1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
use unicode_width::{UnicodeWidthChar, UnicodeWidthStr};
/// returns printed length of string
/// if ansi feature enabled, takes into account escape codes
#[inline(always)]
pub fn measure_text_width(s: &str) -> usize {
s.width()
}
/// Split a line into its individual parts along the given delimiter.
pub fn split_line_by_delimiter(line: &str, delimiter: char) -> Vec<String> {
line.split(delimiter)
.map(ToString::to_string)
.collect::<Vec<String>>()
}
/// Splits a long word at a given character width.
/// This needs some special logic, as we have to take multi-character UTF-8 symbols into account.
/// When simply splitting at a certain char position, we might end up with a string that's has a
/// wider display width than allowed.
pub fn split_long_word(allowed_width: usize, word: &str) -> (String, String) {
let mut current_width = 0;
let mut parts = String::new();
let mut char_iter = word.chars().peekable();
// Check if the string might be too long, one character at a time.
// Peek into the next char and check the exit condition.
// That is, pushing the next character would result in the string being too long.
while let Some(c) = char_iter.peek() {
if (current_width + c.width().unwrap_or(1)) > allowed_width {
break;
}
// We can unwrap, as we just checked that a suitable character is next in line.
let c = char_iter.next().unwrap();
// We default to 1 char, if the character length cannot be determined.
// The user has to live with this, if they decide to add control characters or some fancy
// stuff into their tables. This is considered undefined behavior and we try to handle this
// to the best of our capabilities.
let character_width = c.width().unwrap_or(1);
current_width += character_width;
parts.push(c);
}
// Collect the remaining characters.
let remaining = char_iter.collect();
(parts, remaining)
}