diff options
author | Blaž Hrastnik | 2021-04-05 07:35:04 +0000 |
---|---|---|
committer | Blaž Hrastnik | 2021-04-05 07:35:04 +0000 |
commit | 59a0fc7b59186b3bedb01dd5b958d3b97b9fbba2 (patch) | |
tree | 820018ef4c0719be2daef34cab3ccc5febd67850 /helix-core/src | |
parent | 16350399ac4eb15274e065f113ad808c96dc600f (diff) |
w, b, e: Match kakoune's behavior in selecting by default.
I initially preferred only moving the cursor, but selecting the whole
word is a lot nicer for things like wd (instead of vwd).
Diffstat (limited to 'helix-core/src')
-rw-r--r-- | helix-core/src/movement.rs | 105 |
1 files changed, 61 insertions, 44 deletions
diff --git a/helix-core/src/movement.rs b/helix-core/src/movement.rs index 88d6df0a..02e5b8bf 100644 --- a/helix-core/src/movement.rs +++ b/helix-core/src/movement.rs @@ -61,99 +61,110 @@ pub fn move_vertically( range } -pub fn move_next_word_start(slice: RopeSlice, mut pos: usize, count: usize) -> usize { +pub fn move_next_word_start(slice: RopeSlice, mut begin: usize, count: usize) -> Option<Range> { + let mut end = begin; + for _ in 0..count { - if pos + 1 == slice.len_chars() { - return pos; + if begin + 1 == slice.len_chars() { + return None; } - let mut ch = slice.char(pos); - let next = slice.char(pos + 1); + let mut ch = slice.char(begin); + let next = slice.char(begin + 1); // if we're at the end of a word, or on whitespce right before new one if categorize(ch) != categorize(next) { - pos += 1; - ch = next; + begin += 1; } + // return if not skip while? + skip_over_next(slice, &mut begin, |ch| ch == '\n'); + ch = slice.char(begin); + + end = begin + 1; + if is_word(ch) { - skip_over_next(slice, &mut pos, is_word); + skip_over_next(slice, &mut end, is_word); } else if ch.is_ascii_punctuation() { - skip_over_next(slice, &mut pos, |ch| ch.is_ascii_punctuation()); + skip_over_next(slice, &mut end, |ch| ch.is_ascii_punctuation()); } - // TODO: don't include newline? - skip_over_next(slice, &mut pos, |ch| ch.is_ascii_whitespace()); + skip_over_next(slice, &mut end, is_horiz_blank); } - pos + Some(Range::new(begin, end - 1)) } -pub fn move_prev_word_start(slice: RopeSlice, mut pos: usize, count: usize) -> usize { +pub fn move_prev_word_start(slice: RopeSlice, mut begin: usize, count: usize) -> Option<Range> { + let mut with_end = false; + let mut end = begin; + for _ in 0..count { - if pos == 0 { - return pos; + if begin == 0 { + return None; } - let ch = slice.char(pos); - let prev = slice.char(pos - 1); + let ch = slice.char(begin); + let prev = slice.char(begin - 1); if categorize(ch) != categorize(prev) { - pos -= 1; + begin -= 1; } - // match (category c1, category c2) => { - // if c1 != c2 { - // } - // } + // return if not skip while? + skip_over_prev(slice, &mut begin, |ch| ch == '\n'); - // TODO: skip while eol + end = begin; - // TODO: don't include newline? - skip_over_prev(slice, &mut pos, |ch| ch.is_ascii_whitespace()); + with_end = skip_over_prev(slice, &mut end, is_horiz_blank); // refetch - let ch = slice.char(pos); + let ch = slice.char(end); if is_word(ch) { - skip_over_prev(slice, &mut pos, is_word); + with_end = skip_over_prev(slice, &mut end, is_word); } else if ch.is_ascii_punctuation() { - skip_over_prev(slice, &mut pos, |ch| ch.is_ascii_punctuation()); + with_end = skip_over_prev(slice, &mut end, |ch| ch.is_ascii_punctuation()); } - pos = pos.saturating_add(1) } - pos + // we want to include begin + Some(Range::new(begin + 1, if with_end { end } else { end + 1 })) } -pub fn move_next_word_end(slice: RopeSlice, mut pos: usize, count: usize) -> usize { +pub fn move_next_word_end(slice: RopeSlice, mut begin: usize, count: usize) -> Option<Range> { + let mut end = begin; + for _ in 0..count { - if pos + 1 == slice.len_chars() { - return pos; + if begin + 1 == slice.len_chars() { + return None; } - let ch = slice.char(pos); - let next = slice.char(pos + 1); + let ch = slice.char(begin); + let next = slice.char(begin + 1); if categorize(ch) != categorize(next) { - pos += 1; + begin += 1; } - // TODO: don't include newline? - skip_over_next(slice, &mut pos, |ch| ch.is_ascii_whitespace()); + // return if not skip while? + skip_over_next(slice, &mut begin, |ch| ch == '\n'); + + end = begin; + + skip_over_next(slice, &mut end, is_horiz_blank); // refetch - let ch = slice.char(pos); + let ch = slice.char(end); if is_word(ch) { - skip_over_next(slice, &mut pos, is_word); + skip_over_next(slice, &mut end, is_word); } else if ch.is_ascii_punctuation() { - skip_over_next(slice, &mut pos, |ch| ch.is_ascii_punctuation()); + skip_over_next(slice, &mut end, |ch| ch.is_ascii_punctuation()); } - pos -= 1 } - pos + Some(Range::new(begin, end - 1)) } // ---- util ------------ @@ -164,6 +175,10 @@ fn is_word(ch: char) -> bool { ch.is_alphanumeric() || ch == '_' } +fn is_horiz_blank(ch: char) -> bool { + matches!(ch, ' ' | '\t') +} + #[derive(Debug, Eq, PartialEq)] enum Category { Whitespace, @@ -201,7 +216,8 @@ where } #[inline] -pub fn skip_over_prev<F>(slice: RopeSlice, pos: &mut usize, fun: F) +/// Returns true if the final pos matches the predicate. +pub fn skip_over_prev<F>(slice: RopeSlice, pos: &mut usize, fun: F) -> bool where F: Fn(char) -> bool, { @@ -214,6 +230,7 @@ where } *pos = pos.saturating_sub(1); } + return fun(slice.char(*pos)); } #[cfg(test)] |