//! Some utilities helpful when parsing and compiling markdown.
use crate::constant::{CHARACTER_REFERENCE_NAMES, CHARACTER_REFERENCE_VALUES};
use crate::tokenizer::{Code, Event, EventType};
/// Encode dangerous html characters.
///
/// This ensures that certain characters which have special meaning in HTML are
/// dealt with.
/// Technically, we can skip `>` and `"` in many cases, but CM includes them.
///
/// This behavior is not explained in prose in `CommonMark` but can be inferred
/// from the input/output test cases.
///
/// ## Examples
///
/// ```rust ignore
/// use micromark::util::encode;
///
/// assert_eq!(encode("I <3 🦀"), "I <3 🦀");
/// ```
///
/// ## References
///
/// * [`micromark-util-encode` in `micromark`](https://github.com/micromark/micromark/tree/main/packages/micromark-util-encode)
pub fn encode(value: &str) -> String {
value
.replace('&', "&")
.replace('"', """)
.replace('<', "<")
.replace('>', ">")
}
/// Make a value safe for injection as a URL.
///
/// This encodes unsafe characters with percent-encoding and skips already
/// encoded sequences (see `normalize_uri` below).
/// Further unsafe characters are encoded as character references (see
/// `encode`).
///
/// Then, a vec of (lowercase) allowed protocols can be given, in which case
/// the URL is sanitized.
///
/// For example, `Some(vec!["http", "https", "irc", "ircs", "mailto", "xmpp"])`
/// can be used for `a[href]`, or `Some(vec!["http", "https"])` for `img[src]`.
/// If the URL includes an unknown protocol (one not matched by `protocol`, such
/// as a dangerous example, `javascript:`), the value is ignored.
pub fn sanitize_uri(value: &str, protocols: &Option<Vec<&str>>) -> String {
let value = encode(&normalize_uri(value));
if let Some(protocols) = protocols {
let chars: Vec<char> = value.chars().collect();
let mut index = 0;
let mut colon: Option<usize> = None;
while index < chars.len() {
let char = chars[index];
match char {
':' => {
colon = Some(index);
break;
}
'?' | '#' | '/' => break,
_ => {}
}
index += 1;
}
// If there is no protocol, or the first colon is after `?`, `#`, or `/`, it’s relative.
// It is a protocol, it should be allowed.
if let Some(colon) = colon {
let protocol = chars[0..colon].iter().collect::<String>().to_lowercase();
if !protocols.contains(&protocol.as_str()) {
return "".to_string();
}
}
}
value
}
/// Normalize a URL (such as used in definitions).
///
/// Encode unsafe characters with percent-encoding, skipping already encoded
/// sequences.
fn normalize_uri(value: &str) -> String {
let chars: Vec<char> = value.chars().collect();
let mut result: Vec<String> = vec![];
let mut index = 0;
let mut start = 0;
let mut buff = [0; 4];
while index < chars.len() {
let char = chars[index];
// A correct percent encoded value.
if char == '%'
&& index + 2 < chars.len()
&& chars[index + 1].is_ascii_alphanumeric()
&& chars[index + 2].is_ascii_alphanumeric()
{
index += 3;
continue;
}
// Note: Rust already takes care of lone astral surrogates.
// Non-ascii or not allowed ascii.
if char >= '\u{0080}'
|| !matches!(char, '!' | '#' | '$' | '&'..=';' | '=' | '?'..='Z' | '_' | 'a'..='z' | '~')
{
result.push(chars[start..index].iter().collect::<String>());
char.encode_utf8(&mut buff);
result.push(
buff[0..char.len_utf8()]
.iter()
.map(|&byte| format!("%{:X}", byte))
.collect::<String>(),
);
start = index + 1;
}
index += 1;
}
result.push(chars[start..].iter().collect::<String>());
result.join("")
}
/// Decode numeric character references.
///
/// Turn the number (in string form as either hexadecimal or decimal) coming
/// from a numeric character reference into a character.
/// Whether the base of the string form is `10` (decimal) or `16` (hexadecimal)
/// must be passed as the `radix` parameter.
///
/// This returns the `char` associated with that number or a replacement
/// character for C0 control characters (except for ASCII whitespace), C1
/// control characters, lone surrogates, noncharacters, and out of range
/// characters.
///
/// ## Examples
///
/// ```rust ignore
/// use micromark::util::decode_numeric_character_reference;
///
/// assert_eq!(decode_numeric_character_reference("123", 10), '{');
/// assert_eq!(decode_numeric_character_reference("9", 16), '\t');
/// assert_eq!(decode_numeric_character_reference("0", 10), '�'); // Not allowed.
/// ```
///
/// ## Panics
///
/// This function panics if a invalid string or an out of bounds valid string
/// is given.
/// It is expected that figuring out whether a number is allowed is handled in
/// the parser.
/// When `micromark` is used, this function never panics.
///
/// ## References
///
/// * [`micromark-util-decode-numeric-character-reference` in `micromark`](https://github.com/micromark/micromark/tree/main/packages/micromark-util-decode-numeric-character-reference)
/// * [*§ 2.5 Entity and numeric character references* in `CommonMark`](https://spec.commonmark.org/0.30/#entity-and-numeric-character-references)
pub fn decode_numeric_character_reference(value: &str, radix: u32) -> char {
let code = u32::from_str_radix(value, radix).expect("expected `value` to be an int");
if
// C0 except for HT, LF, FF, CR, space
code < 0x09 ||
code == 0x0B ||
(code > 0x0D && code < 0x20) ||
// Control character (DEL) of the basic block and C1 controls.
(code > 0x7E && code < 0xA0) ||
// Lone high surrogates and low surrogates.
(code > 0xd7ff && code < 0xe000) ||
// Noncharacters.
(code > 0xfdcf && code < 0xfdf0) ||
((code & 0xffff) == 0xffff) ||
((code & 0xffff) == 0xfffe) ||
// Out of range
code > 0x0010_ffff
{
'�'
} else {
char::from_u32(code).expect("expected valid `code`")
}
}
/// Decode named character references.
///
/// Turn the name coming from a named character reference (without the `&` or
/// `;`) into a string.
/// This looks the given string up in [`CHARACTER_REFERENCE_NAMES`][] and then
/// takes the corresponding value from [`CHARACTER_REFERENCE_VALUES`][].
///
/// The result is `String` instead of `char` because named character references
/// can expand into multiple characters.
///
/// ## Examples
///
/// ```rust ignore
/// use micromark::util::decode_named_character_reference;
///
/// assert_eq!(decode_named_character_reference("amp"), "&");
/// assert_eq!(decode_named_character_reference("AElig"), "Æ");
/// assert_eq!(decode_named_character_reference("aelig"), "æ");
/// ```
///
/// ## Panics
///
/// This function panics if a name not in [`CHARACTER_REFERENCE_NAMES`][] is
/// given.
/// It is expected that figuring out whether a name is allowed is handled in
/// the parser.
/// When `micromark` is used, this function never panics.
///
/// ## References
///
/// * [`wooorm/decode-named-character-reference`](https://github.com/wooorm/decode-named-character-reference)
/// * [*§ 2.5 Entity and numeric character references* in `CommonMark`](https://spec.commonmark.org/0.30/#entity-and-numeric-character-references)
pub fn decode_named_character_reference(value: &str) -> String {
let position = CHARACTER_REFERENCE_NAMES.iter().position(|&x| x == value);
if let Some(index) = position {
CHARACTER_REFERENCE_VALUES[index].to_string()
} else {
unreachable!("expected valid `name`")
}
}
/// A struct representing the span of an opening and closing event of a token.
#[derive(Debug)]
pub struct Span {
// To do: probably needed in the future.
// start: Point,
/// Absolute offset (and `index` in `codes`) of where this span starts.
pub start_index: usize,
// To do: probably needed in the future.
// end: Point,
/// Absolute offset (and `index` in `codes`) of where this span ends.
pub end_index: usize,
// To do: probably needed in the future.
// token_type: TokenType,
}
/// Get a span from an event.
///
/// Get the span of an `exit` event, by looking backwards through the events to
/// find the corresponding `enter` event.
/// This assumes that tokens with the same are not nested.
///
/// ## Panics
///
/// This function panics if an enter event is given.
/// When `micromark` is used, this function never panics.
pub fn get_span(events: &[Event], index: usize) -> Span {
let exit = &events[index];
// let end = exit.point.clone();
let end_index = exit.index;
let token_type = exit.token_type.clone();
// To do: support `enter` events if needed and walk forwards?
assert_eq!(
exit.event_type,
EventType::Exit,
"expected `get_span` to be called on `exit` event"
);
let mut enter_index = index - 1;
loop {
let enter = &events[enter_index];
if enter.event_type == EventType::Enter && enter.token_type == token_type {
return Span {
// start: enter.point.clone(),
start_index: enter.index,
// end,
end_index,
// token_type,
};
}
enter_index -= 1;
}
}
/// Serialize a span, optionally expanding tabs.
pub fn slice_serialize(codes: &[Code], span: &Span, expand_tabs: bool) -> String {
serialize_chunks(slice_codes(codes, span), expand_tabs)
}
/// Get a slice of codes from a span.
pub fn slice_codes<'a>(codes: &'a [Code], span: &Span) -> &'a [Code] {
&codes[span.start_index..span.end_index]
}
/// Serialize a slice of codes, optionally expanding tabs.
pub fn serialize_chunks(codes: &[Code], expand_tabs: bool) -> String {
let mut at_tab = false;
let mut index = 0;
let mut value: Vec<char> = vec![];
while index < codes.len() {
let code = codes[index];
let mut at_tab_next = false;
match code {
Code::CarriageReturnLineFeed => {
value.push('\r');
value.push('\n');
}
Code::Char(char) if char == '\n' || char == '\r' => {
value.push(char);
}
Code::Char(char) if char == '\t' => {
at_tab_next = true;
value.push(if expand_tabs { ' ' } else { char });
}
Code::VirtualSpace => {
if !expand_tabs && at_tab {
index += 1;
continue;
}
value.push(' ');
}
Code::Char(char) => {
value.push(char);
}
Code::None => {
unreachable!("unexpected EOF code in codes");
}
}
at_tab = at_tab_next;
index += 1;
}
value.into_iter().collect()
}