feat: add size level aggregation

cmd/internal/flags/flags.go

@@ -36,6 +36,9 @@ var CheckLinkArgs bool
 var CheckLLFiles bool
 var GenLLFiles bool
 var ForceEspClang bool
+var SizeReport bool
+var SizeFormat string
+var SizeLevel string
 
 func AddCommonFlags(fs *flag.FlagSet) {
 	fs.BoolVar(&Verbose, "v", false, "Verbose output")
@@ -51,6 +54,10 @@ func AddBuildFlags(fs *flag.FlagSet) {
 		fs.BoolVar(&GenLLFiles, "gen-llfiles", false, "generate .ll files for pkg export")
 		fs.BoolVar(&ForceEspClang, "force-espclang", false, "force to use esp-clang")
 	}
+
+	fs.BoolVar(&SizeReport, "size", false, "Print size report after build")
+	fs.StringVar(&SizeFormat, "size:format", "", "Size report format (text,json)")
+	fs.StringVar(&SizeLevel, "size:level", "", "Size report aggregation level (full,module,package)")
 }
 
 func AddBuildModeFlags(fs *flag.FlagSet) {
@@ -79,6 +86,15 @@ func UpdateConfig(conf *build.Config) error {
 	conf.Target = Target
 	conf.Port = Port
 	conf.BaudRate = BaudRate
+	if SizeReport || SizeFormat != "" || SizeLevel != "" {
+		conf.SizeReport = true
+		if SizeFormat != "" {
+			conf.SizeFormat = SizeFormat
+		}
+		if SizeLevel != "" {
+			conf.SizeLevel = SizeLevel
+		}
+	}
 
 	switch conf.Mode {
 	case build.ModeBuild:

doc/size-report.md

@@ -0,0 +1,58 @@
+# Size Report Options
+
+The `llgo build -size` flag emits a TinyGo-style table showing how much code,
+rodata, data, and BSS each component contributes to the final binary. This
+document captures the parsing strategy and new aggregation controls.
+
+## Parsing Strategy
+
+- We invoke `llvm-readelf --all <binary>` and parse the textual output with an
+  indentation-sensitive state machine (no JSON). Only the `Sections` and
+  `Symbols` blocks are inspected.
+- Section metadata records the index, address, size, name, and segment. Each
+  section is classified into text/rodata/data/bss buckets.
+- Symbols are attached to their containing sections with start addresses. By
+  sorting symbols and walking their ranges, we compute byte spans that can be
+  attributed to packages/modules.
+- Sections with no symbols fall back to `(unknown <section>)`, and gaps become
+  `(padding <section>)` entries so totals still add up.
+
+## Aggregation Levels
+
+`-size:level` controls how symbol names are grouped prior to reporting:
+
+| Level     | Behavior                                                                 |
+|-----------|---------------------------------------------------------------------------|
+| `full`    | Keeps the raw owner from the symbol name (previous behavior).             |
+| `package` | Uses the list of packages built in `build.Do` and groups by `pkg.ID`.     |
+| `module`* | Default. Groups by `pkg.Module.Path` (or `pkg.ID` if the module is nil).   |
+
+Matching is performed by checking whether the demangled symbol name begins with
+`pkg.ID + "."`. Symbols that do not match any package and contain `llgo` are
+bucketed into `llgo-stubs`; other unmatched entries keep their original owner
+names so we can inspect them later.
+
+Examples:
+
+```sh
+llgo build -size .                     # module-level aggregation (default)
+llgo build -size -size:level=package . # collapse by package ID
+llgo build -size -size:level=full .    # show raw symbol owners
+llgo build -size -size:format=json .   # JSON output (works with all levels)
+```
+
+## Validation
+
+1. Unit tests: `go test ./internal/build -run TestParseReadelfOutput -count=1`.
+2. Real binary test:
+   ```sh
+   cd cl/_testgo/rewrite
+   ../../../llgo.sh build .
+   LLGO_SIZE_REPORT_BIN=$(pwd)/rewrite \
+     go test ./internal/build -run TestParseReadelfRealBinary -count=1
+   ```
+3. Manual smoke test: `../../../llgo.sh build -size -size:level=module .` (or
+   `package`/`full` as desired).
+
+The parser works across Mach-O and ELF targets as long as `llvm-readelf` is in
+`PATH`.

internal/build/build.go

@@ -132,6 +132,9 @@ type Config struct {
 	CheckLinkArgs bool // check linkargs valid
 	ForceEspClang bool // force to use esp-clang
 	Tags          string
+	SizeReport    bool   // print size report after successful build
+	SizeFormat    string // size report format: text,json
+	SizeLevel     string // size aggregation level: full,module,package
 	// GlobalRewrites specifies compile-time overrides for global string variables.
 	// Keys are fully qualified package paths (e.g. "main" or "github.com/user/pkg").
 	// Each Rewrites entry maps variable names to replacement string values. Only
@@ -205,6 +208,15 @@ func Do(args []string, conf *Config) ([]Package, error) {
 	if conf.BuildMode == "" {
 		conf.BuildMode = BuildModeExe
 	}
+	if conf.SizeReport && conf.SizeFormat == "" {
+		conf.SizeFormat = "text"
+	}
+	if conf.SizeReport && conf.SizeLevel == "" {
+		conf.SizeLevel = "module"
+	}
+	if err := ensureSizeReporting(conf); err != nil {
+		return nil, err
+	}
 	// Handle crosscompile configuration first to set correct GOOS/GOARCH
 	forceEspClang := conf.ForceEspClang || conf.Target != ""
 	export, err := crosscompile.Use(conf.Goos, conf.Goarch, conf.Target, IsWasiThreadsEnabled(), forceEspClang)
@@ -380,6 +392,11 @@ func Do(args []string, conf *Config) ([]Package, error) {
 			if err != nil {
 				return nil, err
 			}
+			if conf.Mode == ModeBuild && conf.SizeReport {
+				if err := reportBinarySize(outFmts.Out, conf.SizeFormat, conf.SizeLevel, allPkgs); err != nil {
+					return nil, err
+				}
+			}
 
 			// Generate C headers for c-archive and c-shared modes before linking
 			if ctx.buildConf.BuildMode == BuildModeCArchive || ctx.buildConf.BuildMode == BuildModeCShared {

internal/build/resolver.go

@@ -0,0 +1,109 @@
+package build
+
+import "strings"
+
+// nameResolver maps symbol names to aggregation buckets based on the requested level.
+type nameResolver struct {
+	level      string
+	pkgs       []Package
+	moduleMap  map[string]string
+	packageMap map[string]string
+}
+
+func newNameResolver(level string, pkgs []Package) *nameResolver {
+	lvl := strings.ToLower(strings.TrimSpace(level))
+	if lvl == "" {
+		lvl = "module"
+	}
+	return &nameResolver{
+		level:      lvl,
+		pkgs:       pkgs,
+		moduleMap:  make(map[string]string),
+		packageMap: make(map[string]string),
+	}
+}
+
+func (r *nameResolver) resolve(sym string) string {
+	base := moduleNameFromSymbol(sym)
+	symbol := trimSymbolForMatch(sym)
+	switch r.level {
+	case "full":
+		return base
+	case "package":
+		if pkg := r.matchPackage(symbol); pkg != "" {
+			return pkg
+		}
+	case "module":
+		if mod := r.matchModule(symbol); mod != "" {
+			return mod
+		}
+	}
+	if strings.Contains(symbol, "llgo") {
+		return "llgo-stubs"
+	}
+	return base
+}
+
+func (r *nameResolver) matchPackage(symbol string) string {
+	if symbol == "" {
+		return ""
+	}
+	if cached := r.packageMap[symbol]; cached != "" {
+		return cached
+	}
+	for _, pkg := range r.pkgs {
+		if pkg == nil || pkg.Package == nil {
+			continue
+		}
+		id := pkg.PkgPath
+		if id == "" {
+			continue
+		}
+		if strings.HasPrefix(symbol, id+".") {
+			r.packageMap[symbol] = id
+			return id
+		}
+	}
+	return ""
+}
+
+func (r *nameResolver) matchModule(symbol string) string {
+	if symbol == "" {
+		return ""
+	}
+	if cached := r.moduleMap[symbol]; cached != "" {
+		return cached
+	}
+	for _, pkg := range r.pkgs {
+		if pkg == nil || pkg.Package == nil {
+			continue
+		}
+		path := pkg.PkgPath
+		if path == "" {
+			continue
+		}
+		if strings.HasPrefix(symbol, path+".") {
+			mod := path
+			if pkg.Module != nil && pkg.Module.Path != "" {
+				mod = pkg.Module.Path
+			}
+			r.moduleMap[symbol] = mod
+			return mod
+		}
+	}
+	return ""
+}
+
+func trimSymbolForMatch(sym string) string {
+	name := strings.TrimSpace(sym)
+	for len(name) > 0 && (name[0] == '_' || name[0] == '.') {
+		name = name[1:]
+	}
+	if idx := strings.Index(name, " "); idx >= 0 {
+		name = name[:idx]
+	}
+	if idx := strings.Index(name, "@"); idx >= 0 {
+		name = name[:idx]
+	}
+	return name
+}

internal/build/size_report.go

@@ -0,0 +1,546 @@
+package build
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"sort"
+	"strconv"
+	"strings"
+)
+
+type sectionKind int
+
+const (
+	sectionUnknown sectionKind = iota
+	sectionText
+	sectionROData
+	sectionData
+	sectionBSS
+)
+
+type sectionInfo struct {
+	Index   int
+	Name    string
+	Segment string
+	Address uint64
+	Size    uint64
+	Kind    sectionKind
+}
+
+type symbolInfo struct {
+	Name         string
+	SectionIndex int
+	Address      uint64
+}
+
+type readelfData struct {
+	sections map[int]*sectionInfo
+	symbols  map[int][]symbolInfo
+}
+
+type moduleSize struct {
+	Name   string
+	Code   uint64
+	ROData uint64
+	Data   uint64
+	BSS    uint64
+}
+
+func (m *moduleSize) Flash() uint64 {
+	return m.Code + m.ROData + m.Data
+}
+
+func (m *moduleSize) RAM() uint64 {
+	return m.Data + m.BSS
+}
+
+type sizeReport struct {
+	Binary  string
+	Modules map[string]*moduleSize
+	Total   moduleSize
+}
+
+func (r *sizeReport) module(name string) *moduleSize {
+	if name == "" {
+		name = "(anonymous)"
+	}
+	if r.Modules == nil {
+		r.Modules = make(map[string]*moduleSize)
+	}
+	m, ok := r.Modules[name]
+	if !ok {
+		m = &moduleSize{Name: name}
+		r.Modules[name] = m
+	}
+	return m
+}
+
+func (r *sizeReport) add(name string, kind sectionKind, size uint64) {
+	if size == 0 {
+		return
+	}
+	m := r.module(name)
+	switch kind {
+	case sectionText:
+		m.Code += size
+		r.Total.Code += size
+	case sectionROData:
+		m.ROData += size
+		r.Total.ROData += size
+	case sectionData:
+		m.Data += size
+		r.Total.Data += size
+	case sectionBSS:
+		m.BSS += size
+		r.Total.BSS += size
+	}
+}
+
+func reportBinarySize(path, format, level string, pkgs []Package) error {
+	report, err := collectBinarySize(path, pkgs, level)
+	if err != nil {
+		return err
+	}
+	switch format {
+	case "", "text":
+		printTextReport(os.Stdout, report)
+	case "json":
+		return emitJSONReport(os.Stdout, report)
+	default:
+		return fmt.Errorf("unknown size format %q (valid: text,json)", format)
+	}
+	return nil
+}
+
+func collectBinarySize(path string, pkgs []Package, level string) (*sizeReport, error) {
+	cmd := exec.Command("llvm-readelf", "--all", path)
+	var stderr bytes.Buffer
+	cmd.Stderr = &stderr
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return nil, fmt.Errorf("llvm-readelf stdout: %w", err)
+	}
+	if err := cmd.Start(); err != nil {
+		return nil, fmt.Errorf("failed to execute llvm-readelf: %w", err)
+	}
+	parsed, parseErr := parseReadelfOutput(stdout)
+	closeErr := stdout.Close()
+	waitErr := cmd.Wait()
+	if parseErr != nil {
+		if waitErr != nil {
+			return nil, fmt.Errorf("llvm-readelf failed: %w\n%s", waitErr, stderr.String())
+		}
+		return nil, parseErr
+	}
+	if closeErr != nil {
+		return nil, closeErr
+	}
+	if waitErr != nil {
+		return nil, fmt.Errorf("llvm-readelf failed: %w\n%s", waitErr, stderr.String())
+	}
+	report := buildSizeReport(path, parsed, pkgs, level)
+	if report == nil || len(report.Modules) == 0 {
+		return nil, fmt.Errorf("size report: no allocatable sections found in %s", path)
+	}
+	return report, nil
+}
+
+func parseReadelfOutput(r io.Reader) (*readelfData, error) {
+	scanner := bufio.NewScanner(r)
+	scanner.Buffer(make([]byte, 0, 64*1024), 64*1024*1024)
+
+	type ctxKind int
+	const (
+		ctxRoot ctxKind = iota
+		ctxSections
+		ctxSection
+		ctxSymbols
+		ctxSymbol
+	)
+
+	type ctx struct {
+		kind   ctxKind
+		indent int
+	}
+
+	stack := []ctx{{kind: ctxRoot, indent: -1}}
+	push := func(kind ctxKind, indent int) {
+		stack = append(stack, ctx{kind: kind, indent: indent})
+	}
+	pop := func(expected ctxKind, indent int) bool {
+		top := stack[len(stack)-1]
+		if top.kind != expected || top.indent != indent {
+			return false
+		}
+		stack = stack[:len(stack)-1]
+		return true
+	}
+	current := func() ctx {
+		return stack[len(stack)-1]
+	}
+
+	data := &readelfData{
+		sections: make(map[int]*sectionInfo),
+		symbols:  make(map[int][]symbolInfo),
+	}
+
+	var currentSection *sectionInfo
+	var currentSymbol *symbolInfo
+
+	for scanner.Scan() {
+		raw := scanner.Text()
+		trimmed := strings.TrimSpace(raw)
+		if trimmed == "" {
+			continue
+		}
+		indent := countLeadingSpaces(raw)
+		top := current()
+
+		switch {
+		case strings.HasPrefix(trimmed, "Sections [") && top.kind == ctxRoot:
+			push(ctxSections, indent)
+			continue
+		case strings.HasPrefix(trimmed, "Symbols [") && top.kind == ctxRoot:
+			push(ctxSymbols, indent)
+			continue
+		case trimmed == "Section {" && top.kind == ctxSections && indent == top.indent+2:
+			currentSection = &sectionInfo{Index: -1}
+			push(ctxSection, indent)
+			continue
+		case trimmed == "Symbol {" && top.kind == ctxSymbols && indent == top.indent+2:
+			currentSymbol = &symbolInfo{SectionIndex: -1}
+			push(ctxSymbol, indent)
+			continue
+		case trimmed == "}" && pop(ctxSection, indent):
+			if currentSection != nil && currentSection.Index >= 0 {
+				currentSection.Kind = classifySection(currentSection.Name, currentSection.Segment)
+				data.sections[currentSection.Index] = currentSection
+			}
+			currentSection = nil
+			continue
+		case trimmed == "}" && pop(ctxSymbol, indent):
+			if currentSymbol != nil && currentSymbol.SectionIndex >= 0 {
+				data.symbols[currentSymbol.SectionIndex] = append(data.symbols[currentSymbol.SectionIndex], *currentSymbol)
+			}
+			currentSymbol = nil
+			continue
+		case trimmed == "]" && (top.kind == ctxSections || top.kind == ctxSymbols) && indent == top.indent:
+			stack = stack[:len(stack)-1]
+			continue
+		}
+
+		switch top.kind {
+		case ctxSection:
+			if currentSection == nil {
+				continue
+			}
+			switch {
+			case strings.HasPrefix(trimmed, "Index: "):
+				if idx, err := strconv.Atoi(strings.TrimSpace(trimmed[len("Index: "):])); err == nil {
+					currentSection.Index = idx
+				}
+			case strings.HasPrefix(trimmed, "Name: "):
+				currentSection.Name = parseNameField(trimmed[len("Name: "):])
+			case strings.HasPrefix(trimmed, "Segment: "):
+				currentSection.Segment = parseNameField(trimmed[len("Segment: "):])
+			case strings.HasPrefix(trimmed, "Address: "):
+				if val, err := parseUintField(trimmed[len("Address: "):]); err == nil {
+					currentSection.Address = val
+				}
+			case strings.HasPrefix(trimmed, "Size: "):
+				if val, err := parseUintField(trimmed[len("Size: "):]); err == nil {
+					currentSection.Size = val
+				}
+			}
+		case ctxSymbol:
+			if currentSymbol == nil {
+				continue
+			}
+			switch {
+			case strings.HasPrefix(trimmed, "Name: "):
+				currentSymbol.Name = parseNameField(trimmed[len("Name: "):])
+			case strings.HasPrefix(trimmed, "Section: "):
+				name, idx := parseSectionRef(trimmed[len("Section: "):])
+				currentSymbol.SectionIndex = idx
+				if currentSymbol.Name == "" {
+					currentSymbol.Name = name
+				}
+			case strings.HasPrefix(trimmed, "Value: "):
+				if val, err := parseUintField(trimmed[len("Value: "):]); err == nil {
+					currentSymbol.Address = val
+				}
+			}
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		return nil, err
+	}
+	return data, nil
+}
+
+func countLeadingSpaces(line string) int {
+	count := 0
+	for _, ch := range line {
+		if ch != ' ' {
+			break
+		}
+		count++
+	}
+	return count
+}
+
+func classifySection(name, segment string) sectionKind {
+	ln := strings.ToLower(name)
+	ls := strings.ToLower(segment)
+	switch {
+	case strings.Contains(ln, "text"), strings.Contains(ln, "code"), strings.Contains(ln, "plt"):
+		return sectionText
+	case strings.Contains(ln, "rodata"), strings.Contains(ln, "const"), strings.Contains(ln, "literal"), strings.Contains(ln, "cstring"):
+		return sectionROData
+	case strings.Contains(ln, "bss"), strings.Contains(ln, "tbss"), strings.Contains(ln, "sbss"), strings.Contains(ln, "common"), strings.Contains(ln, "zerofill"):
+		return sectionBSS
+	case strings.Contains(ln, "data"), strings.Contains(ln, "got"), strings.Contains(ln, "init_array"), strings.Contains(ln, "cfstring"), strings.Contains(ln, "tdata"):
+		return sectionData
+	}
+	switch {
+	case strings.Contains(ls, "__text"):
+		return sectionText
+	case strings.Contains(ls, "data_const"):
+		return sectionROData
+	case strings.Contains(ls, "__data"):
+		return sectionData
+	}
+	return sectionUnknown
+}
+
+func buildSizeReport(path string, data *readelfData, pkgs []Package, level string) *sizeReport {
+	report := &sizeReport{Binary: path, Modules: make(map[string]*moduleSize)}
+	if data == nil {
+		return report
+	}
+	res := newNameResolver(level, pkgs)
+	var recognized bool
+	for idx, sec := range data.sections {
+		if sec == nil || sec.Size == 0 {
+			continue
+		}
+		if sec.Kind == sectionUnknown {
+			continue
+		}
+		recognized = true
+		end := sec.Address + sec.Size
+		syms := data.symbols[idx]
+		if len(syms) == 0 {
+			report.add("(unknown "+sec.Name+")", sec.Kind, sec.Size)
+			continue
+		}
+		sort.Slice(syms, func(i, j int) bool {
+			if syms[i].Address == syms[j].Address {
+				return syms[i].Name < syms[j].Name
+			}
+			return syms[i].Address < syms[j].Address
+		})
+		cursor := sec.Address
+		for i := 0; i < len(syms); i++ {
+			sym := syms[i]
+			if sym.Address >= end {
+				continue
+			}
+			addr := sym.Address
+			if addr < sec.Address {
+				addr = sec.Address
+			}
+			if addr > cursor {
+				report.add("(padding "+sec.Name+")", sec.Kind, addr-cursor)
+				cursor = addr
+			}
+			next := end
+			for j := i + 1; j < len(syms); j++ {
+				if syms[j].Address > addr {
+					next = syms[j].Address
+					break
+				}
+			}
+			if next > end {
+				next = end
+			}
+			if next <= addr {
+				continue
+			}
+			mod := res.resolve(sym.Name)
+			report.add(mod, sec.Kind, next-addr)
+			cursor = next
+		}
+		if cursor < end {
+			report.add("(padding "+sec.Name+")", sec.Kind, end-cursor)
+		}
+	}
+	if !recognized {
+		return nil
+	}
+	return report
+}
+
+func emitJSONReport(w io.Writer, report *sizeReport) error {
+	type moduleJSON struct {
+		Name   string `json:"name"`
+		Code   uint64 `json:"code"`
+		ROData uint64 `json:"rodata"`
+		Data   uint64 `json:"data"`
+		BSS    uint64 `json:"bss"`
+		Flash  uint64 `json:"flash"`
+		RAM    uint64 `json:"ram"`
+	}
+	mods := report.sortedModules()
+	jsonMods := make([]moduleJSON, 0, len(mods))
+	for _, m := range mods {
+		jsonMods = append(jsonMods, moduleJSON{
+			Name:   m.Name,
+			Code:   m.Code,
+			ROData: m.ROData,
+			Data:   m.Data,
+			BSS:    m.BSS,
+			Flash:  m.Flash(),
+			RAM:    m.RAM(),
+		})
+	}
+	payload := struct {
+		Binary  string       `json:"binary"`
+		Modules []moduleJSON `json:"modules"`
+		Total   moduleJSON   `json:"total"`
+	}{
+		Binary:  filepath.Clean(report.Binary),
+		Modules: jsonMods,
+		Total: moduleJSON{
+			Name:   "total",
+			Code:   report.Total.Code,
+			ROData: report.Total.ROData,
+			Data:   report.Total.Data,
+			BSS:    report.Total.BSS,
+			Flash:  report.Total.Flash(),
+			RAM:    report.Total.RAM(),
+		},
+	}
+	enc := json.NewEncoder(w)
+	enc.SetIndent("", "  ")
+	return enc.Encode(payload)
+}
+
+func printTextReport(w io.Writer, report *sizeReport) {
+	fmt.Fprintf(w, "\nSize report for %s\n", filepath.Clean(report.Binary))
+	fmt.Fprintln(w, "   code  rodata    data     bss |   flash     ram | module")
+	fmt.Fprintln(w, "------------------------------- | --------------- | ----------------")
+	for _, m := range report.sortedModules() {
+		fmt.Fprintf(w, "%7d %7d %7d %7d | %7d %7d | %s\n", m.Code, m.ROData, m.Data, m.BSS, m.Flash(), m.RAM(), m.Name)
+	}
+	fmt.Fprintln(w, "------------------------------- | --------------- | ----------------")
+	fmt.Fprintf(w, "%7d %7d %7d %7d | %7d %7d | total\n", report.Total.Code, report.Total.ROData, report.Total.Data, report.Total.BSS, report.Total.Flash(), report.Total.RAM())
+}
+
+func (r *sizeReport) sortedModules() []*moduleSize {
+	mods := make([]*moduleSize, 0, len(r.Modules))
+	for _, m := range r.Modules {
+		mods = append(mods, m)
+	}
+	sort.Slice(mods, func(i, j int) bool {
+		if mods[i].Flash() == mods[j].Flash() {
+			return mods[i].Name < mods[j].Name
+		}
+		return mods[i].Flash() > mods[j].Flash()
+	})
+	return mods
+}
+
+func moduleNameFromSymbol(raw string) string {
+	name := strings.TrimSpace(raw)
+	name = strings.TrimPrefix(name, "_")
+	name = strings.TrimPrefix(name, ".")
+	if name == "" {
+		return "(anonymous)"
+	}
+	if idx := strings.Index(name, " "); idx > 0 {
+		name = name[:idx]
+	}
+	if idx := strings.Index(name, "@"); idx > 0 {
+		name = name[:idx]
+	}
+	lastDot := strings.LastIndex(name, ".")
+	if lastDot > 0 {
+		pkg := name[:lastDot]
+		if paren := strings.Index(pkg, "("); paren > 0 {
+			pkg = pkg[:paren]
+		}
+		pkg = strings.Trim(pkg, " ")
+		if pkg != "" {
+			return pkg
+		}
+	}
+	return name
+}
+
+func parseNameField(field string) string {
+	val := strings.TrimSpace(field)
+	if idx := strings.Index(val, "("); idx >= 0 {
+		val = strings.TrimSpace(val[:idx])
+	}
+	return val
+}
+
+func parseSectionRef(field string) (string, int) {
+	name := parseNameField(field)
+	idx := strings.Index(field, "(")
+	if idx < 0 {
+		return name, -1
+	}
+	end := strings.Index(field[idx:], ")")
+	if end < 0 {
+		return name, -1
+	}
+	val := strings.TrimSpace(field[idx+1 : idx+end])
+	val = strings.TrimPrefix(val, "0x")
+	if val == "" {
+		return name, -1
+	}
+	num, err := strconv.ParseUint(val, 16, 64)
+	if err != nil {
+		return name, -1
+	}
+	if num == 0 {
+		return name, -1
+	}
+	return name, int(num - 1)
+}
+
+func parseUintField(field string) (uint64, error) {
+	val := strings.TrimSpace(field)
+	if strings.HasPrefix(val, "0x") || strings.HasPrefix(val, "0X") {
+		return strconv.ParseUint(val[2:], 16, 64)
+	}
+	return strconv.ParseUint(val, 10, 64)
+}
+
+func ensureSizeReporting(conf *Config) error {
+	if !conf.SizeReport {
+		return nil
+	}
+	switch strings.ToLower(conf.SizeLevel) {
+	case "", "module":
+		conf.SizeLevel = "module"
+	case "package", "full":
+		conf.SizeLevel = strings.ToLower(conf.SizeLevel)
+	default:
+		return fmt.Errorf("invalid size level %q (valid: full,module,package)", conf.SizeLevel)
+	}
+	if _, err := exec.LookPath("llvm-readelf"); err != nil {
+		return errors.New("llvm-readelf not found in PATH")
+	}
+	return nil
+}

internal/build/size_report_test.go

@@ -0,0 +1,156 @@
+//go:build !llgo
+
+package build
+
+import (
+	"bytes"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"golang.org/x/tools/go/packages"
+)
+
+const sampleReadelf = `Sections [
+  Section {
+    Index: 0
+    Name: __text (5F)
+    Segment: __TEXT (5F)
+    Address: 0x1000
+    Size: 0x20
+  }
+  Section {
+    Index: 1
+    Name: __data
+    Segment: __DATA
+    Address: 0x2000
+    Size: 0x10
+  }
+  Section {
+    Index: 2
+    Name: __common
+    Segment: __DATA
+    Address: 0x3000
+    Size: 0x8
+  }
+]
+Symbols [
+  Symbol {
+    Name: _main.main
+    Section: __text (0x1)
+    Value: 0x1000
+  }
+  Symbol {
+    Name: _runtime.init
+    Section: __text (0x1)
+    Value: 0x1010
+  }
+  Symbol {
+    Name: _main.dataVar
+    Section: __data (0x2)
+    Value: 0x2000
+  }
+  Symbol {
+    Name: _runtime.dataVar
+    Section: __data (0x2)
+    Value: 0x2008
+  }
+  Symbol {
+    Name: _runtime.bssVar
+    Section: __common (0x3)
+    Value: 0x3000
+  }
+]
+`
+
+func TestParseReadelfOutput(t *testing.T) {
+	parsed, err := parseReadelfOutput(strings.NewReader(sampleReadelf))
+	if err != nil {
+		t.Fatalf("parseReadelfOutput: %v", err)
+	}
+	report := buildSizeReport("fake.bin", parsed, nil, "")
+	if report == nil {
+		t.Fatal("expected report")
+	}
+	modules := report.Modules
+	if len(modules) == 0 {
+		t.Fatal("expected modules in report")
+	}
+	mainMod, ok := modules["main"]
+	if !ok {
+		t.Fatalf("expected main module, got %v", modules)
+	}
+	if mainMod.Code != 0x10 {
+		t.Fatalf("unexpected main code size: %d", mainMod.Code)
+	}
+	if mainMod.Data != 0x8 {
+		t.Fatalf("unexpected main data size: %d", mainMod.Data)
+	}
+	runtimeMod := modules["runtime"]
+	if runtimeMod.Code != 0x10 {
+		t.Fatalf("unexpected runtime code size: %d", runtimeMod.Code)
+	}
+	if runtimeMod.Data != 0x8 {
+		t.Fatalf("unexpected runtime data size: %d", runtimeMod.Data)
+	}
+	if runtimeMod.BSS != 0x8 {
+		t.Fatalf("unexpected runtime bss size: %d", runtimeMod.BSS)
+	}
+	if report.Total.Flash() != 0x10+0x10+0x8+0x8 {
+		t.Fatalf("unexpected flash total: %d", report.Total.Flash())
+	}
+	if report.Total.RAM() != 0x8+0x8+0x8 {
+		t.Fatalf("unexpected ram total: %d", report.Total.RAM())
+	}
+}
+
+func TestParseReadelfRealBinary(t *testing.T) {
+	path := os.Getenv("LLGO_SIZE_REPORT_BIN")
+	if path == "" {
+		return
+	}
+	absPath, err := filepath.Abs(path)
+	if err != nil {
+		t.Fatalf("abs path: %v", err)
+	}
+	cmd := exec.Command("llvm-readelf", "--all", absPath)
+	var buf bytes.Buffer
+	cmd.Stdout = &buf
+	cmd.Stderr = os.Stderr
+	if err := cmd.Run(); err != nil {
+		t.Fatalf("llvm-readelf failed: %v", err)
+	}
+	parsed, err := parseReadelfOutput(bytes.NewReader(buf.Bytes()))
+	if err != nil {
+		t.Fatalf("parseReadelfOutput(real): %v", err)
+	}
+	if len(parsed.sections) == 0 {
+		t.Fatal("expected sections in real binary")
+	}
+	report := buildSizeReport(absPath, parsed, nil, "")
+	if len(report.Modules) == 0 {
+		t.Fatalf("expected modules for %s", path)
+	}
+}
+
+func TestNameResolver(t *testing.T) {
+	pkgs := []Package{
+		&aPackage{Package: &packages.Package{PkgPath: "github.com/foo/bar", Module: &packages.Module{Path: "github.com/foo"}}},
+	}
+	symbol := "_github.com/foo/bar.Type.method"
+	if got := newNameResolver("package", pkgs).resolve(symbol); got != "github.com/foo/bar" {
+		t.Fatalf("package level want github.com/foo/bar, got %q", got)
+	}
+	if got := newNameResolver("module", pkgs).resolve(symbol); got != "github.com/foo" {
+		t.Fatalf("module level want github.com/foo, got %q", got)
+	}
+	full := newNameResolver("full", pkgs).resolve(symbol)
+	if full != "github.com/foo/bar.Type" {
+		t.Fatalf("full level unexpected: %q", full)
+	}
+	if got := newNameResolver("package", nil).resolve("_llgo_stub.foo"); got != "llgo-stubs" {
+		t.Fatalf("llgo default grouping failed: %q", got)
+	}
+}