gasm-sdk/lsp/handlers.go

// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
// SPDX-License-Identifier: BSD-3-Clause

package lsp

import (
	"cmp"
	"fmt"
	"os"
	"path/filepath"
	"regexp"
	"runtime"
	"slices"
	"strconv"
	"strings"
	"unicode"
	"unicode/utf16"
	"unicode/utf8"

	"sourcedock.dev/petrbalvin/gasm-devkit/arch"
	"sourcedock.dev/petrbalvin/gasm-devkit/ast"
	"sourcedock.dev/petrbalvin/gasm-devkit/format"
	"sourcedock.dev/petrbalvin/gasm-devkit/lexer"
	"sourcedock.dev/petrbalvin/gasm-devkit/lint"
	"sourcedock.dev/petrbalvin/gasm-devkit/parser"
	"sourcedock.dev/petrbalvin/gasm-devkit/token"
)

// textflagMacros are the flag names defined by runtime/textflag.h; they are
// highlighted as macros and offered as completions after a TEXT/GLOBL
// directive.
var textflagMacros = map[string]bool{
	"NOPROF": true, "DUPOK": true, "NOSPLIT": true, "RODATA": true,
	"NOPTR": true, "WRAPPER": true, "NEEDCTXT": true, "TLSBSS": true,
	"NOFRAME": true, "REFLECTMETHOD": true, "TOPFRAME": true, "ABIWRAPPER": true,
}

// completion builds the completion list for a document.
func (s *Server) completion(p completionParams) []CompletionItem {
	a := arch.ForArch(arch.FromFilename(uriPath(p.TextDocument.URI)))
	tab := a
	items := []CompletionItem{
		{Label: "TEXT", Kind: ciKeyword, Detail: "define a function"},
		{Label: "DATA", Kind: ciKeyword, Detail: "initialise a data symbol"},
		{Label: "GLOBL", Kind: ciKeyword, Detail: "declare a global symbol"},
	}
	for name := range textflagMacros {
		items = append(items, CompletionItem{Label: name, Kind: ciKeyword, Detail: "textflag.h flag"})
	}
	for name, desc := range map[string]string{
		"FP": "frame pointer (arguments/results)", "SP": "stack pointer",
		"SB": "static base (globals)", "PC": "program counter",
	} {
		items = append(items, CompletionItem{Label: name, Kind: ciConstant, Detail: desc})
	}
	for _, in := range tab.Instructions() {
		items = append(items, CompletionItem{
			Label: in.Name, Kind: ciFunction, Detail: in.Summary, Documentation: in.Summary,
		})
	}
	for _, r := range tab.Registers() {
		kind := ciVariable
		if r.Class == arch.Vector || r.Class == arch.Mask || r.Class == arch.Float || r.Class == arch.VecARM {
			kind = ciClass
		}
		items = append(items, CompletionItem{Label: r.Name, Kind: kind, Detail: r.Desc})
	}
	// Local labels defined in the document.
	if f, _ := parser.Parse("", s.docs[p.TextDocument.URI]); f != nil {
		for _, name := range labelNames(f) {
			items = append(items, CompletionItem{Label: name, Kind: ciModule, Detail: "local label"})
		}
	}
	slices.SortFunc(items, func(a, b CompletionItem) int { return cmp.Compare(a.Label, b.Label) })
	return items
}

// hover returns documentation for the symbol under the cursor.
func (s *Server) hover(p hoverParams) *Hover {
	text := s.docs[p.TextDocument.URI]
	word, rng := wordAt(text, p.Position)
	if word == "" {
		return nil
	}
	a := arch.ForArch(arch.FromFilename(uriPath(p.TextDocument.URI)))

	var md string
	if in, ok := a.Lookup(word); ok {
		md = "**" + in.Name + "**: " + in.Summary
	} else if r, ok := a.Register(word); ok {
		md = "**" + r.Name + "**: " + r.Class.String() + " register. " + r.Desc
	} else if desc, ok := arch.PseudoRegDesc(word); ok {
		md = "**" + strings.ToUpper(word) + "**: pseudo-register. " + desc
	} else if textflagMacros[strings.ToUpper(word)] {
		md = "**" + strings.ToUpper(word) + "**: textflag.h flag"
	} else {
		return nil
	}
	return &Hover{
		Contents: markupContent{Kind: "markdown", Value: md},
		Range:    clientRange(text, rng),
	}
}

// openAST is one open document with its parsed file.
type openAST struct {
	uri  string
	file *ast.File
}

// openASTs parses every open document, in URI order for deterministic
// results.  Parsing is tolerant: a buffer with syntax errors still
// contributes its usable declarations to the workspace scans.
func (s *Server) openASTs() []openAST {
	uris := make([]string, 0, len(s.docs))
	for uri := range s.docs {
		uris = append(uris, uri)
	}
	slices.Sort(uris)
	out := make([]openAST, 0, len(uris))
	for _, uri := range uris {
		if f, _ := parser.Parse(uriPath(uri), s.docs[uri]); f != nil {
			out = append(out, openAST{uri: uri, file: f})
		}
	}
	return out
}

// definition returns the location of the named label or function: a local
// label in the current document wins, then the TEXT functions of every open
// document and indexed workspace file are searched, so a `CALL ·helper(SB)`
// jumps to its definition even when the defining file was never opened.
func (s *Server) definition(p definitionParams) []Location {
	text := s.docs[p.TextDocument.URI]
	word, _ := wordAt(text, p.Position)
	if word == "" {
		return nil
	}
	name := strings.TrimPrefix(word, "\u00B7")

	// The local label definition.
	f, _ := parser.Parse(uriPath(p.TextDocument.URI), text)
	if f != nil {
		for _, d := range f.Decls {
			if t, ok := d.(*ast.Text); ok {
				for _, stmt := range t.Body {
					if lbl, ok := stmt.(*ast.Label); ok {
						if lbl.Name.Text == name || lbl.Name.Text == word {
							return []Location{{
								URI:   p.TextDocument.URI,
								Range: clientRange(text, tokenRange(lbl.Name)),
							}}
						}
					}
				}
			}
		}
	}

	// Function definitions across the workspace.
	for _, u := range s.scanUnits() {
		for _, d := range u.file.Decls {
			t, ok := d.(*ast.Text)
			if !ok || !realSymbol(t.Name) {
				continue
			}
			if t.Name.Name == name {
				return []Location{{URI: u.uri, Range: clientRange(u.text, symRange(t.Name))}}
			}
		}
	}
	return nil
}

// references returns all locations where the symbol under the cursor appears
// across every open document and indexed workspace file.  The current
// document matches labels and any operand name, as before; every other
// source only matches SB-qualified operand references and the definition
// itself, because a bare name is a function-local label whose repeats in
// other files are unrelated.
func (s *Server) references(p referenceParams) []Location {
	text := s.docs[p.TextDocument.URI]
	word, _ := wordAt(text, p.Position)
	if word == "" {
		return nil
	}
	name := strings.TrimPrefix(word, "\u00B7")
	uri := p.TextDocument.URI

	var out []Location
	for _, u := range s.scanUnits() {
		sameDoc := u.uri == uri
		for _, d := range u.file.Decls {
			t, ok := d.(*ast.Text)
			if !ok || !realSymbol(t.Name) {
				continue
			}
			// Include the definition if requested.
			if p.Context.IncludeDeclaration && t.Name.Name == name {
				out = append(out, Location{URI: u.uri, Range: clientRange(u.text, symRange(t.Name))})
			}
			for _, stmt := range t.Body {
				switch st := stmt.(type) {
				case *ast.Label:
					if sameDoc && st.Name.Text == name {
						out = append(out, Location{URI: u.uri, Range: clientRange(u.text, tokenRange(st.Name))})
					}
				case *ast.Instr:
					for _, op := range st.Operands {
						if op.Addr.Sym == nil || op.Addr.Sym.Name != name {
							continue
						}
						if !sameDoc && op.Addr.Sym.Pseudo != "SB" {
							continue
						}
						// The range covers the operand's verbatim
						// identifier, `·`/package prefix included, so a
						// rename replaces the whole spelling.
						out = append(out, Location{
							URI: u.uri,
							Range: clientRange(u.text, Range{
								Start: Position{Line: op.Pos.Line - 1, Character: op.Pos.Column - 1},
								End:   Position{Line: op.Pos.Line - 1, Character: op.Pos.Column - 1 + symIdentLen(op.Addr.Sym)},
							}),
						})
					}
				}
			}
		}
	}
	return out
}

// rename renames a label across every document the references reach.  Each
// edit is grouped under the URI it belongs to: a range collected in another
// file must never be applied to the document that started the rename.
func (s *Server) rename(p renameParams) *WorkspaceEdit {
	text := s.docs[p.TextDocument.URI]
	word, _ := wordAt(text, p.Position)
	if word == "" || word == p.NewName {
		return nil
	}
	refs := s.references(referenceParams{
		TextDocument: p.TextDocument,
		Position:     p.Position,
		Context:      referenceContext{IncludeDeclaration: true},
	})
	if len(refs) == 0 {
		return nil
	}
	edits := make(map[string][]TextEdit, len(refs))
	for _, loc := range refs {
		edits[loc.URI] = append(edits[loc.URI], TextEdit{Range: loc.Range, NewText: p.NewName})
	}
	return &WorkspaceEdit{Changes: edits}
}

// documentFormatting returns a single edit that replaces the whole document
// with its formatted version (canonical indentation and spacing).
func (s *Server) documentFormatting(p documentFormattingParams) []TextEdit {
	text := s.docs[p.TextDocument.URI]
	if text == "" {
		return nil
	}
	formatted := format.Source(text)
	if formatted == text {
		return nil
	}
	lines := strings.Split(text, "\n")
	endLine := len(lines) - 1
	endChar := 0
	if endLine >= 0 {
		endChar = utf16Len(lines[endLine])
	}
	return []TextEdit{{
		Range:   Range{Start: Position{Line: 0, Character: 0}, End: Position{Line: endLine, Character: endChar}},
		NewText: formatted,
	}}
}

// inlayHints returns hints for frame sizes and argument areas.
func (s *Server) inlayHints(p inlayHintParams) []InlayHint {
	text := s.docs[p.TextDocument.URI]
	f, _ := parser.Parse(uriPath(p.TextDocument.URI), text)
	if f == nil {
		return nil
	}
	var out []InlayHint
	for _, d := range f.Decls {
		t, ok := d.(*ast.Text)
		if !ok {
			continue
		}
		// Hint after the args size: show frame size.
		if t.Frame != nil && t.Frame.Imm.HasVal && t.Args != nil && t.Args.Imm.HasVal {
			// Place hint after the args operand using its raw text length.
			line := t.Args.Pos.Line - 1
			col := t.Args.Pos.Column - 1 + runeLen(t.Args.Raw)
			out = append(out, InlayHint{
				Position: Position{Line: line, Character: utf16Column(lineAt(text, line), col)},
				Label:    fmt.Sprintf(" frame=%d", t.Frame.Imm.Val),
				Kind:     inlayHintTypeParameter,
				Tooltip:  fmt.Sprintf("local frame size: %d bytes", t.Frame.Imm.Val),
			})
		}
	}
	return out
}

// codeActions returns quick fixes for diagnostics in the selected range.
func (s *Server) codeActions(p codeActionParams) []CodeAction {
	text := s.docs[p.TextDocument.URI]
	if text == "" {
		return nil
	}
	var actions []CodeAction
	for _, diag := range p.Context.Diagnostics {
		switch diag.Code {
		case "missing-ret":
			// Offer to add RET at the end of the flagged function only: the
			// diagnostic's range covers the TEXT keyword, so a line match
			// picks the function the diagnostic belongs to.
			f, _ := parser.Parse(uriPath(p.TextDocument.URI), text)
			if f == nil {
				continue
			}
			for _, d := range f.Decls {
				t, ok := d.(*ast.Text)
				if !ok {
					continue
				}
				if t.Keyword.Pos.Line-1 != int(diag.Range.Start.Line) {
					continue
				}
				if len(t.Body) == 0 {
					continue
				}
				last := t.Body[len(t.Body)-1]
				line := last.Pos().Line
				// Insert RET after the last statement.
				lines := strings.Split(text, "\n")
				if line-1 < len(lines) {
					insertLine := line // zero-based index for insertion after last stmt
					newLines := make([]string, 0, len(lines)+1)
					newLines = append(newLines, lines[:insertLine]...)
					newLines = append(newLines, "\tRET")
					newLines = append(newLines, lines[insertLine:]...)
					newText := strings.Join(newLines, "\n")
					endLine := len(lines) - 1
					endChar := utf16Len(lines[endLine])
					actions = append(actions, CodeAction{
						Title: "Add RET to " + t.Name.Name,
						Kind:  "quickfix",
						Edit: &WorkspaceEdit{
							Changes: map[string][]TextEdit{p.TextDocument.URI: {
								{Range: Range{Start: Position{Line: 0, Character: 0}, End: Position{Line: endLine, Character: endChar}}, NewText: newText},
							}},
						},
					})
				}
			}
		case "unused-label":
			// Offer to remove the unused label.
			label := strings.TrimPrefix(diag.Message, "label \"")
			label = strings.TrimSuffix(label, "\" is defined but never referenced")
			if label != "" {
				lines := strings.Split(text, "\n")
				for i, line := range lines {
					trimmed := strings.TrimSpace(line)
					if trimmed == label+":" {
						newLines := make([]string, 0, len(lines)-1)
						newLines = append(newLines, lines[:i]...)
						newLines = append(newLines, lines[i+1:]...)
						newText := strings.Join(newLines, "\n")
						endLine := len(lines) - 1
						endChar := utf16Len(lines[endLine])
						actions = append(actions, CodeAction{
							Title: "Remove unused label \"" + label + "\"",
							Kind:  "quickfix",
							Edit: &WorkspaceEdit{
								Changes: map[string][]TextEdit{p.TextDocument.URI: {
									{Range: Range{Start: Position{Line: 0, Character: 0}, End: Position{Line: endLine, Character: endChar}}, NewText: newText},
								}},
							},
						})
						break
					}
				}
			}
		case "missing-textflag-include":
			// Offer to #include the header that defines the flag macros:
			// after the last existing include, or at the top of the file
			// when there is none.
			insertLine := 0
			lines := strings.Split(text, "\n")
			for i, line := range lines {
				if includeRe.MatchString(line) {
					insertLine = i + 1
				}
			}
			actions = append(actions, CodeAction{
				Title: `Add #include "textflag.h"`,
				Kind:  "quickfix",
				Edit: &WorkspaceEdit{
					Changes: map[string][]TextEdit{p.TextDocument.URI: {{
						Range:   Range{Start: Position{Line: insertLine, Character: 0}, End: Position{Line: insertLine, Character: 0}},
						NewText: "#include \"textflag.h\"\n",
					}}},
				},
			})
		case "abi-argsize":
			// Offer to set the TEXT argument area to the size the // func
			// signature implies.  The diagnostic's range covers the TEXT
			// keyword, so a line match picks the function it belongs to.
			f, _ := parser.Parse(uriPath(p.TextDocument.URI), text)
			if f == nil {
				continue
			}
			for _, d := range f.Decls {
				t, ok := d.(*ast.Text)
				if !ok || t.Keyword.Pos.Line-1 != int(diag.Range.Start.Line) {
					continue
				}
				want, ok := lint.ExpectedArgSize(t.Doc)
				if !ok || t.Args == nil || !t.Args.Imm.HasVal || t.Args.Imm.Val == want {
					continue
				}
				// The parser records the argument area with its leading
				// minus, so the replacement spans the whole `-16` shape.
				rng := clientRange(text, Range{
					Start: Position{Line: t.Args.Pos.Line - 1, Character: t.Args.Pos.Column - 1},
					End:   Position{Line: t.Args.Pos.Line - 1, Character: t.Args.Pos.Column - 1 + runeLen(t.Args.Raw)},
				})
				actions = append(actions, CodeAction{
					Title: fmt.Sprintf("Set arg size to %d", want),
					Kind:  "quickfix",
					Edit: &WorkspaceEdit{
						Changes: map[string][]TextEdit{p.TextDocument.URI: {{
							Range:   rng,
							NewText: "-" + strconv.FormatInt(want, 10),
						}}},
					},
				})
			}
		}
	}
	return actions
}

// signatureHelp returns signature information for a CALL instruction.
func (s *Server) signatureHelp(p signatureHelpParams) *SignatureHelp {
	text := s.docs[p.TextDocument.URI]
	if text == "" {
		return nil
	}
	// Check if the cursor is on a CALL/BL line.
	lines := strings.Split(text, "\n")
	if p.Position.Line >= len(lines) {
		return nil
	}
	line := lines[p.Position.Line]
	trimmed := strings.TrimSpace(line)
	upper := strings.ToUpper(trimmed)
	if !strings.HasPrefix(upper, "CALL ") && !strings.HasPrefix(upper, "BL ") {
		return nil
	}
	// Extract the function name from the CALL operand.
	parts := strings.Fields(trimmed)
	if len(parts) < 2 {
		return nil
	}
	funcName := parts[1]
	// Strip (SB) suffix.
	funcName = strings.TrimSuffix(funcName, "(SB)")

	// Look up the function in the document.
	f, _ := parser.Parse(uriPath(p.TextDocument.URI), text)
	if f == nil {
		return nil
	}
	// Strip middle-dot prefix if present (Plan 9 package separator).
	bareName := strings.TrimPrefix(funcName, "·")
	for _, d := range f.Decls {
		t, ok := d.(*ast.Text)
		if !ok {
			continue
		}
		if t.Name.Name == bareName || t.Name.Name == funcName || strings.HasSuffix(t.Name.Name, "·"+bareName) {
			label := "TEXT " + t.Name.Name + "(SB)"
			sig := SignatureInformation{
				Label: label,
			}
			if t.Doc != "" {
				sig.Documentation = t.Doc
			}
			// Add frame and args as parameters.
			frame := int64(0)
			if t.Frame != nil && t.Frame.Imm.HasVal {
				frame = t.Frame.Imm.Val
			}
			args := int64(0)
			if t.Args != nil && t.Args.Imm.HasVal {
				args = t.Args.Imm.Val
			}
			sig.Parameters = []ParameterInformation{
				{Label: fmt.Sprintf("$%d", frame)},
				{Label: fmt.Sprintf("-%d", args)},
			}
			return &SignatureHelp{Signatures: []SignatureInformation{sig}}
		}
	}
	return nil
}

// documentHighlights returns all occurrences of the symbol under the cursor.
func (s *Server) documentHighlights(p documentHighlightParams) []DocumentHighlight {
	text := s.docs[p.TextDocument.URI]
	word, _ := wordAt(text, p.Position)
	if word == "" {
		return nil
	}
	// Symbol names are stored without the middle dot, while the word under
	// the cursor keeps it; trim once and compare the trimmed form, the way
	// references and definition do.
	name := strings.TrimPrefix(word, "\u00B7")
	f, errs := parser.Parse(uriPath(p.TextDocument.URI), text)
	if f == nil || len(errs) > 0 {
		return nil
	}
	var out []DocumentHighlight
	for _, d := range f.Decls {
		t, ok := d.(*ast.Text)
		if !ok {
			continue
		}
		// Highlight the definition.
		if t.Name != nil && t.Name.Name == name {
			out = append(out, DocumentHighlight{
				Range: clientRange(text, symRange(t.Name)),
				Kind:  highlightWrite,
			})
		}
		for _, stmt := range t.Body {
			switch st := stmt.(type) {
			case *ast.Label:
				if st.Name.Text == name {
					out = append(out, DocumentHighlight{
						Range: clientRange(text, tokenRange(st.Name)),
						Kind:  highlightWrite,
					})
				}
			case *ast.Instr:
				for _, op := range st.Operands {
					if op.Addr.Sym != nil && op.Addr.Sym.Name == name {
						out = append(out, DocumentHighlight{
							Range: clientRange(text, Range{
								Start: Position{Line: op.Pos.Line - 1, Character: op.Pos.Column - 1},
								End:   Position{Line: op.Pos.Line - 1, Character: op.Pos.Column - 1 + symIdentLen(op.Addr.Sym)},
							}),
							Kind: highlightRead,
						})
					}
				}
			}
		}
	}
	return out
}

// workspaceSymbols searches all open documents and indexed workspace files
// for symbols matching the query.
func (s *Server) workspaceSymbols(p workspaceSymbolParams) []WorkspaceSymbol {
	if p.Query == "" {
		return nil
	}
	query := strings.ToLower(p.Query)
	var out []WorkspaceSymbol
	for _, u := range s.scanUnits() {
		for _, d := range u.file.Decls {
			switch dd := d.(type) {
			case *ast.Text:
				// A malformed TEXT line is kept in the tree under a "?"
				// placeholder; it is not a symbol, and skipping it keeps
				// the answer serving the rest of a mid-edit buffer.
				if !realSymbol(dd.Name) {
					continue
				}
				if strings.Contains(strings.ToLower(dd.Name.Name), query) {
					out = append(out, WorkspaceSymbol{
						Name:     dd.Name.Name,
						Kind:     symFunction,
						Location: Location{URI: u.uri, Range: clientRange(u.text, symRange(dd.Name))},
					})
				}
			case *ast.Globl:
				if dd.Name != nil && strings.Contains(strings.ToLower(dd.Name.Name), query) {
					out = append(out, WorkspaceSymbol{
						Name:     dd.Name.Name,
						Kind:     symConstant,
						Location: Location{URI: u.uri, Range: clientRange(u.text, symRange(dd.Name))},
					})
				}
			case *ast.Data:
				if dd.Name != nil && strings.Contains(strings.ToLower(dd.Name.Name), query) {
					out = append(out, WorkspaceSymbol{
						Name:     dd.Name.Name,
						Kind:     symConstant,
						Location: Location{URI: u.uri, Range: clientRange(u.text, symRange(dd.Name))},
					})
				}
			}
		}
	}
	return out
}

// documentSymbols returns functions and their labels, plus global symbols.
func (s *Server) documentSymbols(p documentSymbolParams) []DocumentSymbol {
	text := s.docs[p.TextDocument.URI]
	f, _ := parser.Parse(uriPath(p.TextDocument.URI), text)
	if f == nil {
		return nil
	}
	var out []DocumentSymbol
	for _, d := range f.Decls {
		switch dd := d.(type) {
		case *ast.Text:
			// As in workspaceSymbols: a "?" placeholder is not a symbol,
			// and the remaining declarations are still listed.
			if !realSymbol(dd.Name) {
				continue
			}
			sym := DocumentSymbol{
				Name:           dd.Name.Name,
				Detail:         "TEXT " + strings.Join(dd.Flags, " "),
				Kind:           symFunction,
				Range:          clientRange(text, textRange(dd)),
				SelectionRange: clientRange(text, symRange(dd.Name)),
			}
			for _, st := range dd.Body {
				if l, ok := st.(*ast.Label); ok {
					sym.Children = append(sym.Children, DocumentSymbol{
						Name:           l.Name.Text,
						Kind:           symVariable,
						Range:          clientRange(text, tokenRange(l.Name)),
						SelectionRange: clientRange(text, tokenRange(l.Name)),
					})
				}
			}
			out = append(out, sym)
		case *ast.Globl:
			if dd.Name == nil {
				continue
			}
			out = append(out, DocumentSymbol{
				Name: dd.Name.Name, Detail: "GLOBL", Kind: symConstant,
				Range: clientRange(text, symRange(dd.Name)), SelectionRange: clientRange(text, symRange(dd.Name)),
			})
		case *ast.Data:
			if dd.Name == nil {
				continue
			}
			out = append(out, DocumentSymbol{
				Name: dd.Name.Name, Detail: "DATA", Kind: symConstant,
				Range: clientRange(text, symRange(dd.Name)), SelectionRange: clientRange(text, symRange(dd.Name)),
			})
		}
	}
	return out
}

// semTok is one classified token before delta encoding.
type semTok struct {
	line, char, length, typ int
}

// semanticTokens encodes syntax highlighting as LSP semantic tokens.
func (s *Server) semanticTokens(p semanticTokensParams) SemanticTokens {
	text := s.docs[p.TextDocument.URI]
	a := arch.ForArch(arch.FromFilename(uriPath(p.TextDocument.URI)))
	f, _ := parser.Parse("", text)
	labels := map[string]bool{}
	for _, name := range labelNames(f) {
		labels[name] = true
	}

	toks := lexer.Tokenize(text)
	lines := groupLines(toks)
	srcLines := strings.Split(text, "\n")

	var encoded []semTok
	for _, line := range lines {
		for _, st := range classifyLine(line, a, labels) {
			// Columns cross the protocol boundary in UTF-16 code units.
			if st.line >= 0 && st.line < len(srcLines) {
				st.char = utf16Column(srcLines[st.line], st.char)
			}
			encoded = append(encoded, st)
		}
	}

	return SemanticTokens{Data: deltaEncode(encoded)}
}

// classifyLine assigns a semantic token type to each significant token on a line.
func classifyLine(line []token.Token, a *arch.Table, labels map[string]bool) []semTok {
	if len(line) == 0 {
		return nil
	}
	var out []semTok
	first := firstSignificant(line)
	if first < 0 {
		return nil
	}

	isDirective := line[first].Kind == token.Ident &&
		(line[first].Text == "TEXT" || line[first].Text == "DATA" || line[first].Text == "GLOBL")
	isLabel := line[first].Kind == token.Ident && first+1 < len(line) &&
		line[first+1].Kind == token.Colon
	isInstr := !isDirective && !isLabel && line[first].Kind == token.Ident

	mnemonicDone := false
	for i, t := range line {
		typ := -1
		switch t.Kind {
		case token.Comment:
			typ = stComment
		case token.Number:
			typ = stNumber
		case token.String, token.Rune:
			typ = stString
		case token.Hash:
			typ = stMacro
		case token.Ident:
			typ = classifyIdent(i, first, t.Text, a, labels, isDirective, isLabel, isInstr, &mnemonicDone)
		case token.Colon, token.Comma, token.LParen, token.RParen,
			token.Plus, token.Minus, token.Star, token.Slash, token.Dollar,
			token.LAngle, token.RAngle, token.LShift, token.RShift, token.Arrow,
			token.At, token.Pipe:
			typ = stOperator
		}
		if typ >= 0 {
			out = append(out, semTok{
				line:   t.Pos.Line - 1,
				char:   t.Pos.Column - 1,
				length: utf16Len(t.Text),
				typ:    typ,
			})
		}
	}
	return out
}

// classifyIdent decides the semantic type of an identifier token.
func classifyIdent(i, first int, text string, a *arch.Table, labels map[string]bool, isDirective, isLabel, isInstr bool, mnemonicDone *bool) int {

	upper := strings.ToUpper(text)
	switch {
	case isDirective && i == first:
		return stKeyword
	case isDirective && textflagMacros[upper]:
		return stMacro
	case isLabel && i == first:
		return stNamespace
	case arch.IsPseudoReg(text):
		return stProperty
	case labels[text]:
		return stNamespace
	}
	if r, ok := a.Register(text); ok {
		switch r.Class {
		case arch.Vector, arch.Mask, arch.Float, arch.VecARM:
			return stType
		default:
			return stVariable
		}
	}
	if isInstr && i == first && !*mnemonicDone {
		*mnemonicDone = true
		return stFunction
	}
	// Argument/symbol names and anything else.
	return stVariable
}

// deltaEncode converts absolute token positions to the LSP relative encoding.
func deltaEncode(toks []semTok) []int {
	data := make([]int, 0, len(toks)*5)
	prevLine, prevChar := 0, 0
	for _, t := range toks {
		dLine := t.line - prevLine
		dChar := t.char
		if dLine == 0 {
			dChar = t.char - prevChar
		}
		data = append(data, dLine, dChar, t.length, t.typ, 0)
		prevLine, prevChar = t.line, t.char
	}
	return data
}

// --- shared helpers ---------------------------------------------------------

// wordAt extracts the identifier surrounding pos and its range.  The
// incoming character offset is UTF-16 code units, as LSP defines it, and is
// converted to the rune index the scanning works in.
func wordAt(text string, pos Position) (string, Range) {
	lines := strings.Split(text, "\n")
	if pos.Line < 0 || pos.Line >= len(lines) {
		return "", Range{}
	}
	runes := []rune(lines[pos.Line])
	col := runeColumn(lines[pos.Line], pos.Character)
	if col < 0 || col > len(runes) {
		return "", Range{}
	}
	isWord := func(r rune) bool {
		return r == '_' || r == '\u00B7' || unicode.IsLetter(r) || unicode.IsDigit(r)
	}
	start, end := col, col
	for start > 0 && isWord(runes[start-1]) {
		start--
	}
	for end < len(runes) && isWord(runes[end]) {
		end++
	}
	if start == end {
		return "", Range{}
	}
	rng := Range{
		Start: Position{Line: pos.Line, Character: start},
		End:   Position{Line: pos.Line, Character: end},
	}
	return string(runes[start:end]), rng
}

// labelNames collects every label defined in a file.
func labelNames(f *ast.File) []string {
	if f == nil {
		return nil
	}
	seen := map[string]bool{}
	var out []string
	collect := func(body []ast.Stmt) {
		for _, st := range body {
			if l, ok := st.(*ast.Label); ok && !seen[l.Name.Text] {
				seen[l.Name.Text] = true
				out = append(out, l.Name.Text)
			}
		}
	}
	for _, d := range f.Decls {
		if t, ok := d.(*ast.Text); ok {
			collect(t.Body)
		}
	}
	collect(f.Orphans)
	return out
}

// groupLines splits a token stream into lines, keeping Newline boundaries but
// dropping the Newline and EOF tokens themselves.
func groupLines(toks []token.Token) [][]token.Token {
	var lines [][]token.Token
	var cur []token.Token
	for _, t := range toks {
		if t.Kind == token.EOF {
			break
		}
		if t.Kind == token.Newline {
			lines = append(lines, cur)
			cur = nil
			continue
		}
		cur = append(cur, t)
	}
	if len(cur) > 0 {
		lines = append(lines, cur)
	}
	return lines
}

func firstSignificant(line []token.Token) int {
	for i, t := range line {
		if t.Kind != token.Comment {
			return i
		}
	}
	return -1
}

func runeLen(s string) int { return len([]rune(s)) }

// utf16Len returns the length of s in UTF-16 code units, the unit LSP
// positions count: an astral rune (an emoji in a comment) is two of them.
func utf16Len(s string) int {
	n := 0
	for _, r := range s {
		n += utf16.RuneLen(r)
	}
	return n
}

// utf16Column converts a rune-based column on line into a UTF-16 code-unit
// offset.  Rune columns are the lexer's convention; code units are the
// protocol's, and the two diverge once an astral rune precedes the column.
func utf16Column(line string, col int) int {
	if col <= 0 {
		return 0
	}
	units := 0
	seen := 0
	for _, r := range line {
		if seen >= col {
			break
		}
		units += utf16.RuneLen(r)
		seen++
	}
	return units
}

// runeColumn converts a UTF-16 code-unit offset on line into a rune column,
// the inverse of utf16Column, applied to positions arriving from the client.
func runeColumn(line string, units int) int {
	if units <= 0 {
		return 0
	}
	col := 0
	u := 0
	for _, r := range line {
		if u >= units {
			break
		}
		u += utf16.RuneLen(r)
		col++
	}
	return col
}

// lineAt returns the n-th zero-based line of text, or "" when out of range.
func lineAt(text string, n int) string {
	if n < 0 {
		return ""
	}
	lines := strings.Split(text, "\n")
	if n >= len(lines) {
		return ""
	}
	return lines[n]
}

// clientRange re-encodes a rune-based range (the columns the lexer, parser
// and the helpers above produce) in the UTF-16 code units LSP mandates.
// Every range leaving the server passes through here.
func clientRange(text string, r Range) Range {
	return Range{
		Start: Position{Line: r.Start.Line, Character: utf16Column(lineAt(text, r.Start.Line), r.Start.Character)},
		End:   Position{Line: r.End.Line, Character: utf16Column(lineAt(text, r.End.Line), r.End.Character)},
	}
}

// symIdentLen returns the rune length of a symbol's identifier as written:
// the verbatim spelling up to the ABI marker, offset or pseudo-register
// group, so `pkg·name<ABIInternal>(SB)` counts the package prefix and the
// middle dot.  A range built from it covers the whole token a rename
// replaces; the stripped Name alone would stop one character short.
func symIdentLen(sym *ast.Symbol) int {
	if i := strings.IndexAny(sym.Raw, "<+-("); i >= 0 {
		return runeLen(sym.Raw[:i])
	}
	return runeLen(sym.Raw)
}

// symRange builds a range covering a symbol as written, prefix included.
func symRange(sym *ast.Symbol) Range {
	start := Position{Line: sym.Pos.Line - 1, Character: sym.Pos.Column - 1}
	end := start
	end.Character += symIdentLen(sym)
	return Range{Start: start, End: end}
}

// realSymbol reports whether sym names something a client can act on.  The
// parser keeps a malformed TEXT in the tree under a "?" placeholder so the
// rest of the buffer stays servable; that placeholder is not a symbol.
func realSymbol(sym *ast.Symbol) bool {
	return sym != nil && sym.Name != "?"
}

// tokenRange builds a range covering one token.
func tokenRange(t token.Token) Range {
	return Range{
		Start: Position{Line: t.Pos.Line - 1, Character: t.Pos.Column - 1},
		End:   Position{Line: t.End.Line - 1, Character: t.End.Column - 1},
	}
}

// textRange spans a TEXT function from its keyword to the end of its body.
func textRange(t *ast.Text) Range {
	start := Position{Line: t.Keyword.Pos.Line - 1, Character: t.Keyword.Pos.Column - 1}
	end := start
	end.Character += runeLen(t.Keyword.Text)
	if n := len(t.Body); n > 0 {
		last := t.Body[n-1]
		if in, ok := last.(*ast.Instr); ok {
			end = Position{Line: in.Mnemonic.End.Line - 1, Character: in.Mnemonic.End.Column - 1}
		} else {
			lp := last.Pos()
			end = Position{Line: lp.Line - 1, Character: lp.Column - 1}
		}
	}
	return Range{Start: start, End: end}
}

// diagnosticsFor computes the LSP diagnostics of one document; the push
// (publishDiagnostics) and pull (textDocument/diagnostic) paths share it.
// Parse errors surface as error-severity diagnostics so a malformed line is
// visible in the editor instead of only breaking derived features.
func (s *Server) diagnosticsFor(uri string) []Diagnostic {
	text := s.docs[uri]
	f, errs := parser.Parse(uriPath(uri), text)
	cfg := lint.Config{Arch: arch.FromFilename(uriPath(uri))}
	diags := lint.File(f, cfg)

	out := make([]Diagnostic, 0, len(diags)+len(errs))
	for _, e := range errs {
		pos := token.Position{Line: 1, Column: 1}
		if pe, ok := e.(parser.Error); ok && pe.Pos.IsValid() {
			pos = pe.Pos
		}
		out = append(out, Diagnostic{
			Range:    clientRange(text, toRange(pos.Line, pos.Column, token.Position{})),
			Severity: sevError,
			Code:     "syntax",
			Source:   "gasm",
			Message:  e.Error(),
		})
	}
	for _, d := range diags {
		out = append(out, Diagnostic{
			Range:    clientRange(text, toRange(d.Pos.Line, d.Pos.Column, d.End)),
			Severity: lintSeverity(d.Severity),
			Code:     d.Code,
			Source:   "gasm",
			Message:  d.Message,
		})
	}
	return out
}

// includeRe matches a #include directive's quoted path.
var includeRe = regexp.MustCompile(`^\s*#include\s+"([^"]+)"`)

// documentLinks links each #include path to the header file it resolves to:
// first relative to the document's directory, then in $GOROOT/pkg/include
// (where textflag.h and friends live).  Unresolvable paths get no link.
func (s *Server) documentLinks(uri string) []DocumentLink {
	text := s.docs[uri]
	if text == "" {
		return nil
	}
	docDir := filepath.Dir(uriPath(uri))
	goroot := runtime.GOROOT()

	var out []DocumentLink
	lineNo := -1
	for line := range strings.SplitSeq(text, "\n") {
		lineNo++
		m := includeRe.FindStringSubmatch(line)
		if m == nil {
			continue
		}
		target := resolveInclude(docDir, goroot, m[1])
		if target == "" {
			continue
		}
		// strings.Index is a byte offset; columns are runes.
		start := utf8.RuneCountInString(line[:strings.Index(line, "\"")])
		out = append(out, DocumentLink{
			Range: clientRange(text, Range{
				Start: Position{Line: lineNo, Character: start},
				End:   Position{Line: lineNo, Character: start + runeLen(m[1]) + 2},
			}),
			Target: "file://" + target,
		})
	}
	return out
}

// resolveInclude returns the existing file a #include path refers to, or "".
func resolveInclude(docDir, goroot, path string) string {
	for _, dir := range []string{docDir, filepath.Join(goroot, "pkg", "include")} {
		cand := filepath.Join(dir, path)
		if _, err := os.Stat(cand); err == nil {
			return cand
		}
	}
	return ""
}

// foldingRanges marks each TEXT function's body as a collapsible region.
func (s *Server) foldingRanges(uri string) []FoldingRange {
	text := s.docs[uri]
	f, _ := parser.Parse(uriPath(uri), text)
	if f == nil {
		return nil
	}
	var out []FoldingRange
	for _, d := range f.Decls {
		t, ok := d.(*ast.Text)
		if !ok || len(t.Body) == 0 {
			continue
		}
		start := t.Keyword.Pos.Line - 1
		end := t.Body[len(t.Body)-1].Pos().Line - 1
		if end <= start {
			continue
		}
		out = append(out, FoldingRange{StartLine: start, EndLine: end, Kind: "region"})
	}
	return out
}