agent-codinglanguage-mapper 1.0.0__py3-none-any.whl

agent_codinglanguage_mapper/adapters/delphi_frontend/lsp_server.py

@@ -0,0 +1,1832 @@
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Iterable, Optional
+from urllib.parse import unquote, urlparse
+import os
+import re
+
+from .semantic import (
+    ReferenceKind,
+    Scope,
+    ScopeKind,
+    SourceRange,
+    Symbol,
+    SymbolIndex,
+    SymbolKind,
+    SymbolReference,
+    TypeRef,
+    UnknownTypeRef,
+    Visibility,
+)
+from .semantic_builder import SemanticBuilder, SemanticModel
+from .source_reader import read_source_text
+from .workspace import WorkspaceSemanticResult, build_workspace_semantics
+from .project_discovery import discover_delphi_project
+from ..._version import __version__
+
+
+@dataclass
+class DocumentSnapshot:
+    uri: str
+    file_name: str
+    text: str
+    semantic: Optional[SemanticModel] = None
+
+
+@dataclass
+class FileSnapshot:
+    path: str
+    text: str
+    mtime: float
+
+
+@dataclass
+class WorkspaceConfig:
+    roots: list[str] = field(default_factory=list)
+    include_paths: list[str] = field(default_factory=list)
+    search_paths: list[str] = field(default_factory=list)
+    defines: list[str] = field(default_factory=list)
+    extensions: tuple[str, ...] = ('.pas', '.dpr', '.dpk', '.inc')
+    eager_index: bool = False
+    auto_discover_paths: bool = True
+    discovered_include_paths: list[str] = field(default_factory=list)
+    discovered_search_paths: list[str] = field(default_factory=list)
+    discovered_defines: list[str] = field(default_factory=list)
+
+
+@dataclass
+class LspWorkspaceState:
+    config: WorkspaceConfig = field(default_factory=WorkspaceConfig)
+    documents: dict[str, DocumentSnapshot] = field(default_factory=dict)
+    workspace: Optional[WorkspaceSemanticResult] = None
+    workspace_symbol_index: Optional[WorkspaceSemanticResult] = None
+    workspace_symbol_query_cache: dict[str, WorkspaceSemanticResult] = field(default_factory=dict)
+    workspace_files: set[str] = field(default_factory=set)
+    file_cache: dict[str, FileSnapshot] = field(default_factory=dict)
+
+    def configure(self, config: WorkspaceConfig) -> None:
+        self.config = self._with_discovered_config(config)
+        self.workspace_files = set()
+        self.file_cache = {}
+        self.workspace = None
+        self.workspace_symbol_index = None
+        self.workspace_symbol_query_cache = {}
+        if config.eager_index:
+            self.index_workspace()
+
+    def _with_discovered_config(self, config: WorkspaceConfig) -> WorkspaceConfig:
+        if not config.auto_discover_paths or not config.roots:
+            return config
+        include_paths = list(config.include_paths)
+        search_paths = list(config.search_paths)
+        defines = list(config.defines)
+        discovered_include_paths: list[str] = []
+        discovered_search_paths: list[str] = []
+        discovered_defines: list[str] = []
+
+        def add_path(target: list[str], discovered: list[str], value: str) -> None:
+            normalized = str(Path(value).expanduser().resolve())
+            if normalized not in target:
+                target.append(normalized)
+            if normalized not in discovered:
+                discovered.append(normalized)
+
+        def add_define(value: str) -> None:
+            if value not in defines:
+                defines.append(value)
+            if value not in discovered_defines:
+                discovered_defines.append(value)
+
+        for root in config.roots:
+            discovery = discover_delphi_project(root, include_paths=config.include_paths, search_paths=config.search_paths, defines=config.defines)
+            for path in discovery.include_paths:
+                add_path(include_paths, discovered_include_paths, path)
+            for path in discovery.search_paths:
+                add_path(search_paths, discovered_search_paths, path)
+            for define in discovery.defines:
+                add_define(define)
+        return WorkspaceConfig(
+            roots=list(config.roots),
+            include_paths=include_paths,
+            search_paths=search_paths,
+            defines=defines,
+            extensions=config.extensions,
+            eager_index=config.eager_index,
+            auto_discover_paths=config.auto_discover_paths,
+            discovered_include_paths=discovered_include_paths,
+            discovered_search_paths=discovered_search_paths,
+            discovered_defines=discovered_defines,
+        )
+
+    def index_workspace(self) -> None:
+        self.workspace_files = set(self._scan_workspace_files())
+        self._refresh_file_cache()
+        self._rebuild()
+
+    def ensure_workspace_indexed(self) -> None:
+        if self.workspace is None and self.config.roots:
+            self.index_workspace()
+
+    def ensure_workspace_symbol_indexed(self) -> None:
+        if self.workspace_symbol_index is not None:
+            return
+        if not self.workspace_files and self.config.roots:
+            self.workspace_files = set(self._scan_workspace_files())
+        self._refresh_file_cache()
+        sources = {path: snapshot.text for path, snapshot in self.file_cache.items()}
+        for doc in self.documents.values():
+            sources[doc.file_name] = doc.text
+        self.workspace_symbol_index = self._outline_workspace_semantics(sources)
+
+    def workspace_symbols_for_query(self, query: str) -> Optional[WorkspaceSemanticResult]:
+        normalized_query = query.strip().casefold()
+        if not normalized_query:
+            self.ensure_workspace_symbol_indexed()
+            return self.workspace_symbol_index
+        if self.workspace_symbol_index is not None:
+            return self.workspace_symbol_index
+        cached = self.workspace_symbol_query_cache.get(normalized_query)
+        if cached is not None:
+            return cached
+        if not self.workspace_files and self.config.roots:
+            self.workspace_files = set(self._scan_workspace_files())
+        self._refresh_file_cache()
+        sources = {
+            path: snapshot.text
+            for path, snapshot in self.file_cache.items()
+            if normalized_query in snapshot.text.casefold()
+        }
+        for doc in self.documents.values():
+            if normalized_query in doc.text.casefold():
+                sources[doc.file_name] = doc.text
+        result = self._outline_workspace_semantics(sources)
+        if result is None:
+            result = WorkspaceSemanticResult(models={}, index=SymbolIndex())
+        self.workspace_symbol_query_cache[normalized_query] = result
+        return result
+
+    def update_document(self, uri: str, text: str) -> None:
+        file_name = uri_to_path(uri)
+        self.documents[uri] = DocumentSnapshot(uri=uri, file_name=file_name, text=text)
+        self.workspace_symbol_index = None
+        self.workspace_symbol_query_cache = {}
+        self._rebuild()
+
+    def remove_document(self, uri: str) -> None:
+        if uri in self.documents:
+            self.documents.pop(uri)
+            self.workspace_symbol_index = None
+            self.workspace_symbol_query_cache = {}
+            self._rebuild()
+
+    def _scan_workspace_files(self) -> list[str]:
+        files: list[str] = []
+        for root in self.config.roots:
+            root_path = Path(root)
+            if not root_path.exists():
+                continue
+            for ext in self.config.extensions:
+                files.extend(str(path) for path in root_path.rglob(f'*{ext}'))
+        return files
+
+    def _refresh_file_cache(self) -> None:
+        to_remove = [path for path in self.file_cache if path not in self.workspace_files]
+        for path in to_remove:
+            self.file_cache.pop(path, None)
+        for path in self.workspace_files:
+            try:
+                mtime = os.path.getmtime(path)
+            except OSError:
+                continue
+            cached = self.file_cache.get(path)
+            if cached is not None and cached.mtime == mtime:
+                continue
+            try:
+                text = read_source_text(Path(path))
+            except (OSError, UnicodeError):
+                continue
+            self.file_cache[path] = FileSnapshot(path=path, text=text, mtime=mtime)
+
+    def _collect_sources(self) -> dict[str, str]:
+        sources = {
+            path: snapshot.text
+            for path, snapshot in self.file_cache.items()
+        }
+        for doc in self.documents.values():
+            sources[doc.file_name] = doc.text
+        return sources
+
+    def _rebuild(self) -> None:
+        sources = self._collect_sources()
+        self.workspace_symbol_index = None
+        self.workspace_symbol_query_cache = {}
+        if not sources:
+            self.workspace = None
+            return
+        result = self._outline_workspace_semantics(sources)
+        self.workspace = result
+        for doc in self.documents.values():
+            doc.semantic = result.models.get(doc.file_name) if result is not None else None
+
+    def semantic_for_uri(self, uri: str) -> Optional[SemanticModel]:
+        doc = self.documents.get(uri)
+        if doc is not None:
+            return doc.semantic
+        file_name = uri_to_path(uri)
+        model = self.model_for_path(file_name)
+        if model is not None:
+            return model
+        if '://' in file_name:
+            return None
+        path = Path(file_name)
+        if not path.exists():
+            return None
+        try:
+            text = read_source_text(path)
+        except (OSError, UnicodeError):
+            return None
+        return build_outline_semantic_model(text, file_name)
+
+    def full_semantic_for_uri(self, uri: str) -> Optional[SemanticModel]:
+        file_name = uri_to_path(uri)
+        text = self.text_for_uri(uri)
+        if text is None:
+            return None
+        result = build_workspace_semantics(
+            {file_name: text},
+            include_paths=self.config.include_paths,
+            defines=self.config.defines,
+        )
+        return result.models.get(file_name)
+
+    def structure_semantic_for_uri(self, uri: str) -> Optional[SemanticModel]:
+        text = self.text_for_uri(uri)
+        if text is None:
+            return None
+        return build_outline_semantic_model(text, uri_to_path(uri))
+
+    def text_for_uri(self, uri: str) -> Optional[str]:
+        doc = self.documents.get(uri)
+        if doc is not None:
+            return doc.text
+        file_name = uri_to_path(uri)
+        cached = self.file_cache.get(file_name)
+        if cached is not None:
+            return cached.text
+        if '://' in file_name:
+            return None
+        path = Path(file_name)
+        if not path.exists():
+            return None
+        try:
+            return read_source_text(path)
+        except (OSError, UnicodeError):
+            return None
+
+    def _outline_workspace_semantics(
+        self,
+        sources: dict[str, str],
+    ) -> Optional[WorkspaceSemanticResult]:
+        if not sources:
+            return None
+        models = {file_name: build_outline_semantic_model(text, file_name) for file_name, text in sources.items()}
+        index = SymbolIndex()
+        for model in models.values():
+            index.register_unit(model.unit_scope.name, model.unit_scope)
+        for model in models.values():
+            model.index = index
+        return WorkspaceSemanticResult(models=models, index=index)
+
+    def model_for_path(self, path: str) -> Optional[SemanticModel]:
+        if self.workspace is None:
+            return None
+        return self.workspace.models.get(path)
+
+    def iter_models(self) -> Iterable[SemanticModel]:
+        if self.workspace is None:
+            return []
+        return self.workspace.models.values()
+
+    def uri_for_file_name(self, file_name: str) -> Optional[str]:
+        for doc in self.documents.values():
+            if doc.file_name == file_name:
+                return doc.uri
+        return None
+
+    def diagnostics_for_uri(self, uri: str):
+        model = self.semantic_for_uri(uri)
+        return diagnostics_for_model(model) if model else []
+
+
+BUILTIN_TYPES = {
+    'string', 'ansistring', 'widestring', 'unicodestring',
+    'char', 'widechar',
+    'integer', 'smallint', 'shortint', 'longint', 'int64', 'uint64',
+    'byte', 'word', 'cardinal', 'nativeint', 'nativeuint',
+    'boolean', 'bytebool', 'wordbool', 'longbool',
+    'single', 'double', 'extended', 'currency', 'real', 'real48',
+    'variant', 'olevariant', 'pointer', 'pchar', 'tobject', 'tclass',
+}
+
+
+def uri_to_path(uri: str) -> str:
+    parsed = urlparse(uri)
+    if parsed.scheme and parsed.scheme != 'file':
+        return uri
+    path = unquote(parsed.path)
+    if path.startswith('/') and len(path) > 3 and path[2] == ':':
+        path = path[1:]
+    return path or uri
+
+
+def outline_source(text: str) -> str:
+    return _blank_compound_statement_bodies(text)
+
+
+def multiline_string_block_end(text: str, start: int) -> int | None:
+    for quote_count in (5, 3):
+        delimiter = "'" * quote_count
+        if not text.startswith(delimiter, start):
+            continue
+        content_start = start + quote_count
+        if text.startswith('\r\n', content_start):
+            content_start += 2
+        elif content_start < len(text) and text[content_start] == '\n':
+            content_start += 1
+        else:
+            continue
+        close = text.find(delimiter, content_start)
+        return len(text) if close < 0 else close + quote_count
+    return None
+
+
+# These are the statement grammar alternatives that close with END.
+_END_TERMINATED_STATEMENTS = frozenset({'begin', 'asm', 'case', 'try'})
+# Inline variable type_spec declarations can place these inside a body.
+_END_TERMINATED_STRUCTURED_TYPES = frozenset(
+    {'class', 'record', 'object', 'interface', 'dispinterface'}
+)
+_CLASS_MEMBER_PREFIX_FOLLOWERS = frozenset(
+    {
+        'const',
+        'constructor',
+        'destructor',
+        'function',
+        'of',
+        'operator',
+        'procedure',
+        'property',
+        'threadvar',
+        'type',
+        'var',
+    }
+)
+_CLASS_TYPE_PREDECESSORS = frozenset({':', '=', 'of', 'packed', 'to', 'type', '^'})
+_CLASS_MEMBER_BOUNDARIES = frozenset(
+    {
+        ';',
+        ')',
+        ']',
+        'automated',
+        'class',
+        'dispinterface',
+        'interface',
+        'object',
+        'private',
+        'protected',
+        'public',
+        'published',
+        'record',
+    }
+)
+_GENERIC_ROUTINE_HEADINGS = frozenset(
+    {'constructor', 'destructor', 'function', 'operator', 'procedure'}
+)
+
+
+def _blank_compound_statement_bodies(text: str) -> str:
+    chars = list(text)
+    end_stack: list[str] = []
+    body_start: int | None = None
+    previous_token: str | None = None
+    previous_token_is_identifier = False
+    paren_depth = 0
+    bracket_depth = 0
+    angle_depth = 0
+    generic_owner_ready = False
+    routine_heading_active = False
+    type_constructors: dict[tuple[int, int], str] = {}
+    i = 0
+    n = len(text)
+    while i < n:
+        ch = text[i]
+        nxt = text[i + 1] if i + 1 < n else ''
+
+        if end_stack and (
+            (ch == '{' and nxt == '$')
+            or (ch == '(' and nxt == '*' and i + 2 < n and text[i + 2] == '$')
+        ):
+            # Raw conditional branches can disagree on END nesting, so keep the file all-or-safe.
+            return text
+
+        if ch.isspace():
+            i += 1
+            while i < n and text[i].isspace():
+                i += 1
+            continue
+
+        if ch == "'":
+            block_end = multiline_string_block_end(text, i)
+            if block_end is not None:
+                i = block_end
+                previous_token = 'literal'
+                previous_token_is_identifier = False
+                generic_owner_ready = False
+                continue
+            i += 1
+            while i < n:
+                if text[i] == "'":
+                    if i + 1 < n and text[i + 1] == "'":
+                        i += 2
+                        continue
+                    i += 1
+                    break
+                i += 1
+            previous_token = 'literal'
+            previous_token_is_identifier = False
+            generic_owner_ready = False
+            continue
+
+        if ch == '/' and nxt == '/':
+            i += 2
+            while i < n and text[i] not in {'\n', '\r'}:
+                i += 1
+            continue
+
+        if ch == '{':
+            i += 1
+            while i < n and text[i] != '}':
+                i += 1
+            if i < n:
+                i += 1
+            continue
+
+        if ch == '(' and nxt == '*':
+            i += 2
+            while i + 1 < n and not (text[i] == '*' and text[i + 1] == ')'):
+                i += 1
+            if i + 1 < n:
+                i += 2
+            continue
+
+        if ch == '&' and (nxt.isalpha() or nxt == '_'):
+            start = i
+            i += 2
+            while i < n and (text[i].isalnum() or text[i] == '_'):
+                i += 1
+            previous_token = text[start:i].casefold()
+            previous_token_is_identifier = True
+            generic_owner_ready = False
+            continue
+
+        if ch.isalpha() or ch == '_':
+            start = i
+            i += 1
+            while i < n and (text[i].isalnum() or text[i] == '_'):
+                i += 1
+            word = text[start:i].casefold()
+            generic_owner_ready = False
+            opens_construct: bool | None = False
+            if not end_stack:
+                if word in {'begin', 'asm'}:
+                    body_start = i
+                    end_stack.append(word)
+            elif word == 'end':
+                end_stack.pop()
+                if not end_stack and body_start is not None:
+                    _blank_preserving_newlines(chars, body_start, start)
+                    body_start = None
+            else:
+                if angle_depth > 0 and word in _END_TERMINATED_STRUCTURED_TYPES:
+                    # Constraint keywords are names inside a confirmed generic angle.
+                    opens_construct = False
+                else:
+                    opens_construct = _opens_end_terminated_construct(
+                        word,
+                        text=text,
+                        word_end=i,
+                        previous_token=previous_token,
+                        end_stack=end_stack,
+                        type_constructor=type_constructors.get((paren_depth, bracket_depth)),
+                    )
+                if opens_construct is None:
+                    return text
+                if opens_construct:
+                    end_stack.append(word)
+            if word == 'class' and opens_construct:
+                generic_owner_ready = True
+            if word in _GENERIC_ROUTINE_HEADINGS:
+                routine_heading_active = True
+            if word in {'begin', 'asm'}:
+                type_constructors.clear()
+                routine_heading_active = False
+            elif word in {'procedure', 'function', 'array'}:
+                type_constructors[(paren_depth, bracket_depth)] = word
+            elif word == 'var':
+                type_constructors.pop((paren_depth, bracket_depth), None)
+            previous_token = word
+            previous_token_is_identifier = True
+            continue
+
+        context_key = (paren_depth, bracket_depth)
+        if ch == '<' and nxt not in {'=', '>'}:
+            if angle_depth > 0:
+                if not previous_token_is_identifier:
+                    return text
+                angle_depth += 1
+            elif generic_owner_ready or (
+                routine_heading_active
+                and previous_token_is_identifier
+                and previous_token != 'operator'
+            ):
+                angle_depth = 1
+            generic_owner_ready = False
+        elif ch == '>' and nxt != '=' and angle_depth > 0:
+            angle_depth -= 1
+            generic_owner_ready = False
+        elif ch == '(':
+            paren_depth += 1
+            generic_owner_ready = False
+            routine_heading_active = False
+        elif ch == ')':
+            type_constructors.pop(context_key, None)
+            paren_depth = max(0, paren_depth - 1)
+            generic_owner_ready = False
+        elif ch == '[':
+            bracket_depth += 1
+            generic_owner_ready = False
+        elif ch == ']':
+            type_constructors.pop(context_key, None)
+            bracket_depth = max(0, bracket_depth - 1)
+            generic_owner_ready = False
+        elif ch == ';':
+            type_constructors.pop(context_key, None)
+            generic_owner_ready = False
+            routine_heading_active = False
+        elif ch not in {'<', '>'}:
+            generic_owner_ready = False
+        previous_token = ch
+        previous_token_is_identifier = False
+        i += 1
+    return text if angle_depth > 0 else ''.join(chars)
+
+
+def _opens_end_terminated_construct(
+    word: str,
+    *,
+    text: str,
+    word_end: int,
+    previous_token: str | None,
+    end_stack: list[str],
+    type_constructor: str | None,
+) -> bool | None:
+    if word in _END_TERMINATED_STRUCTURED_TYPES:
+        if word == 'object' and previous_token == 'of':
+            if type_constructor in {'procedure', 'function'}:
+                return False
+            if type_constructor == 'array':
+                return True
+            return None
+        if word == 'class':
+            next_word = _next_code_word(text, word_end)
+            if next_word == 'of':
+                return False
+            if (
+                end_stack[-1] in _END_TERMINATED_STRUCTURED_TYPES
+                and next_word in _CLASS_MEMBER_PREFIX_FOLLOWERS
+            ):
+                if previous_token in _CLASS_TYPE_PREDECESSORS:
+                    return True
+                if previous_token in _CLASS_MEMBER_BOUNDARIES:
+                    return False
+                return None
+        return True
+    if end_stack[-1] in _END_TERMINATED_STRUCTURED_TYPES:
+        return False
+    return word in _END_TERMINATED_STATEMENTS
+
+
+def _next_code_word(text: str, start: int) -> str | None:
+    i = start
+    n = len(text)
+    while i < n:
+        if text[i].isspace():
+            i += 1
+            continue
+        if text.startswith('//', i):
+            i += 2
+            while i < n and text[i] not in {'\r', '\n'}:
+                i += 1
+            if i < n and text[i] == '\r' and i + 1 < n and text[i + 1] == '\n':
+                i += 2
+            elif i < n:
+                i += 1
+            continue
+        if text[i] == '{':
+            close = text.find('}', i + 1)
+            i = n if close < 0 else close + 1
+            continue
+        if text.startswith('(*', i):
+            close = text.find('*)', i + 2)
+            i = n if close < 0 else close + 2
+            continue
+        break
+    if i >= n or not (text[i].isalpha() or text[i] == '_'):
+        return None
+    end = i + 1
+    while end < n and (text[end].isalnum() or text[end] == '_'):
+        end += 1
+    return text[i:end].casefold()
+
+
+def _blank_preserving_newlines(chars: list[str], start: int, end: int) -> None:
+    for index in range(start, end):
+        if chars[index] not in {'\n', '\r'}:
+            chars[index] = ' '
+
+
+_IDENTIFIER = r'[A-Za-z_][A-Za-z0-9_]*'
+_QUALIFIED_IDENTIFIER = rf'{_IDENTIFIER}(?:\s*\.\s*{_IDENTIFIER})*'
+_UNIT_RE = re.compile(rf'^\s*(?:unit|program|library|package)\s+({_QUALIFIED_IDENTIFIER})\b', re.IGNORECASE)
+_TYPE_RE = re.compile(rf'^\s*({_IDENTIFIER})(?:\s*<[^;=]+>)?\s*=\s*(.*)', re.IGNORECASE)
+_ROUTINE_RE = re.compile(
+    rf'^\s*(?:class\s+)?(procedure|function|constructor|destructor|operator)\s+({_QUALIFIED_IDENTIFIER})\b',
+    re.IGNORECASE,
+)
+_VARIABLE_RE = re.compile(rf'^\s*({_IDENTIFIER})\s*:\s*([^;=]+)', re.IGNORECASE)
+_PROPERTY_RE = re.compile(rf'^\s*property\s+({_IDENTIFIER})\b(?:[^:;]*:\s*([^;]+))?', re.IGNORECASE)
+_CONSTANT_RE = re.compile(rf'^\s*({_IDENTIFIER})\s*(?::[^=]+)?=', re.IGNORECASE)
+_SECTION_RE = re.compile(
+    r'^\s*(type|var|const|threadvar|resourcestring|implementation|initialization|finalization|begin|end)\b',
+    re.IGNORECASE,
+)
+_VISIBILITY_RE = re.compile(
+    r'^\s*(strict\s+private|strict\s+protected|private|protected|public|published)\b',
+    re.IGNORECASE,
+)
+_TYPE_REF_STOP_RE = re.compile(
+    r'\b(read|write|add|remove|stored|default|nodefault|index|implements)\b',
+    re.IGNORECASE,
+)
+
+
+def build_outline_semantic_model(text: str, file_name: str) -> SemanticModel:
+    unit_name = _outline_unit_name(text, file_name)
+    unit_scope = Scope(kind=ScopeKind.UNIT, name=unit_name)
+    unit_range = SourceRange(file_name, 1, 1, 1, max(2, len(unit_name) + 1))
+    unit_symbol = Symbol(
+        name=unit_name,
+        kind=SymbolKind.UNIT,
+        decl_range=unit_range,
+        name_range=unit_range,
+        scope=unit_scope,
+    )
+    unit_scope.owner = unit_symbol
+    unit_scope.define(unit_symbol)
+
+    comment_state: str | None = None
+    section = ''
+    current_routine: Symbol | None = None
+    current_routine_scope: Scope | None = None
+    routine_var_section = False
+    current_type: Symbol | None = None
+    current_type_scope: Scope | None = None
+    type_visibility = Visibility.PUBLIC
+
+    for line_number, raw_line in enumerate(text.splitlines(), start=1):
+        line, comment_state = _mask_delphi_comments_and_strings(raw_line, comment_state)
+        stripped = line.strip()
+        if not stripped:
+            continue
+
+        if section == 'type' and current_type_scope is not None:
+            section_match = _SECTION_RE.match(line)
+            if section_match and section_match.group(1).casefold() == 'end':
+                if current_type is not None:
+                    current_type.decl_range = SourceRange(
+                        file_name,
+                        current_type.decl_range.start_line,
+                        current_type.decl_range.start_col,
+                        line_number,
+                        max(1, len(raw_line) + 1),
+                    )
+                current_type = None
+                current_type_scope = None
+                type_visibility = Visibility.PUBLIC
+                continue
+
+            visibility_match = _VISIBILITY_RE.match(line)
+            if visibility_match:
+                type_visibility = _outline_visibility(visibility_match.group(1))
+                continue
+
+            routine_match = _ROUTINE_RE.match(line)
+            if routine_match:
+                keyword = routine_match.group(1).casefold()
+                _define_outline_symbol(
+                    current_type_scope,
+                    file_name,
+                    _normalize_outline_member_name(routine_match.group(2)),
+                    _routine_symbol_kind(keyword),
+                    line_number,
+                    routine_match.start(2),
+                    routine_match.end(2),
+                    visibility=type_visibility,
+                )
+                continue
+
+            property_match = _PROPERTY_RE.match(line)
+            if property_match:
+                _define_outline_symbol(
+                    current_type_scope,
+                    file_name,
+                    property_match.group(1),
+                    SymbolKind.PROPERTY,
+                    line_number,
+                    property_match.start(1),
+                    property_match.end(1),
+                    visibility=type_visibility,
+                    type_ref=_outline_type_ref(property_match.group(2) or ''),
+                )
+                continue
+
+            field_match = _VARIABLE_RE.match(line)
+            if field_match:
+                _define_outline_symbol(
+                    current_type_scope,
+                    file_name,
+                    field_match.group(1),
+                    SymbolKind.FIELD,
+                    line_number,
+                    field_match.start(1),
+                    field_match.end(1),
+                    visibility=type_visibility,
+                    type_ref=_outline_type_ref(field_match.group(2)),
+                )
+                continue
+
+        routine_match = _ROUTINE_RE.match(line) if section != 'type' else None
+        if routine_match:
+            keyword = routine_match.group(1).casefold()
+            name = _normalize_outline_name(routine_match.group(2))
+            current_routine = _define_outline_symbol(
+                unit_scope,
+                file_name,
+                name,
+                _routine_symbol_kind(keyword),
+                line_number,
+                routine_match.start(2),
+                routine_match.end(2),
+            )
+            current_routine_scope = Scope(
+                kind=ScopeKind.ROUTINE,
+                name=name,
+                parent=unit_scope,
+                owner=current_routine,
+            )
+            current_routine.member_scope = current_routine_scope
+            routine_var_section = False
+            section = ''
+            continue
+
+        section_match = _SECTION_RE.match(line)
+        if section_match:
+            keyword = section_match.group(1).casefold()
+            if current_routine is not None:
+                if keyword in {'var', 'threadvar'}:
+                    routine_var_section = True
+                    continue
+                if keyword == 'begin':
+                    routine_var_section = False
+                    continue
+                if keyword == 'end':
+                    current_routine.decl_range = SourceRange(
+                        file_name,
+                        current_routine.decl_range.start_line,
+                        current_routine.decl_range.start_col,
+                        line_number,
+                        max(1, len(raw_line) + 1),
+                    )
+                    current_routine = None
+                    current_routine_scope = None
+                    routine_var_section = False
+                    continue
+            if keyword == 'type':
+                section = 'type'
+            elif keyword in {'var', 'threadvar', 'const', 'resourcestring'}:
+                section = keyword
+            elif keyword in {'implementation', 'initialization', 'finalization', 'begin'}:
+                section = ''
+            continue
+
+        if current_routine_scope is not None and routine_var_section:
+            var_match = _VARIABLE_RE.match(line)
+            if var_match:
+                _define_outline_symbol(
+                    current_routine_scope,
+                    file_name,
+                    var_match.group(1),
+                    SymbolKind.VARIABLE,
+                    line_number,
+                    var_match.start(1),
+                    var_match.end(1),
+                    type_ref=_outline_type_ref(var_match.group(2)),
+                )
+            continue
+
+        if section == 'type':
+            type_match = _TYPE_RE.match(line)
+            if type_match:
+                kind = _type_symbol_kind(type_match.group(2))
+                type_symbol = _define_outline_symbol(
+                    unit_scope,
+                    file_name,
+                    type_match.group(1),
+                    kind,
+                    line_number,
+                    type_match.start(1),
+                    type_match.end(1),
+                )
+                if kind in {SymbolKind.CLASS, SymbolKind.RECORD, SymbolKind.INTERFACE} and not _is_forward_outline_type(type_match.group(2)):
+                    current_type = type_symbol
+                    current_type_scope = Scope(
+                        kind=ScopeKind.TYPE,
+                        name=type_symbol.name,
+                        parent=unit_scope,
+                        owner=type_symbol,
+                    )
+                    type_symbol.member_scope = current_type_scope
+                    type_visibility = Visibility.PUBLIC
+            continue
+
+        if section in {'const', 'resourcestring'}:
+            const_match = _CONSTANT_RE.match(line)
+            if const_match:
+                _define_outline_symbol(
+                    unit_scope,
+                    file_name,
+                    const_match.group(1),
+                    SymbolKind.CONSTANT,
+                    line_number,
+                    const_match.start(1),
+                    const_match.end(1),
+                )
+            continue
+
+        if section in {'var', 'threadvar'}:
+            var_match = _VARIABLE_RE.match(line)
+            if var_match:
+                _define_outline_symbol(
+                    unit_scope,
+                    file_name,
+                    var_match.group(1),
+                    SymbolKind.VARIABLE,
+                    line_number,
+                    var_match.start(1),
+                    var_match.end(1),
+                    type_ref=_outline_type_ref(var_match.group(2)),
+                )
+
+    index = SymbolIndex()
+    index.register_unit(unit_name, unit_scope)
+    return SemanticModel(unit_scope=unit_scope, index=index)
+
+
+def _outline_unit_name(text: str, file_name: str) -> str:
+    comment_state: str | None = None
+    for raw_line in text.splitlines()[:200]:
+        line, comment_state = _mask_delphi_comments_and_strings(raw_line, comment_state)
+        match = _UNIT_RE.match(line)
+        if match:
+            return _normalize_outline_name(match.group(1))
+    return Path(file_name).stem or 'unit'
+
+
+def _define_outline_symbol(
+    scope: Scope,
+    file_name: str,
+    name: str,
+    kind: SymbolKind,
+    line_number: int,
+    start_col_0: int,
+    end_col_0: int,
+    *,
+    visibility: Visibility = Visibility.UNKNOWN,
+    type_ref: TypeRef | None = None,
+) -> Symbol:
+    symbol_range = SourceRange(
+        file_name,
+        line_number,
+        start_col_0 + 1,
+        line_number,
+        max(start_col_0 + 2, end_col_0 + 1),
+    )
+    symbol = Symbol(
+        name=name,
+        kind=kind,
+        decl_range=symbol_range,
+        name_range=symbol_range,
+        scope=scope,
+        visibility=visibility,
+        type_ref=type_ref or UnknownTypeRef(),
+    )
+    scope.define(symbol)
+    return symbol
+
+
+def _normalize_outline_name(name: str) -> str:
+    return re.sub(r'\s*\.\s*', '.', name.strip())
+
+
+def _normalize_outline_member_name(name: str) -> str:
+    return _normalize_outline_name(name).split('.')[-1]
+
+
+def _outline_visibility(name: str) -> Visibility:
+    normalized = name.casefold().replace(' ', '_')
+    mapping = {
+        'private': Visibility.PRIVATE,
+        'protected': Visibility.PROTECTED,
+        'public': Visibility.PUBLIC,
+        'published': Visibility.PUBLISHED,
+        'strict_private': Visibility.STRICT_PRIVATE,
+        'strict_protected': Visibility.STRICT_PROTECTED,
+    }
+    return mapping.get(normalized, Visibility.UNKNOWN)
+
+
+def _outline_type_ref(type_text: str) -> TypeRef:
+    cleaned = type_text.strip()
+    if not cleaned:
+        return UnknownTypeRef()
+    stop_match = _TYPE_REF_STOP_RE.search(cleaned)
+    if stop_match:
+        cleaned = cleaned[:stop_match.start()].strip()
+    cleaned = cleaned.rstrip(';').strip()
+    if not cleaned:
+        return UnknownTypeRef()
+    return SemanticBuilder()._type_ref_from_name(cleaned)
+
+
+def _is_forward_outline_type(type_text: str) -> bool:
+    rhs = type_text.strip().casefold()
+    return rhs in {'class;', 'record;', 'interface;'} or rhs.endswith(' class;') or rhs.endswith(' interface;')
+
+
+def _routine_symbol_kind(keyword: str) -> SymbolKind:
+    if keyword == 'procedure':
+        return SymbolKind.PROCEDURE
+    if keyword == 'constructor':
+        return SymbolKind.CONSTRUCTOR
+    if keyword == 'destructor':
+        return SymbolKind.DESTRUCTOR
+    return SymbolKind.FUNCTION
+
+
+def _type_symbol_kind(type_text: str) -> SymbolKind:
+    rhs = type_text.lstrip().casefold()
+    if rhs.startswith('packed '):
+        rhs = rhs[7:].lstrip()
+    if rhs.startswith('class') and not rhs.startswith('class of'):
+        return SymbolKind.CLASS
+    if rhs.startswith('record'):
+        return SymbolKind.RECORD
+    if rhs.startswith('interface'):
+        return SymbolKind.INTERFACE
+    if rhs.startswith('('):
+        return SymbolKind.ENUM
+    return SymbolKind.TYPE
+
+
+def _mask_delphi_comments_and_strings(line: str, state: str | None) -> tuple[str, str | None]:
+    chars = list(line)
+    i = 0
+    n = len(line)
+    while i < n:
+        ch = line[i]
+        nxt = line[i + 1] if i + 1 < n else ''
+
+        if state == 'brace':
+            chars[i] = ' '
+            if ch == '}':
+                state = None
+            i += 1
+            continue
+
+        if state == 'paren':
+            chars[i] = ' '
+            if ch == '*' and nxt == ')':
+                chars[i + 1] = ' '
+                state = None
+                i += 2
+            else:
+                i += 1
+            continue
+
+        if ch == "'":
+            chars[i] = ' '
+            i += 1
+            while i < n:
+                chars[i] = ' '
+                if line[i] == "'":
+                    if i + 1 < n and line[i + 1] == "'":
+                        chars[i + 1] = ' '
+                        i += 2
+                        continue
+                    i += 1
+                    break
+                i += 1
+            continue
+
+        if ch == '/' and nxt == '/':
+            for index in range(i, n):
+                chars[index] = ' '
+            break
+
+        if ch == '{':
+            chars[i] = ' '
+            state = 'brace'
+            i += 1
+            continue
+
+        if ch == '(' and nxt == '*':
+            chars[i] = ' '
+            chars[i + 1] = ' '
+            state = 'paren'
+            i += 2
+            continue
+
+        i += 1
+    return ''.join(chars), state
+
+
+def path_to_uri(path: str) -> str:
+    if '://' in path:
+        return path
+    return Path(path).absolute().as_uri()
+
+
+def find_reference_at_position(
+    model: SemanticModel,
+    *,
+    line: int,
+    character: int,
+) -> Optional[SymbolReference]:
+    line_1 = line + 1
+    col_1 = character + 1
+    candidates = [
+        ref for ref in model.references if ref.ref_range.contains(line_1, col_1)
+    ]
+    if not candidates:
+        same_line = [
+            ref for ref in model.references if ref.ref_range.start_line == line_1
+        ]
+        if not same_line:
+            return None
+        return min(
+            same_line,
+            key=lambda ref: abs(ref.ref_range.start_col - col_1),
+        )
+    return min(candidates, key=_reference_span)
+
+
+def find_symbol_at_position(
+    model: SemanticModel,
+    *,
+    line: int,
+    character: int,
+) -> Optional[Symbol]:
+    line_1 = line + 1
+    col_1 = character + 1
+    candidates: list[Symbol] = []
+    for symbol in iter_symbols(model.unit_scope):
+        if symbol.name_range.contains(line_1, col_1):
+            candidates.append(symbol)
+    if not candidates:
+        return None
+    return min(candidates, key=lambda sym: _range_span(sym.name_range))
+
+
+def find_identifier_symbol_at_position(
+    model: SemanticModel,
+    text: str,
+    *,
+    line: int,
+    character: int,
+) -> Optional[Symbol]:
+    identifier = identifier_at_position(text, line, character)
+    if identifier is None:
+        return None
+    scope = scope_at_line(model.unit_scope, line + 1)
+    if scope is None:
+        return None
+    resolved = scope.resolve(identifier)
+    if resolved:
+        return resolved[0]
+    indexed = model.index.lookup(identifier)
+    return indexed[0] if indexed else None
+
+
+def identifier_at_position(text: str, line: int, character: int) -> Optional[str]:
+    lines = text.splitlines()
+    if line < 0 or line >= len(lines):
+        return None
+    line_text = lines[line]
+    if not line_text:
+        return None
+    idx = min(max(character, 0), len(line_text) - 1)
+    if not _is_identifier_char(line_text[idx]) and idx > 0 and _is_identifier_char(line_text[idx - 1]):
+        idx -= 1
+    if not _is_identifier_char(line_text[idx]):
+        return None
+    start = idx
+    while start > 0 and _is_identifier_char(line_text[start - 1]):
+        start -= 1
+    end = idx + 1
+    while end < len(line_text) and _is_identifier_char(line_text[end]):
+        end += 1
+    return line_text[start:end]
+
+
+def scope_at_line(scope: Scope, line_1: int) -> Optional[Scope]:
+    best_scope = scope
+    best_span = float('inf')
+    for symbol in iter_symbols(scope):
+        member_scope = symbol.member_scope
+        if member_scope is None:
+            continue
+        if not (symbol.decl_range.start_line <= line_1 <= symbol.decl_range.end_line):
+            continue
+        span = symbol.decl_range.end_line - symbol.decl_range.start_line
+        if span < best_span:
+            best_span = span
+            best_scope = member_scope
+    return best_scope
+
+
+def _is_identifier_char(ch: str) -> bool:
+    return ch.isalnum() or ch == '_'
+
+
+def _range_span(range_value: SourceRange) -> tuple[int, int]:
+    return (
+        range_value.end_line - range_value.start_line,
+        range_value.end_col - range_value.start_col,
+    )
+
+
+def _reference_span(reference: SymbolReference) -> tuple[int, int]:
+    return _range_span(reference.ref_range)
+
+
+def iter_symbols(scope) -> Iterable[Symbol]:
+    for symbols in scope.symbols.values():
+        for symbol in symbols:
+            yield symbol
+            if symbol.member_scope is not None:
+                yield from iter_symbols(symbol.member_scope)
+
+
+def hover_text(symbol: Symbol) -> str:
+    type_desc = symbol.type_ref.display_name()
+    return f'{symbol.kind.value} {symbol.name}: {type_desc}'
+
+
+def source_range_to_lsp(range_value: SourceRange) -> tuple[int, int, int, int]:
+    return (
+        range_value.start_line - 1,
+        range_value.start_col - 1,
+        range_value.end_line - 1,
+        range_value.end_col - 1,
+    )
+
+
+def resolve_reference(model: SemanticModel, name: str, scope: Scope | None = None) -> Optional[Symbol]:
+    builder = SemanticBuilder()
+    builder._index = model.index
+    return builder._resolve_reference(name, ReferenceKind.VALUE, scope or model.unit_scope)
+
+
+def split_reference_parts(name: str) -> list[str]:
+    builder = SemanticBuilder()
+    return builder._normalized_reference_parts(name)
+
+
+def resolve_base_for_member_completion(model: SemanticModel, name: str, scope: Scope | None = None) -> Optional[Symbol]:
+    lookup_scope = scope or model.unit_scope
+    parts = split_reference_parts(name)
+    if len(parts) <= 1:
+        symbol = resolve_reference(model, name, lookup_scope)
+        if symbol is not None:
+            return symbol
+        symbols = model.index.lookup(name)
+        return symbols[0] if symbols else None
+    base_name = '.'.join(parts[:-1])
+    symbol = resolve_reference(model, base_name, lookup_scope)
+    if symbol is not None:
+        return symbol
+    symbols = model.index.lookup(base_name)
+    return symbols[0] if symbols else None
+
+
+def iter_member_symbols(model: SemanticModel, symbol: Symbol) -> Iterable[Symbol]:
+    builder = SemanticBuilder()
+    builder._index = model.index
+    for scope in builder._iter_member_scopes(symbol, model.unit_scope):
+        for symbols in scope.symbols.values():
+            for member in symbols:
+                yield member
+
+
+def completion_items_for_scope(scope) -> list[Symbol]:
+    seen: set[str] = set()
+    items: list[Symbol] = []
+    for symbols in scope.symbols.values():
+        for symbol in symbols:
+            key = symbol.name.casefold()
+            if key not in seen:
+                seen.add(key)
+                items.append(symbol)
+    for imported in scope.imports:
+        for symbols in imported.symbols.values():
+            for symbol in symbols:
+                key = symbol.name.casefold()
+                if key not in seen:
+                    seen.add(key)
+                    items.append(symbol)
+    return items
+
+
+def references_for_symbol(workspace: WorkspaceSemanticResult, symbol: Symbol) -> list[SymbolReference]:
+    refs: list[SymbolReference] = []
+    for model in workspace.models.values():
+        for ref in model.references:
+            if ref.resolved is symbol:
+                refs.append(ref)
+    return refs
+
+
+def text_references_for_symbol(
+    text: str,
+    symbol: Symbol,
+    *,
+    include_declaration: bool,
+) -> list[SourceRange]:
+    search_range = _text_reference_search_range(symbol)
+    lines = text.splitlines()
+    if search_range.start_line < 1 or search_range.start_line > len(lines):
+        return []
+    end_line = min(search_range.end_line, len(lines))
+    pattern = re.compile(rf'\b{re.escape(symbol.name)}\b', re.IGNORECASE)
+    ranges: list[SourceRange] = []
+    for line_number in range(search_range.start_line, end_line + 1):
+        line_text = lines[line_number - 1]
+        for match in pattern.finditer(line_text):
+            ref_range = SourceRange(
+                symbol.decl_range.file_name,
+                line_number,
+                match.start() + 1,
+                line_number,
+                match.end() + 1,
+            )
+            if not include_declaration and _same_range(ref_range, symbol.name_range):
+                continue
+            ranges.append(ref_range)
+    return ranges
+
+
+def _text_reference_search_range(symbol: Symbol) -> SourceRange:
+    owner = symbol.scope.owner
+    if owner is not None and owner is not symbol:
+        return owner.decl_range
+    return SourceRange(
+        symbol.decl_range.file_name,
+        1,
+        1,
+        max(symbol.decl_range.end_line, 1_000_000_000),
+        1,
+    )
+
+
+def _same_range(left: SourceRange, right: SourceRange) -> bool:
+    return (
+        left.file_name == right.file_name
+        and left.start_line == right.start_line
+        and left.start_col == right.start_col
+        and left.end_line == right.end_line
+        and left.end_col == right.end_col
+    )
+
+
+def diagnostics_for_model(model: SemanticModel):
+    try:
+        from lsprotocol.types import Diagnostic, DiagnosticSeverity, Range, Position
+    except ImportError:
+        return []
+    diagnostics = []
+    for problem in model.problems:
+        start_line, start_col, end_line, end_col = source_range_to_lsp(problem.range)
+        diagnostics.append(
+            Diagnostic(
+                range=Range(
+                    start=Position(line=start_line, character=start_col),
+                    end=Position(line=end_line, character=end_col),
+                ),
+                message=problem.message,
+                severity=DiagnosticSeverity.Error,
+                source='agent-codinglanguage-mapper',
+            )
+        )
+    return diagnostics
+
+
+def extract_completion_base(text: str, line: int, character: int) -> Optional[str]:
+    lines = text.splitlines()
+    if line < 0 or line >= len(lines):
+        return None
+    line_text = lines[line]
+    if character <= 0 or character > len(line_text):
+        return None
+    if line_text[character - 1] != '.':
+        return None
+    idx = character - 2
+    if idx < 0:
+        return None
+    allowed = set('abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_<>.,&')
+    buf = []
+    while idx >= 0:
+        ch = line_text[idx]
+        if ch.isspace():
+            break
+        if ch in allowed:
+            buf.append(ch)
+            idx -= 1
+            continue
+        break
+    if not buf:
+        return None
+    return ''.join(reversed(buf)).strip()
+
+
+# LSP server entrypoint
+
+def create_server():
+    try:
+        from pygls.server import LanguageServer
+        from lsprotocol.types import (
+            CompletionItem,
+            CompletionItemKind,
+            CompletionList,
+            CompletionOptions,
+            CompletionParams,
+            DefinitionParams,
+            Hover,
+            HoverParams,
+            InitializeParams,
+            InitializeResult,
+            Location,
+            Position,
+            Range,
+            ReferenceParams,
+            RenameParams,
+            ServerCapabilities,
+            SymbolInformation,
+            SymbolKind as LspSymbolKind,
+            TextDocumentSyncKind,
+            TextDocumentSyncOptions,
+            TextEdit,
+            WorkspaceEdit,
+            WorkspaceSymbolParams,
+            DocumentSymbolParams,
+            DocumentSymbol,
+            TEXT_DOCUMENT_COMPLETION,
+            TEXT_DOCUMENT_DEFINITION,
+            TEXT_DOCUMENT_DID_CHANGE,
+            TEXT_DOCUMENT_DID_CLOSE,
+            TEXT_DOCUMENT_DID_OPEN,
+            TEXT_DOCUMENT_HOVER,
+            TEXT_DOCUMENT_REFERENCES,
+            TEXT_DOCUMENT_RENAME,
+            TEXT_DOCUMENT_DOCUMENT_SYMBOL,
+            WORKSPACE_SYMBOL,
+            INITIALIZE,
+        )
+    except ImportError as exc:  # pragma: no cover - optional dependency
+        raise RuntimeError('pygls and lsprotocol are required for the LSP server') from exc
+
+    server = LanguageServer('agent-codinglanguage-mapper', __version__)
+    state = LspWorkspaceState()
+
+    def _symbol_kind(symbol: Symbol) -> LspSymbolKind:
+        mapping = {
+            SymbolKind.CLASS: LspSymbolKind.Class,
+            SymbolKind.RECORD: LspSymbolKind.Struct,
+            SymbolKind.INTERFACE: LspSymbolKind.Interface,
+            SymbolKind.ENUM: LspSymbolKind.Enum,
+            SymbolKind.ENUM_VALUE: LspSymbolKind.EnumMember,
+            SymbolKind.FIELD: LspSymbolKind.Field,
+            SymbolKind.PROPERTY: LspSymbolKind.Property,
+            SymbolKind.METHOD: LspSymbolKind.Method,
+            SymbolKind.FUNCTION: LspSymbolKind.Function,
+            SymbolKind.PROCEDURE: LspSymbolKind.Function,
+            SymbolKind.CONSTRUCTOR: LspSymbolKind.Constructor,
+            SymbolKind.DESTRUCTOR: LspSymbolKind.Method,
+            SymbolKind.VARIABLE: LspSymbolKind.Variable,
+            SymbolKind.PARAMETER: LspSymbolKind.Variable,
+            SymbolKind.CONSTANT: LspSymbolKind.Constant,
+            SymbolKind.TYPE: LspSymbolKind.Class,
+        }
+        return mapping.get(symbol.kind, LspSymbolKind.String)
+
+    def _completion_kind(symbol: Symbol) -> CompletionItemKind:
+        mapping = {
+            SymbolKind.CLASS: CompletionItemKind.Class,
+            SymbolKind.RECORD: CompletionItemKind.Struct,
+            SymbolKind.INTERFACE: CompletionItemKind.Interface,
+            SymbolKind.ENUM: CompletionItemKind.Enum,
+            SymbolKind.ENUM_VALUE: CompletionItemKind.EnumMember,
+            SymbolKind.FIELD: CompletionItemKind.Field,
+            SymbolKind.PROPERTY: CompletionItemKind.Property,
+            SymbolKind.METHOD: CompletionItemKind.Method,
+            SymbolKind.FUNCTION: CompletionItemKind.Function,
+            SymbolKind.PROCEDURE: CompletionItemKind.Function,
+            SymbolKind.CONSTRUCTOR: CompletionItemKind.Constructor,
+            SymbolKind.DESTRUCTOR: CompletionItemKind.Method,
+            SymbolKind.VARIABLE: CompletionItemKind.Variable,
+            SymbolKind.PARAMETER: CompletionItemKind.Variable,
+            SymbolKind.CONSTANT: CompletionItemKind.Constant,
+            SymbolKind.TYPE: CompletionItemKind.Class,
+        }
+        return mapping.get(symbol.kind, CompletionItemKind.Text)
+
+    def _document_symbols_for_scope(scope) -> list[DocumentSymbol]:
+        items: list[DocumentSymbol] = []
+        for symbols in scope.symbols.values():
+            for symbol in symbols:
+                if symbol.kind == SymbolKind.UNIT:
+                    continue
+                start_line, start_col, end_line, end_col = source_range_to_lsp(symbol.decl_range)
+                sel_start, sel_col, sel_end, sel_end_col = source_range_to_lsp(symbol.name_range)
+                children = []
+                if symbol.member_scope is not None:
+                    children = _document_symbols_for_scope(symbol.member_scope)
+                items.append(
+                    DocumentSymbol(
+                        name=symbol.name,
+                        kind=_symbol_kind(symbol),
+                        range=Range(
+                            start=Position(line=start_line, character=start_col),
+                            end=Position(line=end_line, character=end_col),
+                        ),
+                        selection_range=Range(
+                            start=Position(line=sel_start, character=sel_col),
+                            end=Position(line=sel_end, character=sel_end_col),
+                        ),
+                        children=children or None,
+                    )
+                )
+        return items
+
+    @server.feature(INITIALIZE)
+    def initialize(ls: LanguageServer, params: InitializeParams) -> InitializeResult:
+        roots: list[str] = []
+        if params.workspace_folders:
+            roots.extend(uri_to_path(folder.uri) for folder in params.workspace_folders)
+        elif params.root_uri:
+            roots.append(uri_to_path(params.root_uri))
+        init_opts = params.initialization_options or {}
+        include_paths = [uri_to_path(path) for path in init_opts.get('includePaths', [])]
+        search_paths = [uri_to_path(path) for path in init_opts.get('searchPaths', [])]
+        defines = init_opts.get('defines', [])
+        auto_discover_paths = init_opts.get('autoDiscoverPaths', True)
+        config = WorkspaceConfig(
+            roots=roots,
+            include_paths=include_paths,
+            search_paths=search_paths,
+            defines=defines,
+            auto_discover_paths=bool(auto_discover_paths),
+        )
+        state.configure(config)
+
+        capabilities = ServerCapabilities(
+            text_document_sync=TextDocumentSyncOptions(
+                open_close=True,
+                change=TextDocumentSyncKind.Full,
+            ),
+            definition_provider=True,
+            references_provider=True,
+            hover_provider=True,
+            completion_provider=CompletionOptions(trigger_characters=['.']),
+            rename_provider=True,
+            document_symbol_provider=True,
+            workspace_symbol_provider=True,
+        )
+        return InitializeResult(capabilities=capabilities)
+
+    def _publish_diagnostics(ls: LanguageServer, uri: str) -> None:
+        diagnostics = state.diagnostics_for_uri(uri)
+        ls.publish_diagnostics(uri, diagnostics)
+
+    def _symbol_at_position_in_model(
+        model: SemanticModel,
+        uri: str,
+        position: Position,
+        text: str | None = None,
+    ) -> Optional[Symbol]:
+        if text is None:
+            text = state.text_for_uri(uri)
+        ref = find_reference_at_position(
+            model,
+            line=position.line,
+            character=position.character,
+        )
+        symbol = ref.resolved if ref and ref.resolved else find_symbol_at_position(
+            model,
+            line=position.line,
+            character=position.character,
+        )
+        if symbol is None and text is not None:
+            symbol = find_identifier_symbol_at_position(
+                model,
+                text,
+                line=position.line,
+                character=position.character,
+            )
+        return symbol
+
+    def _symbol_at_position(uri: str, position: Position) -> Optional[Symbol]:
+        text = state.text_for_uri(uri)
+        model = state.semantic_for_uri(uri)
+        if model is not None:
+            symbol = _symbol_at_position_in_model(model, uri, position, text)
+            if symbol is not None:
+                return symbol
+        model = state.full_semantic_for_uri(uri)
+        if model is not None:
+            symbol = _symbol_at_position_in_model(model, uri, position, text)
+            if symbol is not None:
+                return symbol
+        return None
+
+    def _add_rename_edit(
+        edits: dict[str, list[TextEdit]],
+        seen: set[tuple[str, int, int, int, int]],
+        ref_range: SourceRange,
+        new_text: str,
+    ) -> None:
+        uri = state.uri_for_file_name(ref_range.file_name) or path_to_uri(ref_range.file_name)
+        key = (uri, ref_range.start_line, ref_range.start_col, ref_range.end_line, ref_range.end_col)
+        if key in seen:
+            return
+        seen.add(key)
+        start_line, start_col, end_line, end_col = source_range_to_lsp(ref_range)
+        edits.setdefault(uri, []).append(
+            TextEdit(
+                range=Range(
+                    start=Position(line=start_line, character=start_col),
+                    end=Position(line=end_line, character=end_col),
+                ),
+                new_text=new_text,
+            )
+        )
+
+    @server.feature(TEXT_DOCUMENT_DID_OPEN)
+    def did_open(ls: LanguageServer, params) -> None:
+        state.update_document(params.text_document.uri, params.text_document.text)
+        _publish_diagnostics(ls, params.text_document.uri)
+
+    @server.feature(TEXT_DOCUMENT_DID_CHANGE)
+    def did_change(ls: LanguageServer, params) -> None:
+        if not params.content_changes:
+            return
+        text = params.content_changes[-1].text
+        state.update_document(params.text_document.uri, text)
+        _publish_diagnostics(ls, params.text_document.uri)
+
+    @server.feature(TEXT_DOCUMENT_DID_CLOSE)
+    def did_close(ls: LanguageServer, params) -> None:
+        state.remove_document(params.text_document.uri)
+        ls.publish_diagnostics(params.text_document.uri, [])
+
+    @server.feature(TEXT_DOCUMENT_DEFINITION)
+    def definition(ls: LanguageServer, params: DefinitionParams):
+        symbol = _symbol_at_position(params.text_document.uri, params.position)
+        if symbol is None:
+            return None
+        file_name = symbol.decl_range.file_name
+        uri = state.uri_for_file_name(file_name) or path_to_uri(file_name)
+        start_line, start_col, end_line, end_col = source_range_to_lsp(symbol.decl_range)
+        return Location(
+            uri=uri,
+            range=Range(
+                start=Position(line=start_line, character=start_col),
+                end=Position(line=end_line, character=end_col),
+            ),
+        )
+
+    @server.feature(TEXT_DOCUMENT_HOVER)
+    def hover(ls: LanguageServer, params: HoverParams) -> Optional[Hover]:
+        symbol = _symbol_at_position(params.text_document.uri, params.position)
+        if symbol is None:
+            return None
+        return Hover(contents=hover_text(symbol))
+
+    @server.feature(TEXT_DOCUMENT_REFERENCES)
+    def references(ls: LanguageServer, params: ReferenceParams):
+        symbol = _symbol_at_position(params.text_document.uri, params.position)
+        if symbol is None:
+            return []
+        locations = []
+        if state.workspace is not None:
+            for ref in references_for_symbol(state.workspace, symbol):
+                file_name = ref.ref_range.file_name
+                uri = state.uri_for_file_name(file_name) or path_to_uri(file_name)
+                start_line, start_col, end_line, end_col = source_range_to_lsp(ref.ref_range)
+                locations.append(
+                    Location(
+                        uri=uri,
+                        range=Range(
+                            start=Position(line=start_line, character=start_col),
+                            end=Position(line=end_line, character=end_col),
+                        ),
+                    )
+                )
+        if locations:
+            return locations
+        text = state.text_for_uri(params.text_document.uri)
+        if text is None:
+            return []
+        include_declaration = getattr(params.context, 'include_declaration', True)
+        for ref_range in text_references_for_symbol(
+            text,
+            symbol,
+            include_declaration=include_declaration,
+        ):
+            uri = state.uri_for_file_name(ref_range.file_name) or path_to_uri(ref_range.file_name)
+            start_line, start_col, end_line, end_col = source_range_to_lsp(ref_range)
+            locations.append(
+                Location(
+                    uri=uri,
+                    range=Range(
+                        start=Position(line=start_line, character=start_col),
+                        end=Position(line=end_line, character=end_col),
+                    ),
+                )
+            )
+        return locations
+
+    @server.feature(TEXT_DOCUMENT_RENAME)
+    def rename(ls: LanguageServer, params: RenameParams) -> Optional[WorkspaceEdit]:
+        symbol = _symbol_at_position(params.text_document.uri, params.position)
+        if symbol is None:
+            return None
+
+        text = state.text_for_uri(params.text_document.uri)
+        semantic_symbol: Symbol | None = None
+        semantic_model = state.semantic_for_uri(params.text_document.uri)
+        if semantic_model is not None:
+            semantic_symbol = _symbol_at_position_in_model(
+                semantic_model,
+                params.text_document.uri,
+                params.position,
+                text,
+            )
+
+        edits: dict[str, list[TextEdit]] = {}
+        seen: set[tuple[str, int, int, int, int]] = set()
+        workspace_symbol = semantic_symbol or symbol
+        if state.workspace is not None and semantic_symbol is not None:
+            for ref in references_for_symbol(state.workspace, semantic_symbol):
+                if '.' in ref.name:
+                    continue
+                _add_rename_edit(edits, seen, ref.ref_range, params.new_name)
+
+        if not edits and text is not None:
+            for ref_range in text_references_for_symbol(
+                text,
+                symbol,
+                include_declaration=True,
+            ):
+                _add_rename_edit(edits, seen, ref_range, params.new_name)
+
+        _add_rename_edit(
+            edits,
+            seen,
+            workspace_symbol.name_range,
+            params.new_name,
+        )
+        return WorkspaceEdit(changes=edits)
+
+    @server.feature(TEXT_DOCUMENT_DOCUMENT_SYMBOL)
+    def document_symbols(ls: LanguageServer, params: DocumentSymbolParams):
+        model = state.structure_semantic_for_uri(params.text_document.uri)
+        if model is None:
+            return []
+        return _document_symbols_for_scope(model.unit_scope)
+
+    @server.feature(WORKSPACE_SYMBOL)
+    def workspace_symbols(ls: LanguageServer, params: WorkspaceSymbolParams):
+        query = (params.query or '').strip().casefold()
+        workspace = state.workspace_symbols_for_query(query)
+        if workspace is None:
+            return []
+        items: list[SymbolInformation] = []
+        for model in workspace.models.values():
+            for symbol in iter_symbols(model.unit_scope):
+                if symbol.kind == SymbolKind.UNIT:
+                    continue
+                if query and query not in symbol.name.casefold():
+                    continue
+                file_name = symbol.decl_range.file_name
+                uri = state.uri_for_file_name(file_name) or path_to_uri(file_name)
+                start_line, start_col, end_line, end_col = source_range_to_lsp(symbol.decl_range)
+                items.append(
+                    SymbolInformation(
+                        name=symbol.name,
+                        kind=_symbol_kind(symbol),
+                        location=Location(
+                            uri=uri,
+                            range=Range(
+                                start=Position(line=start_line, character=start_col),
+                                end=Position(line=end_line, character=end_col),
+                            ),
+                        ),
+                    )
+                )
+        return items
+
+    @server.feature(TEXT_DOCUMENT_COMPLETION)
+    def completion(ls: LanguageServer, params: CompletionParams):
+        model = state.structure_semantic_for_uri(params.text_document.uri)
+        if model is None:
+            return CompletionList(is_incomplete=True, items=[])
+        doc = state.documents.get(params.text_document.uri)
+        base_expr = None
+        if doc is not None:
+            base_expr = extract_completion_base(doc.text, params.position.line, params.position.character)
+        active_scope = scope_at_line(model.unit_scope, params.position.line + 1) or model.unit_scope
+        symbols: list[Symbol]
+        if base_expr:
+            base_symbol = resolve_reference(model, base_expr, active_scope)
+            if base_symbol is None:
+                base_symbol = resolve_base_for_member_completion(model, base_expr, active_scope)
+            if base_symbol is not None:
+                symbols = list(iter_member_symbols(model, base_symbol))
+            else:
+                symbols = completion_items_for_scope(active_scope)
+            if not symbols:
+                full_model = state.full_semantic_for_uri(params.text_document.uri)
+                if full_model is not None and full_model is not model:
+                    full_scope = scope_at_line(full_model.unit_scope, params.position.line + 1) or full_model.unit_scope
+                    full_base_symbol = resolve_reference(full_model, base_expr, full_scope)
+                    if full_base_symbol is None:
+                        full_base_symbol = resolve_base_for_member_completion(full_model, base_expr, full_scope)
+                    if full_base_symbol is not None:
+                        symbols = list(iter_member_symbols(full_model, full_base_symbol))
+        else:
+            full_model = state.full_semantic_for_uri(params.text_document.uri)
+            if full_model is not None:
+                full_scope = scope_at_line(full_model.unit_scope, params.position.line + 1) or full_model.unit_scope
+                symbols = completion_items_for_scope(full_scope)
+            else:
+                symbols = completion_items_for_scope(active_scope)
+        items = [CompletionItem(label=symbol.name, kind=_completion_kind(symbol)) for symbol in symbols]
+        return CompletionList(is_incomplete=False, items=items)
+
+    return server
+
+
+def main() -> None:
+    server = create_server()
+    server.start_io()
+
+
+__all__ = [
+    'create_server',
+    'LspWorkspaceState',
+    'WorkspaceConfig',
+    'DocumentSnapshot',
+    'find_reference_at_position',
+    'find_symbol_at_position',
+    'resolve_reference',
+    'extract_completion_base',
+    'outline_source',
+    'multiline_string_block_end',
+    'main',
+]
+
+
+if __name__ == '__main__':
+    main()