
At present, when calculating some ranges, we take the `tuple_windows` to iterate forward, which will cause some problems when the cursor is being front, because `tuple_windows` iteration cannot iterate to the very beginning, so there will be some cases that cannot be calculated, adjust this method, and now it can calculate more perfectly, and the execution speed is about the same Release Notes: - N/A
1564 lines
50 KiB
Rust
1564 lines
50 KiB
Rust
use std::ops::Range;
|
|
|
|
use crate::{
|
|
motion::{coerce_punctuation, right},
|
|
normal::normal_object,
|
|
state::Mode,
|
|
visual::visual_object,
|
|
Vim,
|
|
};
|
|
use editor::{
|
|
display_map::{DisplaySnapshot, ToDisplayPoint},
|
|
movement::{self, FindRange},
|
|
Bias, DisplayPoint,
|
|
};
|
|
|
|
use itertools::Itertools;
|
|
|
|
use gpui::{actions, impl_actions, ViewContext, WindowContext};
|
|
use language::{char_kind, BufferSnapshot, CharKind, Point, Selection};
|
|
use multi_buffer::MultiBufferRow;
|
|
use serde::Deserialize;
|
|
use workspace::Workspace;
|
|
|
|
#[derive(Copy, Clone, Debug, PartialEq, Eq, Deserialize)]
|
|
pub enum Object {
|
|
Word { ignore_punctuation: bool },
|
|
Sentence,
|
|
Paragraph,
|
|
Quotes,
|
|
BackQuotes,
|
|
DoubleQuotes,
|
|
VerticalBars,
|
|
Parentheses,
|
|
SquareBrackets,
|
|
CurlyBrackets,
|
|
AngleBrackets,
|
|
Argument,
|
|
Tag,
|
|
}
|
|
|
|
#[derive(Clone, Deserialize, PartialEq)]
|
|
#[serde(rename_all = "camelCase")]
|
|
struct Word {
|
|
#[serde(default)]
|
|
ignore_punctuation: bool,
|
|
}
|
|
|
|
impl_actions!(vim, [Word]);
|
|
|
|
actions!(
|
|
vim,
|
|
[
|
|
Sentence,
|
|
Paragraph,
|
|
Quotes,
|
|
BackQuotes,
|
|
DoubleQuotes,
|
|
VerticalBars,
|
|
Parentheses,
|
|
SquareBrackets,
|
|
CurlyBrackets,
|
|
AngleBrackets,
|
|
Argument,
|
|
Tag
|
|
]
|
|
);
|
|
|
|
pub fn register(workspace: &mut Workspace, _: &mut ViewContext<Workspace>) {
|
|
workspace.register_action(
|
|
|_: &mut Workspace, &Word { ignore_punctuation }: &Word, cx: _| {
|
|
object(Object::Word { ignore_punctuation }, cx)
|
|
},
|
|
);
|
|
workspace.register_action(|_: &mut Workspace, _: &Tag, cx: _| object(Object::Tag, cx));
|
|
workspace
|
|
.register_action(|_: &mut Workspace, _: &Sentence, cx: _| object(Object::Sentence, cx));
|
|
workspace
|
|
.register_action(|_: &mut Workspace, _: &Paragraph, cx: _| object(Object::Paragraph, cx));
|
|
workspace.register_action(|_: &mut Workspace, _: &Quotes, cx: _| object(Object::Quotes, cx));
|
|
workspace
|
|
.register_action(|_: &mut Workspace, _: &BackQuotes, cx: _| object(Object::BackQuotes, cx));
|
|
workspace.register_action(|_: &mut Workspace, _: &DoubleQuotes, cx: _| {
|
|
object(Object::DoubleQuotes, cx)
|
|
});
|
|
workspace.register_action(|_: &mut Workspace, _: &Parentheses, cx: _| {
|
|
object(Object::Parentheses, cx)
|
|
});
|
|
workspace.register_action(|_: &mut Workspace, _: &SquareBrackets, cx: _| {
|
|
object(Object::SquareBrackets, cx)
|
|
});
|
|
workspace.register_action(|_: &mut Workspace, _: &CurlyBrackets, cx: _| {
|
|
object(Object::CurlyBrackets, cx)
|
|
});
|
|
workspace.register_action(|_: &mut Workspace, _: &AngleBrackets, cx: _| {
|
|
object(Object::AngleBrackets, cx)
|
|
});
|
|
workspace.register_action(|_: &mut Workspace, _: &VerticalBars, cx: _| {
|
|
object(Object::VerticalBars, cx)
|
|
});
|
|
workspace
|
|
.register_action(|_: &mut Workspace, _: &Argument, cx: _| object(Object::Argument, cx));
|
|
}
|
|
|
|
fn object(object: Object, cx: &mut WindowContext) {
|
|
match Vim::read(cx).state().mode {
|
|
Mode::Normal => normal_object(object, cx),
|
|
Mode::Visual | Mode::VisualLine | Mode::VisualBlock => visual_object(object, cx),
|
|
Mode::Insert | Mode::Replace => {
|
|
// Shouldn't execute a text object in insert mode. Ignoring
|
|
}
|
|
}
|
|
}
|
|
|
|
impl Object {
|
|
pub fn is_multiline(self) -> bool {
|
|
match self {
|
|
Object::Word { .. }
|
|
| Object::Quotes
|
|
| Object::BackQuotes
|
|
| Object::VerticalBars
|
|
| Object::DoubleQuotes => false,
|
|
Object::Sentence
|
|
| Object::Paragraph
|
|
| Object::Parentheses
|
|
| Object::Tag
|
|
| Object::AngleBrackets
|
|
| Object::CurlyBrackets
|
|
| Object::SquareBrackets
|
|
| Object::Argument => true,
|
|
}
|
|
}
|
|
|
|
pub fn always_expands_both_ways(self) -> bool {
|
|
match self {
|
|
Object::Word { .. } | Object::Sentence | Object::Paragraph | Object::Argument => false,
|
|
Object::Quotes
|
|
| Object::BackQuotes
|
|
| Object::DoubleQuotes
|
|
| Object::VerticalBars
|
|
| Object::Parentheses
|
|
| Object::SquareBrackets
|
|
| Object::Tag
|
|
| Object::CurlyBrackets
|
|
| Object::AngleBrackets => true,
|
|
}
|
|
}
|
|
|
|
pub fn target_visual_mode(self, current_mode: Mode) -> Mode {
|
|
match self {
|
|
Object::Word { .. }
|
|
| Object::Sentence
|
|
| Object::Quotes
|
|
| Object::BackQuotes
|
|
| Object::DoubleQuotes => {
|
|
if current_mode == Mode::VisualBlock {
|
|
Mode::VisualBlock
|
|
} else {
|
|
Mode::Visual
|
|
}
|
|
}
|
|
Object::Parentheses
|
|
| Object::SquareBrackets
|
|
| Object::CurlyBrackets
|
|
| Object::AngleBrackets
|
|
| Object::VerticalBars
|
|
| Object::Tag
|
|
| Object::Argument => Mode::Visual,
|
|
Object::Paragraph => Mode::VisualLine,
|
|
}
|
|
}
|
|
|
|
pub fn range(
|
|
self,
|
|
map: &DisplaySnapshot,
|
|
selection: Selection<DisplayPoint>,
|
|
around: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
let relative_to = selection.head();
|
|
match self {
|
|
Object::Word { ignore_punctuation } => {
|
|
if around {
|
|
around_word(map, relative_to, ignore_punctuation)
|
|
} else {
|
|
in_word(map, relative_to, ignore_punctuation)
|
|
}
|
|
}
|
|
Object::Sentence => sentence(map, relative_to, around),
|
|
Object::Paragraph => paragraph(map, relative_to, around),
|
|
Object::Quotes => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '\'', '\'')
|
|
}
|
|
Object::BackQuotes => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '`', '`')
|
|
}
|
|
Object::DoubleQuotes => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '"', '"')
|
|
}
|
|
Object::VerticalBars => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '|', '|')
|
|
}
|
|
Object::Parentheses => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '(', ')')
|
|
}
|
|
Object::Tag => surrounding_html_tag(map, selection, around),
|
|
Object::SquareBrackets => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '[', ']')
|
|
}
|
|
Object::CurlyBrackets => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '{', '}')
|
|
}
|
|
Object::AngleBrackets => {
|
|
surrounding_markers(map, relative_to, around, self.is_multiline(), '<', '>')
|
|
}
|
|
Object::Argument => argument(map, relative_to, around),
|
|
}
|
|
}
|
|
|
|
pub fn expand_selection(
|
|
self,
|
|
map: &DisplaySnapshot,
|
|
selection: &mut Selection<DisplayPoint>,
|
|
around: bool,
|
|
) -> bool {
|
|
if let Some(range) = self.range(map, selection.clone(), around) {
|
|
selection.start = range.start;
|
|
selection.end = range.end;
|
|
true
|
|
} else {
|
|
false
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Returns a range that surrounds the word `relative_to` is in.
|
|
///
|
|
/// If `relative_to` is at the start of a word, return the word.
|
|
/// If `relative_to` is between words, return the space between.
|
|
fn in_word(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
ignore_punctuation: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
// Use motion::right so that we consider the character under the cursor when looking for the start
|
|
let scope = map
|
|
.buffer_snapshot
|
|
.language_scope_at(relative_to.to_point(map));
|
|
let start = movement::find_preceding_boundary_display_point(
|
|
map,
|
|
right(map, relative_to, 1),
|
|
movement::FindRange::SingleLine,
|
|
|left, right| {
|
|
coerce_punctuation(char_kind(&scope, left), ignore_punctuation)
|
|
!= coerce_punctuation(char_kind(&scope, right), ignore_punctuation)
|
|
},
|
|
);
|
|
|
|
let end = movement::find_boundary(map, relative_to, FindRange::SingleLine, |left, right| {
|
|
coerce_punctuation(char_kind(&scope, left), ignore_punctuation)
|
|
!= coerce_punctuation(char_kind(&scope, right), ignore_punctuation)
|
|
});
|
|
|
|
Some(start..end)
|
|
}
|
|
|
|
fn surrounding_html_tag(
|
|
map: &DisplaySnapshot,
|
|
selection: Selection<DisplayPoint>,
|
|
around: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
fn read_tag(chars: impl Iterator<Item = char>) -> String {
|
|
chars
|
|
.take_while(|c| c.is_alphanumeric() || *c == ':' || *c == '-' || *c == '_' || *c == '.')
|
|
.collect()
|
|
}
|
|
fn open_tag(mut chars: impl Iterator<Item = char>) -> Option<String> {
|
|
if Some('<') != chars.next() {
|
|
return None;
|
|
}
|
|
Some(read_tag(chars))
|
|
}
|
|
fn close_tag(mut chars: impl Iterator<Item = char>) -> Option<String> {
|
|
if (Some('<'), Some('/')) != (chars.next(), chars.next()) {
|
|
return None;
|
|
}
|
|
Some(read_tag(chars))
|
|
}
|
|
|
|
let snapshot = &map.buffer_snapshot;
|
|
let offset = selection.head().to_offset(map, Bias::Left);
|
|
let excerpt = snapshot.excerpt_containing(offset..offset)?;
|
|
let buffer = excerpt.buffer();
|
|
let offset = excerpt.map_offset_to_buffer(offset);
|
|
|
|
// Find the most closest to current offset
|
|
let mut cursor = buffer.syntax_layer_at(offset)?.node().walk();
|
|
let mut last_child_node = cursor.node();
|
|
while cursor.goto_first_child_for_byte(offset).is_some() {
|
|
last_child_node = cursor.node();
|
|
}
|
|
|
|
let mut last_child_node = Some(last_child_node);
|
|
while let Some(cur_node) = last_child_node {
|
|
if cur_node.child_count() >= 2 {
|
|
let first_child = cur_node.child(0);
|
|
let last_child = cur_node.child(cur_node.child_count() - 1);
|
|
if let (Some(first_child), Some(last_child)) = (first_child, last_child) {
|
|
let open_tag = open_tag(buffer.chars_for_range(first_child.byte_range()));
|
|
let close_tag = close_tag(buffer.chars_for_range(last_child.byte_range()));
|
|
// It needs to be handled differently according to the selection length
|
|
let is_valid = if selection.end.to_offset(map, Bias::Left)
|
|
- selection.start.to_offset(map, Bias::Left)
|
|
<= 1
|
|
{
|
|
offset <= last_child.end_byte()
|
|
} else {
|
|
selection.start.to_offset(map, Bias::Left) >= first_child.start_byte()
|
|
&& selection.end.to_offset(map, Bias::Left) <= last_child.start_byte() + 1
|
|
};
|
|
if open_tag.is_some() && open_tag == close_tag && is_valid {
|
|
let range = if around {
|
|
first_child.byte_range().start..last_child.byte_range().end
|
|
} else {
|
|
first_child.byte_range().end..last_child.byte_range().start
|
|
};
|
|
if excerpt.contains_buffer_range(range.clone()) {
|
|
let result = excerpt.map_range_from_buffer(range);
|
|
return Some(
|
|
result.start.to_display_point(map)..result.end.to_display_point(map),
|
|
);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
last_child_node = cur_node.parent();
|
|
}
|
|
None
|
|
}
|
|
|
|
/// Returns a range that surrounds the word and following whitespace
|
|
/// relative_to is in.
|
|
///
|
|
/// If `relative_to` is at the start of a word, return the word and following whitespace.
|
|
/// If `relative_to` is between words, return the whitespace back and the following word.
|
|
///
|
|
/// if in word
|
|
/// delete that word
|
|
/// if there is whitespace following the word, delete that as well
|
|
/// otherwise, delete any preceding whitespace
|
|
/// otherwise
|
|
/// delete whitespace around cursor
|
|
/// delete word following the cursor
|
|
fn around_word(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
ignore_punctuation: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
let offset = relative_to.to_offset(map, Bias::Left);
|
|
let scope = map.buffer_snapshot.language_scope_at(offset);
|
|
let in_word = map
|
|
.buffer_chars_at(offset)
|
|
.next()
|
|
.map(|(c, _)| char_kind(&scope, c) != CharKind::Whitespace)
|
|
.unwrap_or(false);
|
|
|
|
if in_word {
|
|
around_containing_word(map, relative_to, ignore_punctuation)
|
|
} else {
|
|
around_next_word(map, relative_to, ignore_punctuation)
|
|
}
|
|
}
|
|
|
|
fn around_containing_word(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
ignore_punctuation: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
in_word(map, relative_to, ignore_punctuation)
|
|
.map(|range| expand_to_include_whitespace(map, range, true))
|
|
}
|
|
|
|
fn around_next_word(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
ignore_punctuation: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
let scope = map
|
|
.buffer_snapshot
|
|
.language_scope_at(relative_to.to_point(map));
|
|
// Get the start of the word
|
|
let start = movement::find_preceding_boundary_display_point(
|
|
map,
|
|
right(map, relative_to, 1),
|
|
FindRange::SingleLine,
|
|
|left, right| {
|
|
coerce_punctuation(char_kind(&scope, left), ignore_punctuation)
|
|
!= coerce_punctuation(char_kind(&scope, right), ignore_punctuation)
|
|
},
|
|
);
|
|
|
|
let mut word_found = false;
|
|
let end = movement::find_boundary(map, relative_to, FindRange::MultiLine, |left, right| {
|
|
let left_kind = coerce_punctuation(char_kind(&scope, left), ignore_punctuation);
|
|
let right_kind = coerce_punctuation(char_kind(&scope, right), ignore_punctuation);
|
|
|
|
let found = (word_found && left_kind != right_kind) || right == '\n' && left == '\n';
|
|
|
|
if right_kind != CharKind::Whitespace {
|
|
word_found = true;
|
|
}
|
|
|
|
found
|
|
});
|
|
|
|
Some(start..end)
|
|
}
|
|
|
|
fn argument(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
around: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
let snapshot = &map.buffer_snapshot;
|
|
let offset = relative_to.to_offset(map, Bias::Left);
|
|
|
|
// The `argument` vim text object uses the syntax tree, so we operate at the buffer level and map back to the display level
|
|
let excerpt = snapshot.excerpt_containing(offset..offset)?;
|
|
let buffer = excerpt.buffer();
|
|
|
|
fn comma_delimited_range_at(
|
|
buffer: &BufferSnapshot,
|
|
mut offset: usize,
|
|
include_comma: bool,
|
|
) -> Option<Range<usize>> {
|
|
// Seek to the first non-whitespace character
|
|
offset += buffer
|
|
.chars_at(offset)
|
|
.take_while(|c| c.is_whitespace())
|
|
.map(char::len_utf8)
|
|
.sum::<usize>();
|
|
|
|
let bracket_filter = |open: Range<usize>, close: Range<usize>| {
|
|
// Filter out empty ranges
|
|
if open.end == close.start {
|
|
return false;
|
|
}
|
|
|
|
// If the cursor is outside the brackets, ignore them
|
|
if open.start == offset || close.end == offset {
|
|
return false;
|
|
}
|
|
|
|
// TODO: Is there any better way to filter out string brackets?
|
|
// Used to filter out string brackets
|
|
return matches!(
|
|
buffer.chars_at(open.start).next(),
|
|
Some('(' | '[' | '{' | '<' | '|')
|
|
);
|
|
};
|
|
|
|
// Find the brackets containing the cursor
|
|
let (open_bracket, close_bracket) =
|
|
buffer.innermost_enclosing_bracket_ranges(offset..offset, Some(&bracket_filter))?;
|
|
|
|
let inner_bracket_range = open_bracket.end..close_bracket.start;
|
|
|
|
let layer = buffer.syntax_layer_at(offset)?;
|
|
let node = layer.node();
|
|
let mut cursor = node.walk();
|
|
|
|
// Loop until we find the smallest node whose parent covers the bracket range. This node is the argument in the parent argument list
|
|
let mut parent_covers_bracket_range = false;
|
|
loop {
|
|
let node = cursor.node();
|
|
let range = node.byte_range();
|
|
let covers_bracket_range =
|
|
range.start == open_bracket.start && range.end == close_bracket.end;
|
|
if parent_covers_bracket_range && !covers_bracket_range {
|
|
break;
|
|
}
|
|
parent_covers_bracket_range = covers_bracket_range;
|
|
|
|
// Unable to find a child node with a parent that covers the bracket range, so no argument to select
|
|
if cursor.goto_first_child_for_byte(offset).is_none() {
|
|
return None;
|
|
}
|
|
}
|
|
|
|
let mut argument_node = cursor.node();
|
|
|
|
// If the child node is the open bracket, move to the next sibling.
|
|
if argument_node.byte_range() == open_bracket {
|
|
if !cursor.goto_next_sibling() {
|
|
return Some(inner_bracket_range);
|
|
}
|
|
argument_node = cursor.node();
|
|
}
|
|
// While the child node is the close bracket or a comma, move to the previous sibling
|
|
while argument_node.byte_range() == close_bracket || argument_node.kind() == "," {
|
|
if !cursor.goto_previous_sibling() {
|
|
return Some(inner_bracket_range);
|
|
}
|
|
argument_node = cursor.node();
|
|
if argument_node.byte_range() == open_bracket {
|
|
return Some(inner_bracket_range);
|
|
}
|
|
}
|
|
|
|
// The start and end of the argument range, defaulting to the start and end of the argument node
|
|
let mut start = argument_node.start_byte();
|
|
let mut end = argument_node.end_byte();
|
|
|
|
let mut needs_surrounding_comma = include_comma;
|
|
|
|
// Seek backwards to find the start of the argument - either the previous comma or the opening bracket.
|
|
// We do this because multiple nodes can represent a single argument, such as with rust `vec![a.b.c, d.e.f]`
|
|
while cursor.goto_previous_sibling() {
|
|
let prev = cursor.node();
|
|
|
|
if prev.start_byte() < open_bracket.end {
|
|
start = open_bracket.end;
|
|
break;
|
|
} else if prev.kind() == "," {
|
|
if needs_surrounding_comma {
|
|
start = prev.start_byte();
|
|
needs_surrounding_comma = false;
|
|
}
|
|
break;
|
|
} else if prev.start_byte() < start {
|
|
start = prev.start_byte();
|
|
}
|
|
}
|
|
|
|
// Do the same for the end of the argument, extending to next comma or the end of the argument list
|
|
while cursor.goto_next_sibling() {
|
|
let next = cursor.node();
|
|
|
|
if next.end_byte() > close_bracket.start {
|
|
end = close_bracket.start;
|
|
break;
|
|
} else if next.kind() == "," {
|
|
if needs_surrounding_comma {
|
|
// Select up to the beginning of the next argument if there is one, otherwise to the end of the comma
|
|
if let Some(next_arg) = next.next_sibling() {
|
|
end = next_arg.start_byte();
|
|
} else {
|
|
end = next.end_byte();
|
|
}
|
|
}
|
|
break;
|
|
} else if next.end_byte() > end {
|
|
end = next.end_byte();
|
|
}
|
|
}
|
|
|
|
Some(start..end)
|
|
}
|
|
|
|
let result = comma_delimited_range_at(buffer, excerpt.map_offset_to_buffer(offset), around)?;
|
|
|
|
if excerpt.contains_buffer_range(result.clone()) {
|
|
let result = excerpt.map_range_from_buffer(result);
|
|
Some(result.start.to_display_point(map)..result.end.to_display_point(map))
|
|
} else {
|
|
None
|
|
}
|
|
}
|
|
|
|
fn sentence(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
around: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
let mut start = None;
|
|
let relative_offset = relative_to.to_offset(map, Bias::Left);
|
|
let mut previous_end = relative_offset;
|
|
|
|
let mut chars = map.buffer_chars_at(previous_end).peekable();
|
|
|
|
// Search backwards for the previous sentence end or current sentence start. Include the character under relative_to
|
|
for (char, offset) in chars
|
|
.peek()
|
|
.cloned()
|
|
.into_iter()
|
|
.chain(map.reverse_buffer_chars_at(previous_end))
|
|
{
|
|
if is_sentence_end(map, offset) {
|
|
break;
|
|
}
|
|
|
|
if is_possible_sentence_start(char) {
|
|
start = Some(offset);
|
|
}
|
|
|
|
previous_end = offset;
|
|
}
|
|
|
|
// Search forward for the end of the current sentence or if we are between sentences, the start of the next one
|
|
let mut end = relative_offset;
|
|
for (char, offset) in chars {
|
|
if start.is_none() && is_possible_sentence_start(char) {
|
|
if around {
|
|
start = Some(offset);
|
|
continue;
|
|
} else {
|
|
end = offset;
|
|
break;
|
|
}
|
|
}
|
|
|
|
if char != '\n' {
|
|
end = offset + char.len_utf8();
|
|
}
|
|
|
|
if is_sentence_end(map, end) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
let mut range = start.unwrap_or(previous_end).to_display_point(map)..end.to_display_point(map);
|
|
if around {
|
|
range = expand_to_include_whitespace(map, range, false);
|
|
}
|
|
|
|
Some(range)
|
|
}
|
|
|
|
fn is_possible_sentence_start(character: char) -> bool {
|
|
!character.is_whitespace() && character != '.'
|
|
}
|
|
|
|
const SENTENCE_END_PUNCTUATION: &[char] = &['.', '!', '?'];
|
|
const SENTENCE_END_FILLERS: &[char] = &[')', ']', '"', '\''];
|
|
const SENTENCE_END_WHITESPACE: &[char] = &[' ', '\t', '\n'];
|
|
fn is_sentence_end(map: &DisplaySnapshot, offset: usize) -> bool {
|
|
let mut next_chars = map.buffer_chars_at(offset).peekable();
|
|
if let Some((char, _)) = next_chars.next() {
|
|
// We are at a double newline. This position is a sentence end.
|
|
if char == '\n' && next_chars.peek().map(|(c, _)| c == &'\n').unwrap_or(false) {
|
|
return true;
|
|
}
|
|
|
|
// The next text is not a valid whitespace. This is not a sentence end
|
|
if !SENTENCE_END_WHITESPACE.contains(&char) {
|
|
return false;
|
|
}
|
|
}
|
|
|
|
for (char, _) in map.reverse_buffer_chars_at(offset) {
|
|
if SENTENCE_END_PUNCTUATION.contains(&char) {
|
|
return true;
|
|
}
|
|
|
|
if !SENTENCE_END_FILLERS.contains(&char) {
|
|
return false;
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
/// Expands the passed range to include whitespace on one side or the other in a line. Attempts to add the
|
|
/// whitespace to the end first and falls back to the start if there was none.
|
|
fn expand_to_include_whitespace(
|
|
map: &DisplaySnapshot,
|
|
range: Range<DisplayPoint>,
|
|
stop_at_newline: bool,
|
|
) -> Range<DisplayPoint> {
|
|
let mut range = range.start.to_offset(map, Bias::Left)..range.end.to_offset(map, Bias::Right);
|
|
let mut whitespace_included = false;
|
|
|
|
let mut chars = map.buffer_chars_at(range.end).peekable();
|
|
while let Some((char, offset)) = chars.next() {
|
|
if char == '\n' && stop_at_newline {
|
|
break;
|
|
}
|
|
|
|
if char.is_whitespace() {
|
|
if char != '\n' {
|
|
range.end = offset + char.len_utf8();
|
|
whitespace_included = true;
|
|
}
|
|
} else {
|
|
// Found non whitespace. Quit out.
|
|
break;
|
|
}
|
|
}
|
|
|
|
if !whitespace_included {
|
|
for (char, point) in map.reverse_buffer_chars_at(range.start) {
|
|
if char == '\n' && stop_at_newline {
|
|
break;
|
|
}
|
|
|
|
if !char.is_whitespace() {
|
|
break;
|
|
}
|
|
|
|
range.start = point;
|
|
}
|
|
}
|
|
|
|
range.start.to_display_point(map)..range.end.to_display_point(map)
|
|
}
|
|
|
|
/// If not `around` (i.e. inner), returns a range that surrounds the paragraph
|
|
/// where `relative_to` is in. If `around`, principally returns the range ending
|
|
/// at the end of the next paragraph.
|
|
///
|
|
/// Here, the "paragraph" is defined as a block of non-blank lines or a block of
|
|
/// blank lines. If the paragraph ends with a trailing newline (i.e. not with
|
|
/// EOF), the returned range ends at the trailing newline of the paragraph (i.e.
|
|
/// the trailing newline is not subject to subsequent operations).
|
|
///
|
|
/// Edge cases:
|
|
/// - If `around` and if the current paragraph is the last paragraph of the
|
|
/// file and is blank, then the selection results in an error.
|
|
/// - If `around` and if the current paragraph is the last paragraph of the
|
|
/// file and is not blank, then the returned range starts at the start of the
|
|
/// previous paragraph, if it exists.
|
|
fn paragraph(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
around: bool,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
let mut paragraph_start = start_of_paragraph(map, relative_to);
|
|
let mut paragraph_end = end_of_paragraph(map, relative_to);
|
|
|
|
let paragraph_end_row = paragraph_end.row();
|
|
let paragraph_ends_with_eof = paragraph_end_row == map.max_point().row();
|
|
let point = relative_to.to_point(map);
|
|
let current_line_is_empty = map.buffer_snapshot.is_line_blank(MultiBufferRow(point.row));
|
|
|
|
if around {
|
|
if paragraph_ends_with_eof {
|
|
if current_line_is_empty {
|
|
return None;
|
|
}
|
|
|
|
let paragraph_start_row = paragraph_start.row();
|
|
if paragraph_start_row.0 != 0 {
|
|
let previous_paragraph_last_line_start =
|
|
Point::new(paragraph_start_row.0 - 1, 0).to_display_point(map);
|
|
paragraph_start = start_of_paragraph(map, previous_paragraph_last_line_start);
|
|
}
|
|
} else {
|
|
let next_paragraph_start = Point::new(paragraph_end_row.0 + 1, 0).to_display_point(map);
|
|
paragraph_end = end_of_paragraph(map, next_paragraph_start);
|
|
}
|
|
}
|
|
|
|
let range = paragraph_start..paragraph_end;
|
|
Some(range)
|
|
}
|
|
|
|
/// Returns a position of the start of the current paragraph, where a paragraph
|
|
/// is defined as a run of non-blank lines or a run of blank lines.
|
|
pub fn start_of_paragraph(map: &DisplaySnapshot, display_point: DisplayPoint) -> DisplayPoint {
|
|
let point = display_point.to_point(map);
|
|
if point.row == 0 {
|
|
return DisplayPoint::zero();
|
|
}
|
|
|
|
let is_current_line_blank = map.buffer_snapshot.is_line_blank(MultiBufferRow(point.row));
|
|
|
|
for row in (0..point.row).rev() {
|
|
let blank = map.buffer_snapshot.is_line_blank(MultiBufferRow(row));
|
|
if blank != is_current_line_blank {
|
|
return Point::new(row + 1, 0).to_display_point(map);
|
|
}
|
|
}
|
|
|
|
DisplayPoint::zero()
|
|
}
|
|
|
|
/// Returns a position of the end of the current paragraph, where a paragraph
|
|
/// is defined as a run of non-blank lines or a run of blank lines.
|
|
/// The trailing newline is excluded from the paragraph.
|
|
pub fn end_of_paragraph(map: &DisplaySnapshot, display_point: DisplayPoint) -> DisplayPoint {
|
|
let point = display_point.to_point(map);
|
|
if point.row == map.max_buffer_row().0 {
|
|
return map.max_point();
|
|
}
|
|
|
|
let is_current_line_blank = map.buffer_snapshot.is_line_blank(MultiBufferRow(point.row));
|
|
|
|
for row in point.row + 1..map.max_buffer_row().0 + 1 {
|
|
let blank = map.buffer_snapshot.is_line_blank(MultiBufferRow(row));
|
|
if blank != is_current_line_blank {
|
|
let previous_row = row - 1;
|
|
return Point::new(
|
|
previous_row,
|
|
map.buffer_snapshot.line_len(MultiBufferRow(previous_row)),
|
|
)
|
|
.to_display_point(map);
|
|
}
|
|
}
|
|
|
|
map.max_point()
|
|
}
|
|
|
|
fn surrounding_markers(
|
|
map: &DisplaySnapshot,
|
|
relative_to: DisplayPoint,
|
|
around: bool,
|
|
search_across_lines: bool,
|
|
open_marker: char,
|
|
close_marker: char,
|
|
) -> Option<Range<DisplayPoint>> {
|
|
let point = relative_to.to_offset(map, Bias::Left);
|
|
|
|
let mut matched_closes = 0;
|
|
let mut opening = None;
|
|
|
|
let mut before_ch = match movement::chars_before(map, point).next() {
|
|
Some((ch, _)) => ch,
|
|
_ => '\0',
|
|
};
|
|
if let Some((ch, range)) = movement::chars_after(map, point).next() {
|
|
if ch == open_marker && before_ch != '\\' {
|
|
if open_marker == close_marker {
|
|
let mut total = 0;
|
|
for ((ch, _), (before_ch, _)) in movement::chars_before(map, point).tuple_windows()
|
|
{
|
|
if ch == '\n' {
|
|
break;
|
|
}
|
|
if ch == open_marker && before_ch != '\\' {
|
|
total += 1;
|
|
}
|
|
}
|
|
if total % 2 == 0 {
|
|
opening = Some(range)
|
|
}
|
|
} else {
|
|
opening = Some(range)
|
|
}
|
|
}
|
|
}
|
|
|
|
if opening.is_none() {
|
|
let mut chars_before = movement::chars_before(map, point).peekable();
|
|
while let Some((ch, range)) = chars_before.next() {
|
|
if ch == '\n' && !search_across_lines {
|
|
break;
|
|
}
|
|
|
|
if let Some((before_ch, _)) = chars_before.peek() {
|
|
if *before_ch == '\\' {
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if ch == open_marker {
|
|
if matched_closes == 0 {
|
|
opening = Some(range);
|
|
break;
|
|
}
|
|
matched_closes -= 1;
|
|
} else if ch == close_marker {
|
|
matched_closes += 1
|
|
}
|
|
}
|
|
}
|
|
if opening.is_none() {
|
|
for (ch, range) in movement::chars_after(map, point) {
|
|
if before_ch != '\\' {
|
|
if ch == open_marker {
|
|
opening = Some(range);
|
|
break;
|
|
} else if ch == close_marker {
|
|
break;
|
|
}
|
|
}
|
|
|
|
before_ch = ch;
|
|
}
|
|
}
|
|
|
|
let Some(mut opening) = opening else {
|
|
return None;
|
|
};
|
|
|
|
let mut matched_opens = 0;
|
|
let mut closing = None;
|
|
before_ch = match movement::chars_before(map, opening.end).next() {
|
|
Some((ch, _)) => ch,
|
|
_ => '\0',
|
|
};
|
|
for (ch, range) in movement::chars_after(map, opening.end) {
|
|
if ch == '\n' && !search_across_lines {
|
|
break;
|
|
}
|
|
|
|
if before_ch != '\\' {
|
|
if ch == close_marker {
|
|
if matched_opens == 0 {
|
|
closing = Some(range);
|
|
break;
|
|
}
|
|
matched_opens -= 1;
|
|
} else if ch == open_marker {
|
|
matched_opens += 1;
|
|
}
|
|
}
|
|
|
|
before_ch = ch;
|
|
}
|
|
|
|
let Some(mut closing) = closing else {
|
|
return None;
|
|
};
|
|
|
|
if around && !search_across_lines {
|
|
let mut found = false;
|
|
|
|
for (ch, range) in movement::chars_after(map, closing.end) {
|
|
if ch.is_whitespace() && ch != '\n' {
|
|
found = true;
|
|
closing.end = range.end;
|
|
} else {
|
|
break;
|
|
}
|
|
}
|
|
|
|
if !found {
|
|
for (ch, range) in movement::chars_before(map, opening.start) {
|
|
if ch.is_whitespace() && ch != '\n' {
|
|
opening.start = range.start
|
|
} else {
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if !around && search_across_lines {
|
|
if let Some((ch, range)) = movement::chars_after(map, opening.end).next() {
|
|
if ch == '\n' {
|
|
opening.end = range.end
|
|
}
|
|
}
|
|
|
|
for (ch, range) in movement::chars_before(map, closing.start) {
|
|
if !ch.is_whitespace() {
|
|
break;
|
|
}
|
|
if ch != '\n' {
|
|
closing.start = range.start
|
|
}
|
|
}
|
|
}
|
|
|
|
let result = if around {
|
|
opening.start..closing.end
|
|
} else {
|
|
opening.end..closing.start
|
|
};
|
|
|
|
Some(
|
|
map.clip_point(result.start.to_display_point(map), Bias::Left)
|
|
..map.clip_point(result.end.to_display_point(map), Bias::Right),
|
|
)
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod test {
|
|
use indoc::indoc;
|
|
|
|
use crate::{
|
|
state::Mode,
|
|
test::{NeovimBackedTestContext, VimTestContext},
|
|
};
|
|
|
|
const WORD_LOCATIONS: &str = indoc! {"
|
|
The quick ˇbrowˇnˇ•••
|
|
fox ˇjuˇmpsˇ over
|
|
the lazy dogˇ••
|
|
ˇ
|
|
ˇ
|
|
ˇ
|
|
Thˇeˇ-ˇquˇickˇ ˇbrownˇ•
|
|
ˇ••
|
|
ˇ••
|
|
ˇ fox-jumpˇs over
|
|
the lazy dogˇ•
|
|
ˇ
|
|
"
|
|
};
|
|
|
|
#[gpui::test]
|
|
async fn test_change_word_object(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
cx.simulate_at_each_offset("c i w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("c i shift-w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("c a w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("c a shift-w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_delete_word_object(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
cx.simulate_at_each_offset("d i w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("d i shift-w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("d a w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("d a shift-w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_visual_word_object(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
/*
|
|
cx.set_shared_state("The quick ˇbrown\nfox").await;
|
|
cx.simulate_shared_keystrokes(["v"]).await;
|
|
cx.assert_shared_state("The quick «bˇ»rown\nfox").await;
|
|
cx.simulate_shared_keystrokes(["i", "w"]).await;
|
|
cx.assert_shared_state("The quick «brownˇ»\nfox").await;
|
|
*/
|
|
cx.set_shared_state("The quick brown\nˇ\nfox").await;
|
|
cx.simulate_shared_keystrokes("v").await;
|
|
cx.shared_state()
|
|
.await
|
|
.assert_eq("The quick brown\n«\nˇ»fox");
|
|
cx.simulate_shared_keystrokes("i w").await;
|
|
cx.shared_state()
|
|
.await
|
|
.assert_eq("The quick brown\n«\nˇ»fox");
|
|
|
|
cx.simulate_at_each_offset("v i w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("v i shift-w", WORD_LOCATIONS)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
|
|
const PARAGRAPH_EXAMPLES: &[&'static str] = &[
|
|
// Single line
|
|
"ˇThe quick brown fox jumpˇs over the lazy dogˇ.ˇ",
|
|
// Multiple lines without empty lines
|
|
indoc! {"
|
|
ˇThe quick brownˇ
|
|
ˇfox jumps overˇ
|
|
the lazy dog.ˇ
|
|
"},
|
|
// Heading blank paragraph and trailing normal paragraph
|
|
indoc! {"
|
|
ˇ
|
|
ˇ
|
|
ˇThe quick brown fox jumps
|
|
ˇover the lazy dog.
|
|
ˇ
|
|
ˇ
|
|
ˇThe quick brown fox jumpsˇ
|
|
ˇover the lazy dog.ˇ
|
|
"},
|
|
// Inserted blank paragraph and trailing blank paragraph
|
|
indoc! {"
|
|
ˇThe quick brown fox jumps
|
|
ˇover the lazy dog.
|
|
ˇ
|
|
ˇ
|
|
ˇ
|
|
ˇThe quick brown fox jumpsˇ
|
|
ˇover the lazy dog.ˇ
|
|
ˇ
|
|
ˇ
|
|
ˇ
|
|
"},
|
|
// "Blank" paragraph with whitespace characters
|
|
indoc! {"
|
|
ˇThe quick brown fox jumps
|
|
over the lazy dog.
|
|
|
|
ˇ \t
|
|
|
|
ˇThe quick brown fox jumps
|
|
over the lazy dog.ˇ
|
|
ˇ
|
|
ˇ \t
|
|
\t \t
|
|
"},
|
|
// Single line "paragraphs", where selection size might be zero.
|
|
indoc! {"
|
|
ˇThe quick brown fox jumps over the lazy dog.
|
|
ˇ
|
|
ˇThe quick brown fox jumpˇs over the lazy dog.ˇ
|
|
ˇ
|
|
"},
|
|
];
|
|
|
|
#[gpui::test]
|
|
async fn test_change_paragraph_object(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
for paragraph_example in PARAGRAPH_EXAMPLES {
|
|
cx.simulate_at_each_offset("c i p", paragraph_example)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("c a p", paragraph_example)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_delete_paragraph_object(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
for paragraph_example in PARAGRAPH_EXAMPLES {
|
|
cx.simulate_at_each_offset("d i p", paragraph_example)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("d a p", paragraph_example)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_visual_paragraph_object(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
const EXAMPLES: &[&'static str] = &[
|
|
indoc! {"
|
|
ˇThe quick brown
|
|
fox jumps over
|
|
the lazy dog.
|
|
"},
|
|
indoc! {"
|
|
ˇ
|
|
|
|
ˇThe quick brown fox jumps
|
|
over the lazy dog.
|
|
ˇ
|
|
|
|
ˇThe quick brown fox jumps
|
|
over the lazy dog.
|
|
"},
|
|
indoc! {"
|
|
ˇThe quick brown fox jumps over the lazy dog.
|
|
ˇ
|
|
ˇThe quick brown fox jumps over the lazy dog.
|
|
|
|
"},
|
|
];
|
|
|
|
for paragraph_example in EXAMPLES {
|
|
cx.simulate_at_each_offset("v i p", paragraph_example)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset("v a p", paragraph_example)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
}
|
|
|
|
// Test string with "`" for opening surrounders and "'" for closing surrounders
|
|
const SURROUNDING_MARKER_STRING: &str = indoc! {"
|
|
ˇTh'ˇe ˇ`ˇ'ˇquˇi`ˇck broˇ'wn`
|
|
'ˇfox juˇmps ov`ˇer
|
|
the ˇlazy d'o`ˇg"};
|
|
|
|
const SURROUNDING_OBJECTS: &[(char, char)] = &[
|
|
('"', '"'), // Double Quote
|
|
('(', ')'), // Parentheses
|
|
];
|
|
|
|
#[gpui::test]
|
|
async fn test_change_surrounding_character_objects(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
for (start, end) in SURROUNDING_OBJECTS {
|
|
let marked_string = SURROUNDING_MARKER_STRING
|
|
.replace('`', &start.to_string())
|
|
.replace('\'', &end.to_string());
|
|
|
|
cx.simulate_at_each_offset(&format!("c i {start}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset(&format!("c i {end}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset(&format!("c a {start}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset(&format!("c a {end}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
}
|
|
#[gpui::test]
|
|
async fn test_singleline_surrounding_character_objects(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
cx.set_shared_wrap(12).await;
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"\"ˇhello world\"!"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"\"«hello worldˇ»\"!"
|
|
});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"\"hˇello world\"!"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"\"«hello worldˇ»\"!"
|
|
});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"helˇlo \"world\"!"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"hello \"«worldˇ»\"!"
|
|
});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"hello \"wˇorld\"!"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"hello \"«worldˇ»\"!"
|
|
});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"hello \"wˇorld\"!"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v a \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"hello« \"world\"ˇ»!"
|
|
});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"hello \"wˇorld\" !"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v a \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"hello «\"world\" ˇ»!"
|
|
});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"hello \"wˇorld\"•
|
|
goodbye"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v a \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"hello «\"world\" ˇ»
|
|
goodbye"
|
|
});
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_multiline_surrounding_character_objects(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"func empty(a string) bool {
|
|
if a == \"\" {
|
|
return true
|
|
}
|
|
ˇreturn false
|
|
}"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i {").await;
|
|
cx.shared_state().await.assert_eq(indoc! {"
|
|
func empty(a string) bool {
|
|
« if a == \"\" {
|
|
return true
|
|
}
|
|
return false
|
|
ˇ»}"});
|
|
cx.set_shared_state(indoc! {
|
|
"func empty(a string) bool {
|
|
if a == \"\" {
|
|
ˇreturn true
|
|
}
|
|
return false
|
|
}"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i {").await;
|
|
cx.shared_state().await.assert_eq(indoc! {"
|
|
func empty(a string) bool {
|
|
if a == \"\" {
|
|
« return true
|
|
ˇ» }
|
|
return false
|
|
}"});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"func empty(a string) bool {
|
|
if a == \"\" ˇ{
|
|
return true
|
|
}
|
|
return false
|
|
}"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i {").await;
|
|
cx.shared_state().await.assert_eq(indoc! {"
|
|
func empty(a string) bool {
|
|
if a == \"\" {
|
|
« return true
|
|
ˇ» }
|
|
return false
|
|
}"});
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_singleline_surrounding_character_objects_with_escape(
|
|
cx: &mut gpui::TestAppContext,
|
|
) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
cx.set_shared_state(indoc! {
|
|
"h\"e\\\"lˇlo \\\"world\"!"
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"h\"«e\\\"llo \\\"worldˇ»\"!"
|
|
});
|
|
|
|
cx.set_shared_state(indoc! {
|
|
"hello \"teˇst \\\"inside\\\" world\""
|
|
})
|
|
.await;
|
|
cx.simulate_shared_keystrokes("v i \"").await;
|
|
cx.shared_state().await.assert_eq(indoc! {
|
|
"hello \"«test \\\"inside\\\" worldˇ»\""
|
|
});
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_vertical_bars(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = VimTestContext::new(cx, true).await;
|
|
cx.set_state(
|
|
indoc! {"
|
|
fn boop() {
|
|
baz(ˇ|a, b| { bar(|j, k| { })})
|
|
}"
|
|
},
|
|
Mode::Normal,
|
|
);
|
|
cx.simulate_keystrokes("c i |");
|
|
cx.assert_state(
|
|
indoc! {"
|
|
fn boop() {
|
|
baz(|ˇ| { bar(|j, k| { })})
|
|
}"
|
|
},
|
|
Mode::Insert,
|
|
);
|
|
cx.simulate_keystrokes("escape 1 8 |");
|
|
cx.assert_state(
|
|
indoc! {"
|
|
fn boop() {
|
|
baz(|| { bar(ˇ|j, k| { })})
|
|
}"
|
|
},
|
|
Mode::Normal,
|
|
);
|
|
|
|
cx.simulate_keystrokes("v a |");
|
|
cx.assert_state(
|
|
indoc! {"
|
|
fn boop() {
|
|
baz(|| { bar(«|j, k| ˇ»{ })})
|
|
}"
|
|
},
|
|
Mode::Visual,
|
|
);
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_argument_object(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = VimTestContext::new(cx, true).await;
|
|
|
|
// Generic arguments
|
|
cx.set_state("fn boop<A: ˇDebug, B>() {}", Mode::Normal);
|
|
cx.simulate_keystrokes("v i a");
|
|
cx.assert_state("fn boop<«A: Debugˇ», B>() {}", Mode::Visual);
|
|
|
|
// Function arguments
|
|
cx.set_state(
|
|
"fn boop(ˇarg_a: (Tuple, Of, Types), arg_b: String) {}",
|
|
Mode::Normal,
|
|
);
|
|
cx.simulate_keystrokes("d a a");
|
|
cx.assert_state("fn boop(ˇarg_b: String) {}", Mode::Normal);
|
|
|
|
cx.set_state("std::namespace::test(\"strinˇg\", a.b.c())", Mode::Normal);
|
|
cx.simulate_keystrokes("v a a");
|
|
cx.assert_state("std::namespace::test(«\"string\", ˇ»a.b.c())", Mode::Visual);
|
|
|
|
// Tuple, vec, and array arguments
|
|
cx.set_state(
|
|
"fn boop(arg_a: (Tuple, Ofˇ, Types), arg_b: String) {}",
|
|
Mode::Normal,
|
|
);
|
|
cx.simulate_keystrokes("c i a");
|
|
cx.assert_state(
|
|
"fn boop(arg_a: (Tuple, ˇ, Types), arg_b: String) {}",
|
|
Mode::Insert,
|
|
);
|
|
|
|
cx.set_state("let a = (test::call(), 'p', my_macro!{ˇ});", Mode::Normal);
|
|
cx.simulate_keystrokes("c a a");
|
|
cx.assert_state("let a = (test::call(), 'p'ˇ);", Mode::Insert);
|
|
|
|
cx.set_state("let a = [test::call(ˇ), 300];", Mode::Normal);
|
|
cx.simulate_keystrokes("c i a");
|
|
cx.assert_state("let a = [ˇ, 300];", Mode::Insert);
|
|
|
|
cx.set_state(
|
|
"let a = vec![Vec::new(), vecˇ![test::call(), 300]];",
|
|
Mode::Normal,
|
|
);
|
|
cx.simulate_keystrokes("c a a");
|
|
cx.assert_state("let a = vec![Vec::new()ˇ];", Mode::Insert);
|
|
|
|
// Cursor immediately before / after brackets
|
|
cx.set_state("let a = [test::call(first_arg)ˇ]", Mode::Normal);
|
|
cx.simulate_keystrokes("v i a");
|
|
cx.assert_state("let a = [«test::call(first_arg)ˇ»]", Mode::Visual);
|
|
|
|
cx.set_state("let a = [test::callˇ(first_arg)]", Mode::Normal);
|
|
cx.simulate_keystrokes("v i a");
|
|
cx.assert_state("let a = [«test::call(first_arg)ˇ»]", Mode::Visual);
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_delete_surrounding_character_objects(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = NeovimBackedTestContext::new(cx).await;
|
|
|
|
for (start, end) in SURROUNDING_OBJECTS {
|
|
let marked_string = SURROUNDING_MARKER_STRING
|
|
.replace('`', &start.to_string())
|
|
.replace('\'', &end.to_string());
|
|
|
|
cx.simulate_at_each_offset(&format!("d i {start}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset(&format!("d i {end}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset(&format!("d a {start}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
cx.simulate_at_each_offset(&format!("d a {end}"), &marked_string)
|
|
.await
|
|
.assert_matches();
|
|
}
|
|
}
|
|
|
|
#[gpui::test]
|
|
async fn test_tags(cx: &mut gpui::TestAppContext) {
|
|
let mut cx = VimTestContext::new_html(cx).await;
|
|
|
|
cx.set_state("<html><head></head><body><b>hˇi!</b></body>", Mode::Normal);
|
|
cx.simulate_keystrokes("v i t");
|
|
cx.assert_state(
|
|
"<html><head></head><body><b>«hi!ˇ»</b></body>",
|
|
Mode::Visual,
|
|
);
|
|
cx.simulate_keystrokes("a t");
|
|
cx.assert_state(
|
|
"<html><head></head><body>«<b>hi!</b>ˇ»</body>",
|
|
Mode::Visual,
|
|
);
|
|
cx.simulate_keystrokes("a t");
|
|
cx.assert_state(
|
|
"<html><head></head>«<body><b>hi!</b></body>ˇ»",
|
|
Mode::Visual,
|
|
);
|
|
|
|
// The cursor is before the tag
|
|
cx.set_state(
|
|
"<html><head></head><body> ˇ <b>hi!</b></body>",
|
|
Mode::Normal,
|
|
);
|
|
cx.simulate_keystrokes("v i t");
|
|
cx.assert_state(
|
|
"<html><head></head><body> <b>«hi!ˇ»</b></body>",
|
|
Mode::Visual,
|
|
);
|
|
cx.simulate_keystrokes("a t");
|
|
cx.assert_state(
|
|
"<html><head></head><body> «<b>hi!</b>ˇ»</body>",
|
|
Mode::Visual,
|
|
);
|
|
|
|
// The cursor is in the open tag
|
|
cx.set_state(
|
|
"<html><head></head><body><bˇ>hi!</b><b>hello!</b></body>",
|
|
Mode::Normal,
|
|
);
|
|
cx.simulate_keystrokes("v a t");
|
|
cx.assert_state(
|
|
"<html><head></head><body>«<b>hi!</b>ˇ»<b>hello!</b></body>",
|
|
Mode::Visual,
|
|
);
|
|
cx.simulate_keystrokes("i t");
|
|
cx.assert_state(
|
|
"<html><head></head><body>«<b>hi!</b><b>hello!</b>ˇ»</body>",
|
|
Mode::Visual,
|
|
);
|
|
|
|
// current selection length greater than 1
|
|
cx.set_state(
|
|
"<html><head></head><body><«b>hi!ˇ»</b></body>",
|
|
Mode::Visual,
|
|
);
|
|
cx.simulate_keystrokes("i t");
|
|
cx.assert_state(
|
|
"<html><head></head><body><b>«hi!ˇ»</b></body>",
|
|
Mode::Visual,
|
|
);
|
|
cx.simulate_keystrokes("a t");
|
|
cx.assert_state(
|
|
"<html><head></head><body>«<b>hi!</b>ˇ»</body>",
|
|
Mode::Visual,
|
|
);
|
|
|
|
cx.set_state(
|
|
"<html><head></head><body><«b>hi!</ˇ»b></body>",
|
|
Mode::Visual,
|
|
);
|
|
cx.simulate_keystrokes("a t");
|
|
cx.assert_state(
|
|
"<html><head></head>«<body><b>hi!</b></body>ˇ»",
|
|
Mode::Visual,
|
|
);
|
|
}
|
|
}
|