rolldown_utils 1.1.0

use std::borrow::Cow;
use std::sync::LazyLock;

use memchr::memmem::Finder;
use rolldown_error::{BuildDiagnostic, InvalidOptionType, SingleBuildResult};
use rustc_hash::FxHashMap;

use crate::indexmap::FxIndexSet;

const HASH_PLACEHOLDER_LEFT: &str = "!~{";
const HASH_PLACEHOLDER_RIGHT: &str = "}~";
const HASH_PLACEHOLDER_OVERHEAD: usize = HASH_PLACEHOLDER_LEFT.len() + HASH_PLACEHOLDER_RIGHT.len();

const MAX_HASH_SIZE: usize = 21;
const DEFAULT_HASH_SIZE: usize = 8;

pub static HASH_PLACEHOLDER_LEFT_FINDER: LazyLock<Finder<'static>> =
  LazyLock::new(|| Finder::new(HASH_PLACEHOLDER_LEFT));

/// Checks if a string is a hash placeholder with the pattern "!~{...}~"
/// where ... is 1-17 alphanumeric characters or _ or $
fn is_hash_placeholder(s: &str) -> bool {
  // Check if the string starts with the left placeholder and ends with the right placeholder
  if !s.starts_with(HASH_PLACEHOLDER_LEFT) || !s.ends_with(HASH_PLACEHOLDER_RIGHT) {
    return false;
  }

  // Extract the content between the placeholders
  let content = &s[HASH_PLACEHOLDER_LEFT.len()..s.len() - HASH_PLACEHOLDER_RIGHT.len()];

  // Content must be 1-17 characters long
  if content.is_empty() || content.len() > 17 {
    return false;
  }

  // All characters must be alphanumeric or _ or $
  content.bytes().all(|c| c.is_ascii_alphanumeric() || c == b'_' || c == b'$')
}

/// A lazy iterator over hash placeholders in a string.
pub struct HashPlaceholderIter<'a> {
  s: &'a str,
  finder: &'a Finder<'static>,
  start: usize,
}

impl<'a> Iterator for HashPlaceholderIter<'a> {
  type Item = (usize, usize, &'a str);

  fn next(&mut self) -> Option<Self::Item> {
    loop {
      let left_pos = self.finder.find(&self.s.as_bytes()[self.start..])?;
      let left_pos = self.start + left_pos;
      // Bound the search for `}~` to the maximum possible placeholder length.
      // Use byte-level slicing to avoid panics when `search_end` falls inside a
      // multi-byte UTF-8 character (e.g. Chinese/Japanese characters in the source).
      let search_end = (left_pos + MAX_HASH_SIZE + HASH_PLACEHOLDER_OVERHEAD).min(self.s.len());
      if let Some(right_pos) = memchr::memmem::find(
        &self.s.as_bytes()[left_pos..search_end],
        HASH_PLACEHOLDER_RIGHT.as_bytes(),
      ) {
        let right_pos = left_pos + right_pos + HASH_PLACEHOLDER_RIGHT.len();
        let placeholder = &self.s[left_pos..right_pos];
        self.start = right_pos;
        if is_hash_placeholder(placeholder) {
          return Some((left_pos, right_pos, placeholder));
        }
      } else {
        // No `}~` found within bound; skip past `!~{` and continue
        self.start = left_pos + HASH_PLACEHOLDER_LEFT.len();
      }
    }
  }
}

/// Finds all hash placeholders in a string and returns a lazy iterator over their positions and values.
pub fn find_hash_placeholders<'a>(
  s: &'a str,
  finder: &'a Finder<'static>,
) -> HashPlaceholderIter<'a> {
  HashPlaceholderIter { s, finder, start: 0 }
}

const BASE: u32 = 64;
const CHARS: &[u8; BASE as usize] =
  b"0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ_$";

pub fn to_base64(mut value: u32) -> String {
  let mut buffer = Vec::with_capacity(16);

  loop {
    let current_digit = value % BASE;
    buffer.push(CHARS[current_digit as usize]);
    value /= BASE;

    if value == 0 {
      break;
    }
  }
  buffer.reverse();
  // SAFETY: `buffer` is base64 characters, it is valid utf8 characters
  unsafe { String::from_utf8_unchecked(buffer) }
}

#[derive(Debug, Default)]
pub struct HashPlaceholderGenerator {
  // This is used to generate hash placeholder. Something like `~!{001}~`.
  next_index: u32,
}

impl HashPlaceholderGenerator {
  // Refer to https://github.com/rollup/rollup/blob/1f2d579ccd4b39f223fed14ac7d031a6c848cd80/src/utils/hashPlaceholders.ts#L16-L17
  pub fn generate(&mut self, len: Option<usize>, pattern_name: &str) -> SingleBuildResult<String> {
    let len = len.unwrap_or(DEFAULT_HASH_SIZE);

    if len > MAX_HASH_SIZE {
      return Err(BuildDiagnostic::invalid_option(InvalidOptionType::HashLengthTooLong {
        pattern_name: pattern_name.to_string(),
        received: len,
        max: MAX_HASH_SIZE,
      }));
    }

    let index_in_base64 = to_base64(self.next_index);
    let placeholder_size = index_in_base64.len() + HASH_PLACEHOLDER_OVERHEAD;

    let mut placeholder =
      String::with_capacity(len + HASH_PLACEHOLDER_LEFT.len() + HASH_PLACEHOLDER_RIGHT.len());

    // The placeholder format is `!~{index}~`, requiring at least `HASH_PLACEHOLDER_OVERHEAD + index_in_base64.len()` characters.
    // For example, with index 0 the placeholder is `!~{0}~` (6 chars), with index 100_000_000 it's `!~{5Zu40}~` (10 chars).
    if placeholder_size > len {
      return Err(BuildDiagnostic::invalid_option(InvalidOptionType::HashLengthTooShort {
        pattern_name: pattern_name.to_string(),
        received: len,
        min: placeholder_size,
        chunk_count: self.next_index + 1,
      }));
    }

    placeholder.push_str(HASH_PLACEHOLDER_LEFT);
    placeholder.extend(std::iter::repeat_n('0', len - placeholder_size));
    placeholder.push_str(&index_in_base64);
    placeholder.push_str(HASH_PLACEHOLDER_RIGHT);

    self.next_index += 1;

    Ok(placeholder)
  }
}

/// This function would replace the facade hash placeholder in the given input
/// ```js
/// import { foo } from "foo.!~{000}~.js";
/// ```
/// to
/// ```js
/// import { foo } from "foo.xx__hash.js";
/// ```
#[expect(clippy::implicit_hasher)]
pub fn replace_placeholder_with_hash<'a>(
  source: &'a str,
  final_hashes_by_placeholder: &FxHashMap<String, &'a str>,
  finder: &'a Finder<'static>,
) -> Cow<'a, str> {
  // Check for placeholders directly
  let mut placeholders = find_hash_placeholders(source, finder).peekable();
  if placeholders.peek().is_none() {
    return Cow::Borrowed(source);
  }

  // Create a new string with replacements
  let mut result = String::with_capacity(source.len());
  let mut last_end = 0;

  for (start, end, placeholder) in placeholders {
    // Add the text before this placeholder
    result.push_str(&source[last_end..start]);

    // Add the replacement or the original placeholder if not found
    let replacement = final_hashes_by_placeholder.get(placeholder).unwrap_or(&placeholder);
    result.push_str(replacement);

    last_end = end;
  }

  // Add any remaining text
  if last_end < source.len() {
    result.push_str(&source[last_end..]);
  }

  Cow::Owned(result)
}

pub fn extract_hash_placeholders<'a>(
  source: &'a str,
  finder: &'a Finder<'static>,
) -> FxIndexSet<&'a str> {
  find_hash_placeholders(source, finder).map(|(_, _, placeholder)| placeholder).collect()
}

const NORMALIZED_PLACEHOLDER_INNER: [u8; MAX_HASH_SIZE] = [b'0'; MAX_HASH_SIZE];

/// Walks `source` and feeds it through `visit` byte-slice by byte-slice, replacing each hash
/// placeholder accepted by `is_known_placeholder` with a zero-filled placeholder of the same
/// shape (`!~{000...}~`). Lets a caller stream a content-stable representation into a hasher
/// (or any other sink) without materializing the normalized string in memory — important for
/// chunk content, which can be megabytes.
///
/// Placeholders not recognized by `is_known_placeholder` (typically user source code that just
/// happens to contain a syntactically valid placeholder literal) are emitted verbatim, so a
/// change in their bytes still flows into the hash. Matches Rollup's
/// `replacePlaceholdersWithDefaultAndGetContainedPlaceholders`.
pub fn visit_with_placeholders_defaulted<F, P>(
  source: &str,
  finder: &Finder<'static>,
  is_known_placeholder: P,
  mut visit: F,
) where
  F: FnMut(&[u8]),
  P: Fn(&str) -> bool,
{
  let bytes = source.as_bytes();
  let mut last_end = 0;
  for (start, end, placeholder) in find_hash_placeholders(source, finder) {
    if !is_known_placeholder(placeholder) {
      continue;
    }
    visit(&bytes[last_end..start]);
    visit(HASH_PLACEHOLDER_LEFT.as_bytes());
    visit(&NORMALIZED_PLACEHOLDER_INNER[..placeholder.len() - HASH_PLACEHOLDER_OVERHEAD]);
    visit(HASH_PLACEHOLDER_RIGHT.as_bytes());
    last_end = end;
  }
  visit(&bytes[last_end..]);
}

#[test]
fn test_facade_hash_generator() {
  let mut r#gen = HashPlaceholderGenerator::default();
  assert_eq!(r#gen.generate(None, "").unwrap(), "!~{000}~");
  assert_eq!(r#gen.generate(None, "").unwrap(), "!~{001}~");
}

#[test]
fn test_to_base64() {
  assert_eq!(to_base64(0), "0");
  assert_eq!(to_base64(1), "1");
  assert_eq!(to_base64(10), "a");
  assert_eq!(to_base64(64), "10");
  assert_eq!(to_base64(65), "11");
  assert_eq!(to_base64(128), "20");
  assert_eq!(to_base64(100_000_000), "5Zu40");
}

#[test]
fn test_is_hash_placeholder() {
  assert!(is_hash_placeholder("!~{000}~"));
  assert!(is_hash_placeholder("!~{abc123}~"));
  assert!(is_hash_placeholder("!~{_$ABC123}~"));
  assert!(is_hash_placeholder("!~{12345678901234567}~")); // 17 chars

  assert!(!is_hash_placeholder("!~{}~")); // Empty content
  assert!(!is_hash_placeholder("!~{123456789012345678}~")); // 18 chars (too long)
  assert!(!is_hash_placeholder("!~{abc-123}~")); // Invalid char
  assert!(!is_hash_placeholder("{000}~")); // Missing left
  assert!(!is_hash_placeholder("!~{000}")); // Missing right
  assert!(!is_hash_placeholder("!~000}~")); // Missing {
}

#[test]
fn test_find_hash_placeholders() {
  let s = "prefix!~{000}~middle!~{abc}~suffix";
  let placeholders: Vec<_> = find_hash_placeholders(s, &HASH_PLACEHOLDER_LEFT_FINDER).collect();
  assert_eq!(placeholders.len(), 2);
  assert_eq!(placeholders[0], (6, 14, "!~{000}~"));
  assert_eq!(placeholders[1], (20, 28, "!~{abc}~"));

  let s = "no placeholders here";
  let placeholders: Vec<_> = find_hash_placeholders(s, &HASH_PLACEHOLDER_LEFT_FINDER).collect();
  assert_eq!(placeholders.len(), 0);

  let s = "!~{000}~!~{001}~";
  let placeholders: Vec<_> = find_hash_placeholders(s, &HASH_PLACEHOLDER_LEFT_FINDER).collect();
  assert_eq!(placeholders.len(), 2);
  assert_eq!(placeholders[0], (0, 8, "!~{000}~"));
  assert_eq!(placeholders[1], (8, 16, "!~{001}~"));
}

#[test]
fn test_find_hash_placeholders_multi_byte_chars() {
  // Multi-byte UTF-8 chars near placeholders must not cause a panic.
  let s = "import{C as e}from\"./vue.runtime.esm-bundler-!~{001}~.js\";// 中文级别文字";
  let placeholders: Vec<_> = find_hash_placeholders(s, &HASH_PLACEHOLDER_LEFT_FINDER).collect();
  assert_eq!(placeholders.len(), 1);
  assert_eq!(placeholders[0].2, "!~{001}~");
}

#[test]
fn test_visit_with_placeholders_defaulted() {
  use rustc_hash::FxHashSet;
  fn collect(source: &str, known: &FxHashSet<&str>) -> Vec<u8> {
    let mut buf = Vec::new();
    visit_with_placeholders_defaulted(
      source,
      &HASH_PLACEHOLDER_LEFT_FINDER,
      |p| known.contains(p),
      |bytes| buf.extend_from_slice(bytes),
    );
    buf
  }

  // All placeholders are known and get normalized to their zero-filled shape.
  let all_known: FxHashSet<&str> = ["!~{000}~", "!~{abc12}~"].into_iter().collect();
  assert_eq!(
    collect("prefix!~{000}~middle!~{abc12}~suffix", &all_known),
    b"prefix!~{000}~middle!~{00000}~suffix",
  );

  // An unknown placeholder (e.g. a literal in user source) is emitted verbatim so the hash
  // still reflects changes to its bytes. Only the known one is normalized.
  let only_known: FxHashSet<&str> = std::iter::once("!~{000}~").collect();
  assert_eq!(
    collect("prefix!~{000}~middle!~{user}~suffix", &only_known),
    b"prefix!~{000}~middle!~{user}~suffix",
  );

  let empty: FxHashSet<&str> = FxHashSet::default();
  assert_eq!(collect("no placeholders here", &empty), b"no placeholders here");
  assert_eq!(collect("", &empty), b"");
}