Copy-on-write for record values (#12305)

# Description
This adds a `SharedCow` type as a transparent copy-on-write pointer that
clones to unique on mutate.

As an initial test, the `Record` within `Value::Record` is shared.

There are some pretty big wins for performance. I'll post benchmark
results in a comment. The biggest winner is nested access, as that would
have cloned the records for each cell path follow before and it doesn't
have to anymore.

The reusability of the `SharedCow` type is nice and I think it could be
used to clean up the previous work I did with `Arc` in `EngineState`.
It's meant to be a mostly transparent clone-on-write that just clones on
`.to_mut()` or `.into_owned()` if there are actually multiple
references, but avoids cloning if the reference is unique.

# User-Facing Changes
- `Value::Record` field is a different type (plugin authors)

# Tests + Formatting
- 🟢 `toolkit fmt`
- 🟢 `toolkit clippy`
- 🟢 `toolkit test`
- 🟢 `toolkit test stdlib`

# After Submitting
- [ ] use for `EngineState`
- [ ] use for `Value::List`
This commit is contained in:
Devyn Cairns
2024-04-13 18:42:03 -07:00
committed by GitHub
parent b508d1028c
commit 2ae9ad8676
52 changed files with 328 additions and 222 deletions

View File

@ -177,9 +177,9 @@ fn run_histogram(
match v {
// parse record, and fill valid value to actual input.
Value::Record { val, .. } => {
for (c, v) in *val {
if &c == col_name {
if let Ok(v) = HashableValue::from_value(v, head_span) {
for (c, v) in val.iter() {
if c == col_name {
if let Ok(v) = HashableValue::from_value(v.clone(), head_span) {
inputs.push(v);
}
}

View File

@ -131,7 +131,7 @@ fn into_record(
.collect(),
span,
),
Value::Record { val, .. } => Value::record(*val, span),
Value::Record { .. } => input,
Value::Error { .. } => input,
other => Value::error(
ShellError::TypeMismatch {

View File

@ -108,7 +108,8 @@ impl Iterator for UpdateCellIterator {
let span = val.span();
match val {
Value::Record { val, .. } => Some(Value::record(
val.into_iter()
val.into_owned()
.into_iter()
.map(|(col, val)| match &self.columns {
Some(cols) if !cols.contains(&col) => (col, val),
_ => (

View File

@ -316,7 +316,7 @@ fn insert_value(
match stream_value {
// map each column value into its SQL representation
Value::Record { val, .. } => {
let sql_vals = values_to_sql(val.into_values())?;
let sql_vals = values_to_sql(val.values().cloned())?;
insert_statement
.execute(rusqlite::params_from_iter(sql_vals))

View File

@ -500,7 +500,7 @@ pub fn nu_value_to_params(value: Value) -> Result<NuSqlParams, ShellError> {
Value::Record { val, .. } => {
let mut params = Vec::with_capacity(val.len());
for (mut column, value) in val.into_iter() {
for (mut column, value) in val.into_owned().into_iter() {
let sql_type_erased = value_to_sql(value)?;
if !column.starts_with([':', '@', '$']) {

View File

@ -199,7 +199,7 @@ mod util {
let span = value.span();
match value {
Value::Record { val: record, .. } => {
let (cols, vals): (Vec<_>, Vec<_>) = record.into_iter().unzip();
let (cols, vals): (Vec<_>, Vec<_>) = record.into_owned().into_iter().unzip();
(
cols,
vec![vals

View File

@ -53,7 +53,7 @@ impl Command for LoadEnv {
}
None => match input {
PipelineData::Value(Value::Record { val, .. }, ..) => {
for (env_var, rhs) in *val {
for (env_var, rhs) in val.into_owned() {
let env_var_ = env_var.as_str();
if ["FILE_PWD", "CURRENT_FILE"].contains(&env_var_) {
return Err(ShellError::AutomaticEnvVarSetManually {

View File

@ -118,6 +118,7 @@ fn getcol(
})
}
Value::Record { val, .. } => Ok(val
.into_owned()
.into_iter()
.map(move |(x, _)| Value::string(x, head))
.into_pipeline_data(ctrlc)

View File

@ -85,31 +85,29 @@ fn default(
if let Some(column) = column {
input
.map(
move |item| {
let span = item.span();
match item {
Value::Record {
val: mut record, ..
} => {
let mut found = false;
move |mut item| match item {
Value::Record {
val: ref mut record,
..
} => {
let mut found = false;
for (col, val) in record.iter_mut() {
if *col == column.item {
found = true;
if matches!(val, Value::Nothing { .. }) {
*val = value.clone();
}
for (col, val) in record.to_mut().iter_mut() {
if *col == column.item {
found = true;
if matches!(val, Value::Nothing { .. }) {
*val = value.clone();
}
}
if !found {
record.push(column.item.clone(), value.clone());
}
Value::record(*record, span)
}
_ => item,
if !found {
record.to_mut().push(column.item.clone(), value.clone());
}
item
}
_ => item,
},
ctrlc,
)

View File

@ -106,7 +106,7 @@ fn drop_cols(
Ok(PipelineData::Empty)
}
}
PipelineData::Value(v, ..) => {
PipelineData::Value(mut v, ..) => {
let span = v.span();
match v {
Value::List { mut vals, .. } => {
@ -119,11 +119,12 @@ fn drop_cols(
Ok(Value::list(vals, span).into_pipeline_data_with_metadata(metadata))
}
Value::Record {
val: mut record, ..
val: ref mut record,
..
} => {
let len = record.len().saturating_sub(columns);
record.truncate(len);
Ok(Value::record(*record, span).into_pipeline_data_with_metadata(metadata))
record.to_mut().truncate(len);
Ok(v.into_pipeline_data_with_metadata(metadata))
}
// Propagate errors
Value::Error { error, .. } => Err(*error),
@ -143,7 +144,7 @@ fn drop_cols(
fn drop_cols_set(val: &mut Value, head: Span, drop: usize) -> Result<HashSet<String>, ShellError> {
if let Value::Record { val: record, .. } = val {
let len = record.len().saturating_sub(drop);
Ok(record.drain(len..).map(|(col, _)| col).collect())
Ok(record.to_mut().drain(len..).map(|(col, _)| col).collect())
} else {
Err(unsupported_value_error(val, head))
}
@ -155,7 +156,7 @@ fn drop_record_cols(
drop_cols: &HashSet<String>,
) -> Result<(), ShellError> {
if let Value::Record { val, .. } = val {
val.retain(|col, _| !drop_cols.contains(col));
val.to_mut().retain(|col, _| !drop_cols.contains(col));
Ok(())
} else {
Err(unsupported_value_error(val, head))

View File

@ -156,15 +156,15 @@ fn flat_value(columns: &[CellPath], item: Value, all: bool) -> Vec<Value> {
let mut out = IndexMap::<String, Value>::new();
let mut inner_table = None;
for (column_index, (column, value)) in val.into_iter().enumerate() {
for (column_index, (column, value)) in val.into_owned().into_iter().enumerate() {
let column_requested = columns.iter().find(|c| c.to_string() == column);
let need_flatten = { columns.is_empty() || column_requested.is_some() };
let span = value.span();
match value {
Value::Record { val, .. } => {
Value::Record { ref val, .. } => {
if need_flatten {
for (col, val) in *val {
for (col, val) in val.clone().into_owned() {
if out.contains_key(&col) {
out.insert(format!("{column}_{col}"), val);
} else {
@ -172,9 +172,9 @@ fn flat_value(columns: &[CellPath], item: Value, all: bool) -> Vec<Value> {
}
}
} else if out.contains_key(&column) {
out.insert(format!("{column}_{column}"), Value::record(*val, span));
out.insert(format!("{column}_{column}"), value);
} else {
out.insert(column, Value::record(*val, span));
out.insert(column, value);
}
}
Value::List { vals, .. } => {

View File

@ -149,6 +149,7 @@ fn replace_headers(
if let Value::Record { val: record, .. } = value {
Ok(Value::record(
record
.into_owned()
.into_iter()
.filter_map(|(col, val)| {
old_headers

View File

@ -86,6 +86,7 @@ impl Command for Items {
PipelineData::Empty => Ok(PipelineData::Empty),
PipelineData::Value(v, ..) => match v {
Value::Record { val, .. } => Ok(val
.into_owned()
.into_iter()
.map_while(run_for_each_item)
.into_pipeline_data(ctrlc)),
@ -99,6 +100,7 @@ impl Command for Items {
})?,
};
Ok(record
.into_owned()
.into_iter()
.map_while(run_for_each_item)
.into_pipeline_data(ctrlc))

View File

@ -162,7 +162,7 @@ fn rename(
block_info.clone()
{
record
.into_iter()
.into_owned().into_iter()
.map(|(col, val)| {
stack.with_env(&env_vars, &env_hidden);
@ -191,7 +191,7 @@ fn rename(
// record columns are unique so we can track the number
// of renamed columns to check if any were missed
let mut renamed = 0;
let record = record.into_iter().map(|(col, val)| {
let record = record.into_owned().into_iter().map(|(col, val)| {
let col = if let Some(col) = columns.get(&col) {
renamed += 1;
col.clone()
@ -222,7 +222,7 @@ fn rename(
}
}
None => Ok(record
.into_iter()
.into_owned().into_iter()
.enumerate()
.map(|(i, (col, val))| {
(columns.get(i).cloned().unwrap_or(col), val)
@ -237,7 +237,7 @@ fn rename(
}
}
// Propagate errors by explicitly matching them before the final case.
Value::Error { .. } => item.clone(),
Value::Error { .. } => item,
other => Value::error(
ShellError::OnlySupportsThisInputType {
exp_input_type: "record".into(),

View File

@ -144,7 +144,14 @@ impl Command for Sort {
// Records have two sorting methods, toggled by presence or absence of -v
PipelineData::Value(Value::Record { val, .. }, ..) => {
let sort_by_value = call.has_flag(engine_state, stack, "values")?;
let record = sort_record(*val, span, sort_by_value, reverse, insensitive, natural);
let record = sort_record(
val.into_owned(),
span,
sort_by_value,
reverse,
insensitive,
natural,
);
Ok(record.into_pipeline_data())
}
// Other values are sorted here

View File

@ -187,11 +187,11 @@ pub fn data_split(
let span = v.span();
match v {
Value::Record { val: grouped, .. } => {
for (outer_key, list) in grouped.into_iter() {
for (outer_key, list) in grouped.into_owned() {
match data_group(&list, splitter, span) {
Ok(grouped_vals) => {
if let Value::Record { val: sub, .. } = grouped_vals {
for (inner_key, subset) in sub.into_iter() {
for (inner_key, subset) in sub.into_owned() {
let s: &mut IndexMap<String, Value> =
splits.entry(inner_key).or_default();

View File

@ -195,6 +195,7 @@ fn sort_attributes(val: Value) -> Value {
Value::Record { val, .. } => {
// TODO: sort inplace
let sorted = val
.into_owned()
.into_iter()
.sorted_by(|a, b| a.0.cmp(&b.0))
.collect_vec();

View File

@ -157,7 +157,9 @@ fn values(
}
}
Value::Record { val, .. } => Ok(val
.into_values()
.values()
.cloned()
.collect::<Vec<_>>()
.into_pipeline_data_with_metadata(metadata, ctrlc)),
Value::LazyRecord { val, .. } => {
let record = match val.collect()? {
@ -169,6 +171,7 @@ fn values(
})?,
};
Ok(record
.into_owned()
.into_values()
.into_pipeline_data_with_metadata(metadata, ctrlc))
}

View File

@ -125,6 +125,7 @@ fn local_into_string(value: Value, separator: &str, config: &Config) -> String {
.collect::<Vec<_>>()
.join(separator),
Value::Record { val, .. } => val
.into_owned()
.into_iter()
.map(|(x, y)| format!("{}: {}", x, local_into_string(y, ", ", config)))
.collect::<Vec<_>>()

View File

@ -325,8 +325,8 @@ impl Job {
// alternatives like {tag: a attributes: {} content: []}, {tag: a attribbutes: null
// content: null}, {tag: a}. See to_xml_entry for more
let attrs = match attrs {
Value::Record { val, .. } => val,
Value::Nothing { .. } => Box::new(Record::new()),
Value::Record { val, .. } => val.into_owned(),
Value::Nothing { .. } => Record::new(),
_ => {
return Err(ShellError::CantConvert {
to_type: "XML".into(),
@ -350,7 +350,7 @@ impl Job {
}
};
self.write_tag(entry_span, tag, tag_span, *attrs, content)
self.write_tag(entry_span, tag, tag_span, attrs, content)
}
}

View File

@ -136,7 +136,7 @@ used as the next argument to the closure, otherwise generation stops.
match value {
// {out: ..., next: ...} -> output and continue
Value::Record { val, .. } => {
let iter = val.into_iter();
let iter = val.into_owned().into_iter();
let mut out = None;
let mut next = None;
let mut err = None;

View File

@ -139,19 +139,20 @@ pub fn highlight_search_in_table(
let search_string = search_string.to_folded_case();
let mut matches = vec![];
for record in table {
let span = record.span();
let (mut record, record_span) = if let Value::Record { val, .. } = record {
(val, span)
} else {
for mut value in table {
let Value::Record {
val: ref mut record,
..
} = value
else {
return Err(ShellError::NushellFailedSpanned {
msg: "Expected record".to_string(),
label: format!("got {}", record.get_type()),
span: record.span(),
label: format!("got {}", value.get_type()),
span: value.span(),
});
};
let has_match = record.iter_mut().try_fold(
let has_match = record.to_mut().iter_mut().try_fold(
false,
|acc: bool, (col, val)| -> Result<bool, ShellError> {
if !searched_cols.contains(&col.as_str()) {
@ -180,7 +181,7 @@ pub fn highlight_search_in_table(
)?;
if has_match {
matches.push(Value::record(*record, record_span));
matches.push(value);
}
}

View File

@ -80,15 +80,15 @@ pub fn calculate(
),
_ => mf(vals, span, name),
},
PipelineData::Value(Value::Record { val: record, .. }, ..) => {
let mut record = record;
PipelineData::Value(Value::Record { val, .. }, ..) => {
let mut record = val.into_owned();
record
.iter_mut()
.try_for_each(|(_, val)| -> Result<(), ShellError> {
*val = mf(slice::from_ref(val), span, name)?;
Ok(())
})?;
Ok(Value::record(*record, span))
Ok(Value::record(record, span))
}
PipelineData::Value(Value::Range { val, .. }, ..) => {
let new_vals: Result<Vec<Value>, ShellError> = val

View File

@ -222,7 +222,7 @@ pub fn send_request(
Value::Record { val, .. } if body_type == BodyType::Form => {
let mut data: Vec<(String, String)> = Vec::with_capacity(val.len());
for (col, val) in *val {
for (col, val) in val.into_owned() {
data.push((col, val.coerce_into_string()?))
}

View File

@ -92,6 +92,7 @@ impl Command for SubCommand {
match value {
Value::Record { val, .. } => {
let url_components = val
.into_owned()
.into_iter()
.try_fold(UrlComponents::new(), |url, (k, v)| {
url.add_component(k, v, span, engine_state)
@ -179,6 +180,7 @@ impl UrlComponents {
return match value {
Value::Record { val, .. } => {
let mut qs = val
.into_owned()
.into_iter()
.map(|(k, v)| match v.coerce_into_string() {
Ok(val) => Ok(format!("{k}={val}")),

View File

@ -108,7 +108,7 @@ prints out the list properly."#
// dbg!("value::record");
let mut items = vec![];
for (i, (c, v)) in val.into_iter().enumerate() {
for (i, (c, v)) in val.into_owned().into_iter().enumerate() {
items.push((i, c, v.to_expanded_string(", ", config)))
}

View File

@ -392,7 +392,7 @@ fn handle_table_command(
}
PipelineData::Value(Value::Record { val, .. }, ..) => {
input.data = PipelineData::Empty;
handle_record(input, cfg, *val)
handle_record(input, cfg, val.into_owned())
}
PipelineData::Value(Value::LazyRecord { val, .. }, ..) => {
input.data = val.collect()?.into_pipeline_data();
@ -557,7 +557,7 @@ fn handle_row_stream(
stream.map(move |mut x| match &mut x {
Value::Record { val: record, .. } => {
// Only the name column gets special colors, for now
if let Some(value) = record.get_mut("name") {
if let Some(value) = record.to_mut().get_mut("name") {
let span = value.span();
if let Value::String { val, .. } = value {
if let Some(val) = render_path_name(val, &config, &ls_colors, span)
@ -583,7 +583,7 @@ fn handle_row_stream(
ListStream::from_stream(
stream.map(move |mut x| match &mut x {
Value::Record { val: record, .. } => {
for (rec_col, rec_val) in record.iter_mut() {
for (rec_col, rec_val) in record.to_mut().iter_mut() {
// Every column in the HTML theme table except 'name' is colored
if rec_col != "name" {
continue;