No cursor yet, just always appends to the end of the current buffer. But still!
467 lines
13 KiB
Rust
467 lines
13 KiB
Rust
#![allow(dead_code)]
|
|
|
|
use std::cmp::min;
|
|
use std::mem;
|
|
use std::fmt;
|
|
use std;
|
|
|
|
|
|
fn newline_count(text: &str) -> uint {
|
|
let mut count = 0;
|
|
for c in text.chars() {
|
|
if c == '\n' {
|
|
count += 1;
|
|
}
|
|
}
|
|
return count;
|
|
}
|
|
|
|
fn char_count(text: &str) -> uint {
|
|
let mut count = 0;
|
|
for _ in text.chars() {
|
|
count += 1;
|
|
}
|
|
return count;
|
|
}
|
|
|
|
fn char_and_newline_count(text: &str) -> (uint, uint) {
|
|
let mut char_count = 0;
|
|
let mut newline_count = 0;
|
|
|
|
for c in text.chars() {
|
|
char_count += 1;
|
|
if c == '\n' {
|
|
newline_count += 1;
|
|
}
|
|
}
|
|
|
|
return (char_count, newline_count);
|
|
}
|
|
|
|
fn char_pos_to_byte_pos(text: &str, pos: uint) -> uint {
|
|
let mut i: uint = 0;
|
|
|
|
for (offset, _) in text.char_indices() {
|
|
if i == pos {
|
|
return offset;
|
|
}
|
|
i += 1;
|
|
}
|
|
|
|
if i == pos {
|
|
return text.len();
|
|
}
|
|
|
|
panic!("char_pos_to_byte_pos(): char position off the end of the string.");
|
|
}
|
|
|
|
|
|
/// A block of text, contiguous in memory
|
|
pub struct TextBlock {
|
|
pub data: Vec<u8>,
|
|
}
|
|
|
|
impl TextBlock {
|
|
/// Create a new empty text block.
|
|
pub fn new() -> TextBlock {
|
|
TextBlock {
|
|
data: Vec::<u8>::new()
|
|
}
|
|
}
|
|
|
|
/// Create a new text block with the contents of 'text'.
|
|
pub fn new_from_str(text: &str) -> TextBlock {
|
|
let mut tb = TextBlock {
|
|
data: Vec::<u8>::with_capacity(text.len())
|
|
};
|
|
|
|
for b in text.bytes() {
|
|
tb.data.push(b);
|
|
}
|
|
|
|
return tb;
|
|
}
|
|
|
|
/// Return the length of the text block in bytes.
|
|
pub fn len(&self) -> uint {
|
|
self.data.len()
|
|
}
|
|
|
|
/// Insert 'text' at char position 'pos'.
|
|
pub fn insert_text(&mut self, text: &str, pos: uint) {
|
|
// Find insertion position in bytes
|
|
let byte_pos = char_pos_to_byte_pos(self.as_str(), pos);
|
|
|
|
// Grow data size
|
|
self.data.grow(text.len(), 0);
|
|
|
|
// Move old bytes forward
|
|
let mut from = self.data.len() - text.len();
|
|
let mut to = self.data.len();
|
|
while from > byte_pos {
|
|
from -= 1;
|
|
to -= 1;
|
|
|
|
self.data[to] = self.data[from];
|
|
}
|
|
|
|
// Copy new bytes in
|
|
let mut i = byte_pos;
|
|
for b in text.bytes() {
|
|
self.data[i] = b;
|
|
i += 1
|
|
}
|
|
}
|
|
|
|
/// Remove the text between char positions 'pos_a' and 'pos_b'.
|
|
pub fn remove_text(&mut self, pos_a: uint, pos_b: uint) {
|
|
// Bounds checks
|
|
if pos_a > pos_b {
|
|
panic!("TextBlock::remove_text(): pos_a must be less than or equal to pos_b.");
|
|
}
|
|
|
|
// Find removal positions in bytes
|
|
let byte_pos_a = char_pos_to_byte_pos(self.as_str(), pos_a);
|
|
let byte_pos_b = char_pos_to_byte_pos(self.as_str(), pos_b);
|
|
|
|
// Move bytes to fill in the gap left by the removed bytes
|
|
let mut from = byte_pos_b;
|
|
let mut to = byte_pos_a;
|
|
while from < self.data.len() {
|
|
self.data[to] = self.data[from];
|
|
|
|
from += 1;
|
|
to += 1;
|
|
}
|
|
|
|
// Remove data from the end
|
|
let final_size = self.data.len() + byte_pos_a - byte_pos_b;
|
|
self.data.truncate(final_size);
|
|
}
|
|
|
|
/// Returns an immutable string slice into the text block's memory
|
|
pub fn as_str<'a>(&'a self) -> &'a str {
|
|
unsafe {
|
|
mem::transmute(self.data.as_slice())
|
|
}
|
|
}
|
|
}
|
|
|
|
impl fmt::Show for TextBlock {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
write!(f, "{}", self.as_str())
|
|
}
|
|
}
|
|
|
|
|
|
|
|
|
|
/// A text rope node, using TextBlocks for its underlying text
|
|
/// storage.
|
|
// TODO: record number of graphines as well, to support utf8 properly
|
|
pub struct TextNode {
|
|
pub data: TextNodeData,
|
|
pub char_count: uint,
|
|
pub newline_count: uint,
|
|
}
|
|
|
|
pub enum TextNodeData {
|
|
Leaf(TextBlock),
|
|
Branch(Box<TextNode>, Box<TextNode>)
|
|
}
|
|
|
|
const MIN_LEAF_SIZE: uint = 64;
|
|
const MAX_LEAF_SIZE: uint = MIN_LEAF_SIZE * 2;
|
|
|
|
|
|
impl TextNode {
|
|
pub fn new() -> TextNode {
|
|
TextNode {
|
|
data: TextNodeData::Leaf(TextBlock::new()),
|
|
char_count: 0,
|
|
newline_count: 0,
|
|
}
|
|
}
|
|
|
|
pub fn new_from_str(text: &str) -> TextNode {
|
|
TextNode {
|
|
data: TextNodeData::Leaf(TextBlock::new_from_str(text)),
|
|
char_count: char_count(text),
|
|
newline_count: newline_count(text),
|
|
}
|
|
}
|
|
|
|
/// Recursively splits a leaf node into roughly equal-sized children,
|
|
/// being no larger than 'max_size'.
|
|
pub fn split(&mut self, max_size: uint) {
|
|
if let TextNodeData::Branch(_, _) = self.data {
|
|
panic!("TextNode::split(): attempt to split a non-leaf node.");
|
|
}
|
|
|
|
if self.char_count > max_size {
|
|
// Split data into two new text blocks
|
|
let mut tn1 = box TextNode::new();
|
|
let mut tn2 = box TextNode::new();
|
|
if let TextNodeData::Leaf(ref mut tb) = self.data {
|
|
let pos = tb.len() / 2;
|
|
tn1 = box TextNode::new_from_str(tb.as_str().slice(0, pos));
|
|
tn2 = box TextNode::new_from_str(tb.as_str().slice(pos, tb.len()));
|
|
}
|
|
|
|
tn1.split(max_size);
|
|
tn2.split(max_size);
|
|
|
|
// Swap the old and new data
|
|
let mut new_data = TextNodeData::Branch(tn1, tn2);
|
|
mem::swap(&mut self.data, &mut new_data);
|
|
}
|
|
}
|
|
|
|
/// Merges the data of a non-leaf node to make it a leaf node
|
|
pub fn merge(&mut self) {
|
|
if let TextNodeData::Branch(_, _) = self.data {
|
|
let mut s: String = String::from_str("");
|
|
|
|
if let TextNodeData::Branch(ref mut left, ref mut right) = self.data {
|
|
// Merge left and right children first, to make sure we're dealing
|
|
// with leafs
|
|
if let TextNodeData::Branch(_, _) = left.data { left.merge(); }
|
|
if let TextNodeData::Branch(_, _) = right.data { right.merge(); }
|
|
|
|
// Push data into a string
|
|
if let TextNodeData::Leaf(ref tb) = left.data {
|
|
s.push_str(tb.as_str());
|
|
}
|
|
if let TextNodeData::Leaf(ref tb) = right.data {
|
|
s.push_str(tb.as_str());
|
|
}
|
|
}
|
|
|
|
self.data = TextNodeData::Leaf(TextBlock::new_from_str(s.as_slice()));
|
|
}
|
|
}
|
|
|
|
/// Insert 'text' at position 'pos'.
|
|
pub fn insert_text(&mut self, text: &str, pos: uint) {
|
|
if pos > self.char_count {
|
|
panic!("TextNode::insert_text(): attempt to insert text after end of node text.");
|
|
}
|
|
|
|
match self.data {
|
|
TextNodeData::Leaf(_) => {
|
|
if let TextNodeData::Leaf(ref mut tb) = self.data {
|
|
tb.insert_text(text, pos);
|
|
|
|
let (cc, nlc) = char_and_newline_count(text);
|
|
self.char_count += cc;
|
|
self.newline_count += nlc;
|
|
}
|
|
|
|
if self.char_count > MAX_LEAF_SIZE {
|
|
self.split(MAX_LEAF_SIZE);
|
|
}
|
|
},
|
|
|
|
TextNodeData::Branch(ref mut left, ref mut right) => {
|
|
if pos <= left.char_count {
|
|
left.insert_text(text, pos);
|
|
}
|
|
else {
|
|
right.insert_text(text, pos - left.char_count);
|
|
}
|
|
|
|
self.char_count = left.char_count + right.char_count;
|
|
self.newline_count = left.newline_count + right.newline_count;
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Remove the text between byte positions 'pos_a' and 'pos_b'.
|
|
pub fn remove_text(&mut self, pos_a: uint, pos_b: uint) {
|
|
// Bounds checks
|
|
if pos_a > pos_b {
|
|
panic!("TextNode::remove_text(): pos_a must be less than or equal to pos_b.");
|
|
}
|
|
if pos_b > self.char_count {
|
|
panic!("TextNode::remove_text(): attempt to remove text after end of node text.");
|
|
}
|
|
|
|
match self.data {
|
|
TextNodeData::Leaf(ref mut tb) => {
|
|
tb.remove_text(pos_a, pos_b);
|
|
|
|
let (cc, nlc) = char_and_newline_count(tb.as_str());
|
|
self.char_count = cc;
|
|
self.newline_count = nlc;
|
|
},
|
|
|
|
TextNodeData::Branch(ref mut left, ref mut right) => {
|
|
let lbc = left.char_count;
|
|
|
|
if pos_a < lbc {
|
|
left.remove_text(pos_a, min(pos_b, lbc));
|
|
}
|
|
|
|
if pos_b > lbc {
|
|
right.remove_text(pos_a - min(pos_a, lbc), pos_b - lbc);
|
|
}
|
|
|
|
self.char_count = left.char_count + right.char_count;
|
|
self.newline_count = left.newline_count + right.newline_count;
|
|
}
|
|
}
|
|
|
|
if self.char_count < MIN_LEAF_SIZE {
|
|
self.merge();
|
|
}
|
|
}
|
|
}
|
|
|
|
impl fmt::Show for TextNode {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
match self.data {
|
|
TextNodeData::Leaf(ref tb) => {
|
|
tb.fmt(f)
|
|
},
|
|
|
|
TextNodeData::Branch(ref left, ref right) => {
|
|
try!(left.fmt(f));
|
|
right.fmt(f)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
|
|
|
|
/// A text buffer
|
|
pub struct TextBuffer {
|
|
root: TextNode
|
|
}
|
|
|
|
impl TextBuffer {
|
|
pub fn new() -> TextBuffer {
|
|
TextBuffer {
|
|
root: TextNode::new()
|
|
}
|
|
}
|
|
|
|
pub fn len(&self) -> uint {
|
|
self.root.char_count
|
|
}
|
|
|
|
/// Insert 'text' at char position 'pos'.
|
|
pub fn insert_text(&mut self, text: &str, pos: uint) {
|
|
self.root.insert_text(text, pos);
|
|
}
|
|
|
|
/// Remove the text between char positions 'pos_a' and 'pos_b'.
|
|
pub fn remove_text(&mut self, pos_a: uint, pos_b: uint) {
|
|
self.root.remove_text(pos_a, pos_b);
|
|
}
|
|
|
|
pub fn root_iter<'a>(&'a self) -> TextBufferIter<'a> {
|
|
let mut node_stack: Vec<&'a TextNode> = Vec::new();
|
|
let mut cur_node = &self.root;
|
|
|
|
loop {
|
|
match cur_node.data {
|
|
TextNodeData::Leaf(_) => {
|
|
break;
|
|
},
|
|
|
|
TextNodeData::Branch(ref left, ref right) => {
|
|
node_stack.push(&(**right));
|
|
cur_node = &(**left);
|
|
}
|
|
}
|
|
}
|
|
|
|
TextBufferIter {
|
|
node_stack: node_stack,
|
|
cur_block: match cur_node.data {
|
|
TextNodeData::Leaf(ref tb) => tb.as_str().chars(),
|
|
_ => panic!("This should never happen.")
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
impl fmt::Show for TextBuffer {
|
|
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
|
|
self.root.fmt(f)
|
|
}
|
|
}
|
|
|
|
|
|
|
|
|
|
/// An iterator over a text buffer
|
|
pub struct TextBufferIter<'a> {
|
|
node_stack: Vec<&'a TextNode>,
|
|
cur_block: std::str::Chars<'a>,
|
|
}
|
|
|
|
|
|
impl<'a> TextBufferIter<'a> {
|
|
// Puts the iterator on the next line
|
|
pub fn next_line(&mut self) -> Option<char> {
|
|
// TODO: more efficient implementation, taking advantage of rope
|
|
// structure.
|
|
for c in *self {
|
|
if c == '\n' {
|
|
return Option::Some(c);
|
|
}
|
|
}
|
|
|
|
return Option::None;
|
|
}
|
|
|
|
|
|
// Skips the iterator n characters ahead
|
|
pub fn skip_chars(&mut self, n: uint) {
|
|
// TODO: more efficient implementation, taking advantage of rope
|
|
// structure.
|
|
for _ in range(0, n) {
|
|
if let Option::None = self.next() {
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
impl<'a> Iterator<char> for TextBufferIter<'a> {
|
|
fn next(&mut self) -> Option<char> {
|
|
if let Option::Some(c) = self.cur_block.next() {
|
|
return Option::Some(c);
|
|
}
|
|
|
|
loop {
|
|
if let Option::Some(node) = self.node_stack.pop() {
|
|
match node.data {
|
|
TextNodeData::Leaf(ref tb) => {
|
|
self.cur_block = tb.as_str().chars();
|
|
|
|
if let Option::Some(c) = self.cur_block.next() {
|
|
return Option::Some(c);
|
|
}
|
|
else {
|
|
continue;
|
|
}
|
|
},
|
|
|
|
TextNodeData::Branch(ref left, ref right) => {
|
|
self.node_stack.push(&(**right));
|
|
self.node_stack.push(&(**left));
|
|
continue;
|
|
}
|
|
}
|
|
}
|
|
else {
|
|
return Option::None;
|
|
}
|
|
}
|
|
}
|
|
} |