.into_owned(),
span,
&mut default_value,
columns.as_slice(),
default_when_empty,
)
.unwrap_or_else(|err| Value::error(err, head))
} else {
item
}
_ => item,
},
engine_state.signals(),
)
.map(|x| x.set_metadata(metadata))
})
.into_pipeline_data_with_metadata(head, engine_state.signals().clone(), metadata))
// If columns are given, but input does not use columns, return an error
} else {
Err(ShellError::PipelineMismatch {
exp_input_type: "record, table".to_string(),
dst_span: input_span,
src_span: input_span,
})
}
// Otherwise, if no column name is given, check if value is null
// or an empty string, list, or record when --empty is passed
} else if input.is_nothing()
|| (default_when_empty
&& matches!(input, PipelineData::Value(ref value, _) if value.is_empty()))
{
Ok(value.into_pipeline_data())
default_value.pipeline_data()
} else if default_when_empty && matches!(input, PipelineData::ListStream(..)) {
let PipelineData::ListStream(ls, metadata) = input else {
unreachable!()
@ -153,26 +207,86 @@ fn default(
let span = ls.span();
let mut stream = ls.into_inner().peekable();
if stream.peek().is_none() {
return Ok(value.into_pipeline_data());
return default_value.pipeline_data();
}
// stream's internal state already preserves the original signals config, so if this
// Signals::empty list stream gets interrupted it will be caught by the underlying iterator
let ls = ListStream::new(stream, span, Signals::empty());
Ok(PipelineData::ListStream(ls, metadata))
// Otherwise, return the input as is
} else {
Ok(input)
}
}
/// A wrapper around the default value to handle closures and caching values
enum DefaultValue {
Uncalculated(Spanned<ClosureEval>),
Calculated(Value),
}
impl DefaultValue {
fn new(engine_state: &EngineState, stack: &Stack, value: Value) -> Self {
let span = value.span();
match value {
Value::Closure { val, .. } => {
let closure_eval = ClosureEval::new(engine_state, stack, *val);
DefaultValue::Uncalculated(closure_eval.into_spanned(span))
}
_ => DefaultValue::Calculated(value),
}
}
fn value(&mut self) -> Result<Value, ShellError> {
match self {
DefaultValue::Uncalculated(closure) => {
let value = closure
.item
.run_with_input(PipelineData::Empty)?
.into_value(closure.span)?;
*self = DefaultValue::Calculated(value.clone());
Ok(value)
}
DefaultValue::Calculated(value) => Ok(value.clone()),
}
}
fn pipeline_data(&mut self) -> Result<PipelineData, ShellError> {
self.value().map(|x| x.into_pipeline_data())
}
}
/// Given a record, fill missing columns with a default value
fn fill_record(
mut record: Record,
span: Span,
default_value: &mut DefaultValue,
columns: &[String],
empty: bool,
) -> Result<Value, ShellError> {
for col in columns {
if let Some(val) = record.get_mut(col) {
if matches!(val, Value::Nothing { .. }) || (empty && val.is_empty()) {
*val = default_value.value()?;
}
} else {
record.push(col.clone(), default_value.value()?);
}
}
Ok(Value::record(record, span))
}
#[cfg(test)]
mod test {
use crate::Upsert;
use super::*;
#[test]
fn test_examples() {
use crate::test_examples;
use crate::test_examples_with_commands;
test_examples(Default {})
test_examples_with_commands(Default {}, &[&Upsert]);
}
}

View File

@ -150,3 +150,97 @@ fn do_not_replace_non_empty_list_stream() {
assert_eq!(actual.out, "2");
})
}
#[test]
fn closure_eval_simple() {
let actual = nu!(r#"null | default { 1 }"#);
assert_eq!(actual.out, "1");
}
#[test]
fn closure_eval_complex() {
let actual = nu!(r#"null | default { seq 1 5 | math sum }"#);
assert_eq!(actual.out, "15");
}
#[test]
fn closure_eval_is_lazy() {
let actual = nu!(r#"1 | default { error make -u {msg: foo} }"#);
assert_eq!(actual.out, "1");
}
#[test]
fn column_closure_eval_is_lazy() {
let actual = nu!(r#"{a: 1} | default { error make -u {msg: foo} } a | get a"#);
assert_eq!(actual.out, "1");
}
#[test]
fn closure_eval_replace_empty_string() {
let actual = nu!(r#"'' | default --empty { 1 }"#);
assert_eq!(actual.out, "1");
}
#[test]
fn closure_eval_do_not_replace_empty_string() {
let actual = nu!(r#"'' | default { 1 }"#);
assert_eq!(actual.out, "");
}
#[test]
fn closure_eval_replace_empty_list() {
let actual = nu!(r#"[] | default --empty { 1 }"#);
assert_eq!(actual.out, "1");
}
#[test]
fn closure_eval_do_not_replace_empty_list() {
let actual = nu!(r#"[] | default { 1 } | length"#);
assert_eq!(actual.out, "0");
}
#[test]
fn closure_eval_replace_empty_record() {
let actual = nu!(r#"{} | default --empty { 1 }"#);
assert_eq!(actual.out, "1");
}
#[test]
fn closure_eval_do_not_replace_empty_record() {
let actual = nu!(r#"{} | default { 1 } | columns | length"#);
assert_eq!(actual.out, "0");
}
#[test]
fn closure_eval_add_missing_column_record() {
let actual = nu!(r#"
{a: 1} | default { 2 } b | get b
"#);
assert_eq!(actual.out, "2");
}
#[test]
fn closure_eval_add_missing_column_table() {
let actual = nu!(r#"
[{a: 1, b: 2}, {b: 4}] | default { 3 } a | get a | to json -r
"#);
assert_eq!(actual.out, "[1,3]");
}
#[test]
fn closure_eval_replace_empty_column() {
let actual = nu!(r#"{a: ''} | default -e { 1 } a | get a"#);
assert_eq!(actual.out, "1");
}
#[test]
fn replace_multiple_columns() {
let actual = nu!(r#"{a: ''} | default -e 1 a b | values | to json -r"#);
assert_eq!(actual.out, "[1,1]");
}
#[test]
fn return_closure_value() {
let actual = nu!(r#"null | default { {||} }"#);
assert!(actual.out.starts_with("closure"));
}