const std = @import("std"); const Ast = std.zig.Ast; const DocumentStore = @import("DocumentStore.zig"); const analysis = @import("analysis.zig"); const ast = @import("ast.zig"); const types = @import("types.zig"); const requests = @import("requests.zig"); const offsets = @import("offsets.zig"); pub const Builder = struct { arena: *std.heap.ArenaAllocator, document_store: *DocumentStore, handle: *const DocumentStore.Handle, offset_encoding: offsets.Encoding, pub fn generateCodeAction(builder: *Builder, diagnostic: types.Diagnostic, actions: *std.ArrayListUnmanaged(types.CodeAction)) error{OutOfMemory}!void { const kind = DiagnosticKind.parse(diagnostic.message) orelse return; const loc = offsets.rangeToLoc(builder.handle.text, diagnostic.range, builder.offset_encoding); switch (kind) { .unused => |id| switch (id) { .@"function parameter" => try handleUnusedFunctionParameter(builder, actions, loc), .@"local constant" => try handleUnusedVariableOrConstant(builder, actions, loc), .@"local variable" => try handleUnusedVariableOrConstant(builder, actions, loc), .@"loop index capture" => try handleUnusedIndexCapture(builder, actions, loc), .capture => try handleUnusedCapture(builder, actions, loc), }, .non_camelcase_fn => try handleNonCamelcaseFunction(builder, actions, loc), .pointless_discard => try handlePointlessDiscard(builder, actions, loc), .omit_discard => |id| switch (id) { .@"index capture" => try handleUnusedIndexCapture(builder, actions, loc), .@"error capture" => try handleUnusedCapture(builder, actions, loc), }, // the undeclared identifier may be a discard .undeclared_identifier => try handlePointlessDiscard(builder, actions, loc), .unreachable_code => { // TODO // autofix: comment out code // fix: remove code }, } } pub fn createTextEditLoc(self: *Builder, loc: offsets.Loc, new_text: []const u8) types.TextEdit { const range = offsets.locToRange(self.handle.text, loc, self.offset_encoding); return types.TextEdit{ .range = range, .newText = new_text }; } pub fn createTextEditPos(self: *Builder, index: usize, new_text: []const u8) types.TextEdit { const position = offsets.indexToPosition(self.handle.text, index, self.offset_encoding); return types.TextEdit{ .range = .{ .start = position, .end = position }, .newText = new_text }; } pub fn createWorkspaceEdit(self: *Builder, edits: []const types.TextEdit) error{OutOfMemory}!types.WorkspaceEdit { var text_edits = std.ArrayListUnmanaged(types.TextEdit){}; try text_edits.appendSlice(self.arena.allocator(), edits); var workspace_edit = types.WorkspaceEdit{ .changes = .{} }; try workspace_edit.changes.putNoClobber(self.arena.allocator(), self.handle.uri, text_edits); return workspace_edit; } }; fn handleNonCamelcaseFunction(builder: *Builder, actions: *std.ArrayListUnmanaged(types.CodeAction), loc: offsets.Loc) !void { const identifier_name = offsets.locToSlice(builder.handle.text, loc); if (std.mem.allEqual(u8, identifier_name, '_')) return; const new_text = try createCamelcaseText(builder.arena.allocator(), identifier_name); const action1 = types.CodeAction{ .title = "make function name camelCase", .kind = .QuickFix, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditLoc(loc, new_text)}), }; try actions.append(builder.arena.allocator(), action1); } fn handleUnusedFunctionParameter(builder: *Builder, actions: *std.ArrayListUnmanaged(types.CodeAction), loc: offsets.Loc) !void { const identifier_name = offsets.locToSlice(builder.handle.text, loc); const tree = builder.handle.tree; const node_tags = tree.nodes.items(.tag); const node_datas = tree.nodes.items(.data); const node_tokens = tree.nodes.items(.main_token); const token_starts = tree.tokens.items(.start); const decl = (try analysis.lookupSymbolGlobal( builder.document_store, builder.arena, builder.handle, identifier_name, loc.start, )) orelse return; const payload = switch (decl.decl.*) { .param_payload => |pay| pay, else => return, }; std.debug.assert(node_tags[payload.func] == .fn_decl); const block = node_datas[payload.func].rhs; const new_text = try createDiscardText(builder, identifier_name, token_starts[node_tokens[payload.func]], true); const index = token_starts[node_tokens[block]] + 1; const action1 = types.CodeAction{ .title = "discard function parameter", .kind = .SourceFixAll, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditPos(index, new_text)}), }; // TODO fix formatting const action2 = types.CodeAction{ .title = "remove function parameter", .kind = .QuickFix, .isPreferred = false, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditLoc(getParamRemovalRange(tree, payload.param), "")}), }; try actions.appendSlice(builder.arena.allocator(), &.{ action1, action2 }); } fn handleUnusedVariableOrConstant(builder: *Builder, actions: *std.ArrayListUnmanaged(types.CodeAction), loc: offsets.Loc) !void { const identifier_name = offsets.locToSlice(builder.handle.text, loc); const tree = builder.handle.tree; const token_tags = tree.tokens.items(.tag); const token_starts = tree.tokens.items(.start); const decl = (try analysis.lookupSymbolGlobal( builder.document_store, builder.arena, builder.handle, identifier_name, loc.start, )) orelse return; const node = switch (decl.decl.*) { .ast_node => |node| node, else => return, }; const first_token = tree.firstToken(node); const last_token = ast.lastToken(tree, node) + 1; if (token_tags[last_token] != .semicolon) return; const new_text = try createDiscardText(builder, identifier_name, token_starts[first_token], false); const index = token_starts[last_token] + 1; try actions.append(builder.arena.allocator(), .{ .title = "discard value", .kind = .SourceFixAll, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditPos(index, new_text)}), }); } fn handleUnusedIndexCapture(builder: *Builder, actions: *std.ArrayListUnmanaged(types.CodeAction), loc: offsets.Loc) !void { const capture_locs = getCaptureLoc(builder.handle.text, loc, true) orelse return; // TODO support discarding without modifying the capture // by adding a discard in the block scope const is_value_discarded = std.mem.eql(u8, offsets.locToSlice(builder.handle.text, capture_locs.value), "_"); if (is_value_discarded) { // |_, i| -> // TODO fix formatting try actions.append(builder.arena.allocator(), .{ .title = "remove capture", .kind = .QuickFix, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditLoc(capture_locs.loc, "")}), }); } else { // |v, i| -> |v| // |v, _| -> |v| try actions.append(builder.arena.allocator(), .{ .title = "remove index capture", .kind = .QuickFix, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditLoc( .{ .start = capture_locs.value.end, .end = capture_locs.loc.end - 1 }, "", )}), }); } } fn handleUnusedCapture(builder: *Builder, actions: *std.ArrayListUnmanaged(types.CodeAction), loc: offsets.Loc) !void { const capture_locs = getCaptureLoc(builder.handle.text, loc, false) orelse return; // TODO support discarding without modifying the capture // by adding a discard in the block scope if (capture_locs.index != null) { // |v, i| -> |_, i| try actions.append(builder.arena.allocator(), .{ .title = "discard capture", .kind = .QuickFix, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditLoc(capture_locs.value, "_")}), }); } else { // |v| -> // TODO fix formatting try actions.append(builder.arena.allocator(), .{ .title = "remove capture", .kind = .QuickFix, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{builder.createTextEditLoc(capture_locs.loc, "")}), }); } } fn handlePointlessDiscard(builder: *Builder, actions: *std.ArrayListUnmanaged(types.CodeAction), loc: offsets.Loc) !void { const edit_loc = getDiscardLoc(builder.handle.text, loc) orelse return; try actions.append(builder.arena.allocator(), .{ .title = "remove pointless discard", .kind = .SourceFixAll, .isPreferred = true, .edit = try builder.createWorkspaceEdit(&.{ builder.createTextEditLoc(edit_loc, ""), }), }); } fn detectIndentation(source: []const u8) []const u8 { // Essentially I'm looking for the first indentation in the file. var i: usize = 0; var len = source.len - 1; // I need 1 look-ahead while (i < len) : (i += 1) { if (source[i] != '\n') continue; i += 1; if (source[i] == '\t') return "\t"; var space_count: usize = 0; while (i < source.len and source[i] == ' ') : (i += 1) { space_count += 1; } if (source[i] == '\n') { // Some editors mess up indentation of empty lines i -= 1; continue; } if (space_count == 0) continue; if (source[i] == '/') continue; // Comments sometimes have additional alignment. if (source[i] == '\\') continue; // multi-line strings might as well. return source[i - space_count .. i]; } return " " ** 4; // recommended style } // attempts to converts a slice of text into camelcase 'FUNCTION_NAME' -> 'functionName' fn createCamelcaseText(allocator: std.mem.Allocator, identifier: []const u8) ![]const u8 { // skip initial & ending underscores const trimmed_identifier = std.mem.trim(u8, identifier, "_"); const num_separators = std.mem.count(u8, trimmed_identifier, "_"); const new_text_len = trimmed_identifier.len - num_separators; var new_text = try std.ArrayListUnmanaged(u8).initCapacity(allocator, new_text_len); errdefer new_text.deinit(allocator); var idx: usize = 0; while (idx < trimmed_identifier.len) { const ch = trimmed_identifier[idx]; if (ch == '_') { // the trimmed identifier is guaranteed to not have underscores at the end, // so it can be assumed that ptr dereferences are safe until an alnum char is found while (trimmed_identifier[idx] == '_') : (idx += 1) {} const ch2 = trimmed_identifier[idx]; new_text.appendAssumeCapacity(std.ascii.toUpper(ch2)); } else { new_text.appendAssumeCapacity(std.ascii.toLower(ch)); } idx += 1; } return new_text.toOwnedSlice(allocator); } // returns a discard string `\n{indent}_ = identifier_name;` fn createDiscardText(builder: *Builder, identifier_name: []const u8, declaration_start: usize, add_block_indentation: bool) ![]const u8 { const indent = find_indent: { const line = offsets.lineSliceUntilIndex(builder.handle.text, declaration_start); for (line) |char, i| { if (!std.ascii.isWhitespace(char)) { break :find_indent line[0..i]; } } break :find_indent line; }; const additional_indent = if (add_block_indentation) detectIndentation(builder.handle.text) else ""; const allocator = builder.arena.allocator(); const new_text_len = 1 + indent.len + additional_indent.len + "_ = ;".len + identifier_name.len; var new_text = try std.ArrayListUnmanaged(u8).initCapacity(allocator, new_text_len); errdefer new_text.deinit(allocator); new_text.appendAssumeCapacity('\n'); new_text.appendSliceAssumeCapacity(indent); new_text.appendSliceAssumeCapacity(additional_indent); new_text.appendSliceAssumeCapacity("_ = "); new_text.appendSliceAssumeCapacity(identifier_name); new_text.appendAssumeCapacity(';'); return new_text.toOwnedSlice(allocator); } fn getParamRemovalRange(tree: Ast, param: Ast.full.FnProto.Param) offsets.Loc { var param_start = offsets.tokenToIndex(tree, ast.paramFirstToken(tree, param)); var param_end = offsets.tokenToLoc(tree, ast.paramLastToken(tree, param)).end; var trim_end = false; while (param_start != 0) : (param_start -= 1) { switch (tree.source[param_start - 1]) { ' ', '\n' => continue, ',' => { param_start -= 1; break; }, '(' => { trim_end = true; break; }, else => break, } } var found_comma = false; while (trim_end and param_end < tree.source.len) : (param_end += 1) { switch (tree.source[param_end]) { ' ', '\n' => continue, ',' => if (!found_comma) { found_comma = true; continue; } else { param_end += 1; break; }, ')' => break, else => break, } } return .{ .start = param_start, .end = param_end }; } const DiagnosticKind = union(enum) { unused: IdCat, pointless_discard: IdCat, omit_discard: DiscardCat, non_camelcase_fn, undeclared_identifier, unreachable_code, const IdCat = enum { @"function parameter", @"local constant", @"local variable", @"loop index capture", capture, }; const DiscardCat = enum { // "discard of index capture; omit it instead" @"index capture", // "discard of error capture; omit it instead" @"error capture", }; pub fn parse(diagnostic_message: []const u8) ?DiagnosticKind { const msg = diagnostic_message; if (std.mem.startsWith(u8, msg, "unused ")) { return DiagnosticKind{ .unused = parseEnum(IdCat, msg["unused ".len..]) orelse return null, }; } else if (std.mem.startsWith(u8, msg, "pointless discard of ")) { return DiagnosticKind{ .pointless_discard = parseEnum(IdCat, msg["pointless discard of ".len..]) orelse return null, }; } else if (std.mem.startsWith(u8, msg, "discard of ")) { return DiagnosticKind{ .omit_discard = parseEnum(DiscardCat, msg["discard of ".len..]) orelse return null, }; } else if (std.mem.startsWith(u8, msg, "Functions should be camelCase")) { return .non_camelcase_fn; } else if (std.mem.startsWith(u8, msg, "use of undeclared identifier")) { return .undeclared_identifier; } return null; } fn parseEnum(comptime T: type, message: []const u8) ?T { inline for (std.meta.fields(T)) |field| { if (std.mem.startsWith(u8, message, field.name)) { // is there a better way to achieve this? return @intToEnum(T, field.value); } } return null; } }; /// takes the location of an identifier which is part of a discard `_ = location_here;` /// and returns the location from '_' until ';' or null on failure fn getDiscardLoc(text: []const u8, loc: offsets.Loc) ?offsets.Loc { // check of the loc points to a valid identifier for (offsets.locToSlice(text, loc)) |c| { if (!isSymbolChar(c)) return null; } // check if the identifier is followed by a colon const colon_position = found: { var i = loc.end; while (i < text.len) : (i += 1) { switch (text[i]) { ' ' => continue, ';' => break :found i + 1, else => return null, } } return null; }; // check if the identifier is preceed by a equal sign and then an underscore var i: usize = loc.start - 1; var found_equal_sign = false; const underscore_position = found: { while (true) : (i -= 1) { if (i == 0) return null; switch (text[i]) { ' ' => {}, '=' => { if (found_equal_sign) return null; found_equal_sign = true; }, '_' => if (found_equal_sign) break :found i else return null, else => return null, } } }; // move backwards until we find a newline i = underscore_position - 1; const start_position = found: { while (true) : (i -= 1) { if (i == 0) break :found underscore_position; switch (text[i]) { ' ' => {}, '\n' => break :found i, else => break :found underscore_position, } } }; return offsets.Loc{ .start = start_position, .end = colon_position, }; } const CaptureLocs = struct { loc: offsets.Loc, value: offsets.Loc, index: ?offsets.Loc, }; /// takes the location of an identifier which is part of a payload `|value, index|` /// and returns the location from '|' until '|' or null on failure /// use `is_index_payload` to indicate whether `loc` points to `value` or `index` fn getCaptureLoc(text: []const u8, loc: offsets.Loc, is_index_payload: bool) ?CaptureLocs { const value_end = if (!is_index_payload) loc.end else found: { // move back until we find a comma const comma_position = found_comma: { var i = loc.start - 1; while (i != 0) : (i -= 1) { switch (text[i]) { ' ' => continue, ',' => break :found_comma i, else => return null, } } else return null; }; // trim space var i = comma_position - 1; while (i != 0) : (i -= 1) { switch (text[i]) { ' ' => continue, else => { if (!isSymbolChar(text[i])) return null; break :found i + 1; }, } } else return null; }; const value_start = if (!is_index_payload) loc.start else found: { // move back until we find a non identifier character var i = value_end - 1; while (i != 0) : (i -= 1) { if (isSymbolChar(text[i])) continue; switch (text[i]) { ' ', '|', '*' => break :found i + 1, else => return null, } } else return null; }; var index: ?offsets.Loc = null; if (is_index_payload) { index = loc; } else blk: { // move forward until we find a comma const comma_position = found_comma: { var i = value_end; while (i < text.len) : (i += 1) { switch (text[i]) { ' ' => continue, ',' => break :found_comma i, else => break :blk, } } break :blk; }; // trim space const index_start = found_start: { var i = comma_position + 1; while (i < text.len) : (i += 1) { switch (text[i]) { ' ' => continue, else => { if (!isSymbolChar(text[i])) break :blk; break :found_start i; }, } } break :blk; }; // move forward until we find a non identifier character var i = index_start + 1; while (i < text.len) : (i += 1) { if (isSymbolChar(text[i])) continue; index = offsets.Loc{ .start = index_start, .end = i, }; break; } } const start_pipe_position = found: { var i = value_start - 1; while (i != 0) : (i -= 1) { switch (text[i]) { ' ' => continue, '|' => break :found i, else => return null, } } else return null; }; const end_pipe_position = found: { var i: usize = if (index) |index_loc| index_loc.end else value_end; while (i < text.len) : (i += 1) { switch (text[i]) { ' ' => continue, '|' => break :found i + 1, else => return null, } } else return null; }; return CaptureLocs{ .loc = .{ .start = start_pipe_position, .end = end_pipe_position }, .value = .{ .start = value_start, .end = value_end }, .index = index, }; } fn isSymbolChar(char: u8) bool { return std.ascii.isAlphanumeric(char) or char == '_'; }