capnprotols 0.1.3

//! Compute the next valid `@<n>` ordinal for a field declaration in a struct.
//!
//! Cap'n Proto requires field ordinals to be contiguous starting from zero, scoped to the
//! enclosing struct. Groups and unions share the parent struct's ID space; nested structs
//! get their own. We work entirely from the buffer text (with comments/strings stripped)
//! so this stays usable on broken-but-parseable buffers where the compiler may not have
//! produced a CGR.
//!
//! The hot path is short and bounded by the size of the enclosing struct, so a textual
//! scan is plenty.

use crate::aliases::strip_comments;

/// Block-opening keyword found by walking back from the cursor.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum BlockKind {
  Struct,
  Enum,
  Group,
  Union,
  Other,
}

#[derive(Debug)]
struct OpenBrace {
  /// Byte index of the `{` itself.
  brace_byte: usize,
  kind:       BlockKind,
}

/// Suggest valid ordinal values at `cursor` in ascending order: first any gaps in the
/// existing assignments, then the next-after-max. Returns empty if the cursor isn't
/// inside a struct or enum. Cap'n Proto requires contiguous ordinals starting from 0, so
/// gaps (e.g. from a deleted field) are legal slots to fill; when the user is typing @
/// they're often doing exactly that.
pub fn suggest_ordinals_at(text: &str, cursor: usize) -> Vec<u32> {
  let cleaned = strip_for_scan(text);
  let Some(outer) = enclosing_struct_or_enum(&cleaned, cursor) else {
    return Vec::new();
  };
  let close =
    matching_close(&cleaned, outer.brace_byte).unwrap_or(cleaned.len());
  let body = &cleaned[outer.brace_byte + 1..close];
  let used = collect_ordinals(body, outer.kind);
  let mut out: Vec<u32> = Vec::new();
  if let Some(m) = used.iter().copied().max() {
    for n in 0..=m {
      if !used.contains(&n) {
        out.push(n);
      }
    }
    out.push(m + 1);
  } else {
    out.push(0);
  }
  out
}

/// Strip both `# ...` line comments and `"..."` string literals — replacing their bytes
/// with spaces so byte offsets are preserved.
fn strip_for_scan(src: &str) -> String {
  let mut s = strip_comments(src).into_bytes();
  let mut i = 0;
  while i < s.len() {
    if s[i] == b'"' {
      s[i] = b' ';
      i += 1;
      while i < s.len() && s[i] != b'"' && s[i] != b'\n' {
        s[i] = b' ';
        i += 1;
      }
      if i < s.len() && s[i] == b'"' {
        s[i] = b' ';
        i += 1;
      }
    } else {
      i += 1;
    }
  }
  String::from_utf8(s).unwrap_or_default()
}

/// Walk forward, building a stack of open braces with the kind of block they open. Stop
/// when we pass `cursor`. Returns the topmost (innermost) struct-or-enum brace still open.
fn enclosing_struct_or_enum(text: &str, cursor: usize) -> Option<OpenBrace> {
  let bytes = text.as_bytes();
  let cursor = cursor.min(bytes.len());
  let mut stack: Vec<OpenBrace> = Vec::new();
  let mut i = 0;
  while i < cursor {
    let b = bytes[i];
    if b == b'{' {
      stack.push(OpenBrace {
        brace_byte: i,
        kind:       classify_block(text, i),
      });
    } else if b == b'}' {
      stack.pop();
    }
    i += 1;
  }
  stack
    .into_iter()
    .rev()
    .find(|f| matches!(f.kind, BlockKind::Struct | BlockKind::Enum))
}

/// Classify what kind of block an opening `{` belongs to by scanning the tokens that
/// precede it on the same statement (back to the previous `{`, `}`, or `;`).
fn classify_block(text: &str, brace_byte: usize) -> BlockKind {
  let bytes = text.as_bytes();
  let mut i = brace_byte;
  // Stop when we hit a statement terminator or another brace.
  while i > 0 {
    i -= 1;
    match bytes[i] {
      b';' | b'{' | b'}' => {
        i += 1;
        break;
      }
      _ => {}
    }
  }
  let segment = &text[i..brace_byte];
  // `union {` or `union $Foo.bar {` (annotations between `union` and `{`)
  // `group { ... }` is unusual — capnp's `:group {` is the common form.
  // `struct Name { ... }` or `struct Name $Foo {` etc.
  let words: Vec<&str> = segment
    .split(|c: char| !(c.is_ascii_alphanumeric() || c == '_'))
    .filter(|w| !w.is_empty())
    .collect();
  // Anonymous union has just `union` on its own; named union doesn't exist. `:group` is
  // a field type whose body is a brace-delimited declaration.
  if words.contains(&"struct") {
    BlockKind::Struct
  } else if words.contains(&"enum") {
    BlockKind::Enum
  } else if words.contains(&"union") {
    BlockKind::Union
  } else if words.contains(&"group") {
    BlockKind::Group
  } else {
    BlockKind::Other
  }
}

/// Match a `{` at `open_byte` to its closing `}`. Returns the byte index of the `}`.
fn matching_close(text: &str, open_byte: usize) -> Option<usize> {
  let bytes = text.as_bytes();
  if bytes.get(open_byte) != Some(&b'{') {
    return None;
  }
  let mut depth: i32 = 0;
  let mut i = open_byte;
  while i < bytes.len() {
    match bytes[i] {
      b'{' => depth += 1,
      b'}' => {
        depth -= 1;
        if depth == 0 {
          return Some(i);
        }
      }
      _ => {}
    }
    i += 1;
  }
  None
}

/// Collect every `@<n>` ordinal in `body` that belongs to the enclosing block's ID space.
/// For a struct: groups/unions share the parent's ID space (so we descend through them),
/// but nested `struct { ... }` and `enum { ... }` each open their own space and are skipped.
/// For an enum: every nested block opens a different space, so all of them are skipped.
fn collect_ordinals(body: &str, outer: BlockKind) -> Vec<u32> {
  let bytes = body.as_bytes();
  let mut out: Vec<u32> = Vec::new();
  let mut i = 0;
  while i < bytes.len() {
    if bytes[i] == b'{' {
      let kind = classify_block(body, i);
      let crosses_scope = match outer {
        BlockKind::Struct => {
          matches!(kind, BlockKind::Struct | BlockKind::Enum)
        }
        BlockKind::Enum => true,
        _ => false,
      };
      if crosses_scope {
        if let Some(close) = matching_close(body, i) {
          i = close + 1;
          continue;
        } else {
          break;
        }
      }
    }
    if bytes[i] == b'@' {
      let mut j = i + 1;
      while j < bytes.len() && bytes[j].is_ascii_digit() {
        j += 1;
      }
      if j > i + 1 {
        if let Ok(n) = body[i + 1..j].parse::<u32>() {
          out.push(n);
        }
        i = j;
        continue;
      }
    }
    i += 1;
  }
  out
}

#[cfg(test)]
mod tests {
  use super::*;

  fn ord_at(src: &str) -> Option<u32> {
    let cursor = src
      .find('|')
      .expect("test source needs a `|` cursor marker");
    let stripped = src.replace('|', "");
    suggest_ordinals_at(&stripped, cursor).first().copied()
  }

  fn ords_at(src: &str) -> Vec<u32> {
    let cursor = src
      .find('|')
      .expect("test source needs a `|` cursor marker");
    let stripped = src.replace('|', "");
    suggest_ordinals_at(&stripped, cursor)
  }

  #[test]
  fn empty_struct() {
    assert_eq!(ord_at("struct A { foo @|"), Some(0));
  }

  #[test]
  fn after_existing_fields() {
    let src = "struct A {\n  foo @0 :Text;\n  bar @1 :UInt8;\n  baz @|";
    assert_eq!(ord_at(src), Some(2));
  }

  #[test]
  fn group_shares_parent_id_space() {
    let src = "struct A {\n  foo @0 :Text;\n  inner :group {\n    a @1 :UInt8;\n    b @|";
    assert_eq!(ord_at(src), Some(2));
  }

  #[test]
  fn union_shares_parent_id_space() {
    let src = "struct A {\n  foo @0 :Text;\n  body :union {\n    a @1 :UInt8;\n    b @|";
    assert_eq!(ord_at(src), Some(2));
  }

  #[test]
  fn nested_struct_has_own_id_space() {
    let src = "struct Outer {\n  foo @0 :Text;\n  bar @5 :Int32;\n  struct Inner {\n    a @0 :Bool;\n    b @|";
    // Cursor is inside Inner -> next is 1.
    assert_eq!(ord_at(src), Some(1));
  }

  #[test]
  fn outer_ignores_nested_struct_ordinals() {
    let src = "struct Outer {\n  struct Inner { a @9 :Bool; }\n  foo @0 :Text;\n  bar @|";
    assert_eq!(ord_at(src), Some(1));
  }

  #[test]
  fn outside_struct_returns_none() {
    assert_eq!(ord_at("@|"), None);
  }

  #[test]
  fn gaps_offered_before_next() {
    // existing: 0,2,3,5 — gaps are 1 and 4, plus 6 at end.
    let src = "struct S {\n  a @0 :Text;\n  c @2 :Text;\n  d @3 :Text;\n  e @5 :Text;\n  f @|";
    assert_eq!(ords_at(src), vec![1, 4, 6]);
  }

  #[test]
  fn enum_ordinals() {
    assert_eq!(ord_at("enum Side { buy @0; sell @|"), Some(1));
  }

  #[test]
  fn empty_enum() {
    assert_eq!(ord_at("enum E { first @|"), Some(0));
  }

  #[test]
  fn enum_inside_struct_has_own_space() {
    let src = "struct S {\n  foo @0 :Text;\n  bar @1 :UInt8;\n  enum Kind {\n    a @0;\n    b @|";
    assert_eq!(ord_at(src), Some(1));
  }

  #[test]
  fn struct_ignores_nested_enum_ordinals() {
    let src = "struct S {\n  enum K { a @0; b @1; }\n  foo @0 :Text;\n  bar @|";
    assert_eq!(ord_at(src), Some(1));
  }

  #[test]
  fn ignores_at_in_string() {
    let src = "struct A { foo @0 :Text = \"hello @99 world\"; bar @|";
    assert_eq!(ord_at(src), Some(1));
  }
}