Initial commit: Go 1.23 release state

src/internal/profile/encode.go

@@ -0,0 +1,482 @@
+// Copyright 2014 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package profile
+
+import (
+	"errors"
+	"fmt"
+	"sort"
+)
+
+func (p *Profile) decoder() []decoder {
+	return profileDecoder
+}
+
+// preEncode populates the unexported fields to be used by encode
+// (with suffix X) from the corresponding exported fields. The
+// exported fields are cleared up to facilitate testing.
+func (p *Profile) preEncode() {
+	strings := make(map[string]int)
+	addString(strings, "")
+
+	for _, st := range p.SampleType {
+		st.typeX = addString(strings, st.Type)
+		st.unitX = addString(strings, st.Unit)
+	}
+
+	for _, s := range p.Sample {
+		s.labelX = nil
+		var keys []string
+		for k := range s.Label {
+			keys = append(keys, k)
+		}
+		sort.Strings(keys)
+		for _, k := range keys {
+			vs := s.Label[k]
+			for _, v := range vs {
+				s.labelX = append(s.labelX,
+					Label{
+						keyX: addString(strings, k),
+						strX: addString(strings, v),
+					},
+				)
+			}
+		}
+		var numKeys []string
+		for k := range s.NumLabel {
+			numKeys = append(numKeys, k)
+		}
+		sort.Strings(numKeys)
+		for _, k := range numKeys {
+			vs := s.NumLabel[k]
+			for _, v := range vs {
+				s.labelX = append(s.labelX,
+					Label{
+						keyX: addString(strings, k),
+						numX: v,
+					},
+				)
+			}
+		}
+		s.locationIDX = nil
+		for _, l := range s.Location {
+			s.locationIDX = append(s.locationIDX, l.ID)
+		}
+	}
+
+	for _, m := range p.Mapping {
+		m.fileX = addString(strings, m.File)
+		m.buildIDX = addString(strings, m.BuildID)
+	}
+
+	for _, l := range p.Location {
+		for i, ln := range l.Line {
+			if ln.Function != nil {
+				l.Line[i].functionIDX = ln.Function.ID
+			} else {
+				l.Line[i].functionIDX = 0
+			}
+		}
+		if l.Mapping != nil {
+			l.mappingIDX = l.Mapping.ID
+		} else {
+			l.mappingIDX = 0
+		}
+	}
+	for _, f := range p.Function {
+		f.nameX = addString(strings, f.Name)
+		f.systemNameX = addString(strings, f.SystemName)
+		f.filenameX = addString(strings, f.Filename)
+	}
+
+	p.dropFramesX = addString(strings, p.DropFrames)
+	p.keepFramesX = addString(strings, p.KeepFrames)
+
+	if pt := p.PeriodType; pt != nil {
+		pt.typeX = addString(strings, pt.Type)
+		pt.unitX = addString(strings, pt.Unit)
+	}
+
+	p.stringTable = make([]string, len(strings))
+	for s, i := range strings {
+		p.stringTable[i] = s
+	}
+}
+
+func (p *Profile) encode(b *buffer) {
+	for _, x := range p.SampleType {
+		encodeMessage(b, 1, x)
+	}
+	for _, x := range p.Sample {
+		encodeMessage(b, 2, x)
+	}
+	for _, x := range p.Mapping {
+		encodeMessage(b, 3, x)
+	}
+	for _, x := range p.Location {
+		encodeMessage(b, 4, x)
+	}
+	for _, x := range p.Function {
+		encodeMessage(b, 5, x)
+	}
+	encodeStrings(b, 6, p.stringTable)
+	encodeInt64Opt(b, 7, p.dropFramesX)
+	encodeInt64Opt(b, 8, p.keepFramesX)
+	encodeInt64Opt(b, 9, p.TimeNanos)
+	encodeInt64Opt(b, 10, p.DurationNanos)
+	if pt := p.PeriodType; pt != nil && (pt.typeX != 0 || pt.unitX != 0) {
+		encodeMessage(b, 11, p.PeriodType)
+	}
+	encodeInt64Opt(b, 12, p.Period)
+}
+
+var profileDecoder = []decoder{
+	nil, // 0
+	// repeated ValueType sample_type = 1
+	func(b *buffer, m message) error {
+		x := new(ValueType)
+		pp := m.(*Profile)
+		pp.SampleType = append(pp.SampleType, x)
+		return decodeMessage(b, x)
+	},
+	// repeated Sample sample = 2
+	func(b *buffer, m message) error {
+		x := new(Sample)
+		pp := m.(*Profile)
+		pp.Sample = append(pp.Sample, x)
+		return decodeMessage(b, x)
+	},
+	// repeated Mapping mapping = 3
+	func(b *buffer, m message) error {
+		x := new(Mapping)
+		pp := m.(*Profile)
+		pp.Mapping = append(pp.Mapping, x)
+		return decodeMessage(b, x)
+	},
+	// repeated Location location = 4
+	func(b *buffer, m message) error {
+		x := new(Location)
+		pp := m.(*Profile)
+		pp.Location = append(pp.Location, x)
+		return decodeMessage(b, x)
+	},
+	// repeated Function function = 5
+	func(b *buffer, m message) error {
+		x := new(Function)
+		pp := m.(*Profile)
+		pp.Function = append(pp.Function, x)
+		return decodeMessage(b, x)
+	},
+	// repeated string string_table = 6
+	func(b *buffer, m message) error {
+		err := decodeStrings(b, &m.(*Profile).stringTable)
+		if err != nil {
+			return err
+		}
+		if m.(*Profile).stringTable[0] != "" {
+			return errors.New("string_table[0] must be ''")
+		}
+		return nil
+	},
+	// repeated int64 drop_frames = 7
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).dropFramesX) },
+	// repeated int64 keep_frames = 8
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).keepFramesX) },
+	// repeated int64 time_nanos = 9
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).TimeNanos) },
+	// repeated int64 duration_nanos = 10
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).DurationNanos) },
+	// optional string period_type = 11
+	func(b *buffer, m message) error {
+		x := new(ValueType)
+		pp := m.(*Profile)
+		pp.PeriodType = x
+		return decodeMessage(b, x)
+	},
+	// repeated int64 period = 12
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).Period) },
+	// repeated int64 comment = 13
+	func(b *buffer, m message) error { return decodeInt64s(b, &m.(*Profile).commentX) },
+	// int64 defaultSampleType = 14
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).defaultSampleTypeX) },
+}
+
+// postDecode takes the unexported fields populated by decode (with
+// suffix X) and populates the corresponding exported fields.
+// The unexported fields are cleared up to facilitate testing.
+func (p *Profile) postDecode() error {
+	var err error
+
+	mappings := make(map[uint64]*Mapping)
+	for _, m := range p.Mapping {
+		m.File, err = getString(p.stringTable, &m.fileX, err)
+		m.BuildID, err = getString(p.stringTable, &m.buildIDX, err)
+		mappings[m.ID] = m
+	}
+
+	functions := make(map[uint64]*Function)
+	for _, f := range p.Function {
+		f.Name, err = getString(p.stringTable, &f.nameX, err)
+		f.SystemName, err = getString(p.stringTable, &f.systemNameX, err)
+		f.Filename, err = getString(p.stringTable, &f.filenameX, err)
+		functions[f.ID] = f
+	}
+
+	locations := make(map[uint64]*Location)
+	for _, l := range p.Location {
+		l.Mapping = mappings[l.mappingIDX]
+		l.mappingIDX = 0
+		for i, ln := range l.Line {
+			if id := ln.functionIDX; id != 0 {
+				l.Line[i].Function = functions[id]
+				if l.Line[i].Function == nil {
+					return fmt.Errorf("Function ID %d not found", id)
+				}
+				l.Line[i].functionIDX = 0
+			}
+		}
+		locations[l.ID] = l
+	}
+
+	for _, st := range p.SampleType {
+		st.Type, err = getString(p.stringTable, &st.typeX, err)
+		st.Unit, err = getString(p.stringTable, &st.unitX, err)
+	}
+
+	for _, s := range p.Sample {
+		labels := make(map[string][]string)
+		numLabels := make(map[string][]int64)
+		for _, l := range s.labelX {
+			var key, value string
+			key, err = getString(p.stringTable, &l.keyX, err)
+			if l.strX != 0 {
+				value, err = getString(p.stringTable, &l.strX, err)
+				labels[key] = append(labels[key], value)
+			} else {
+				numLabels[key] = append(numLabels[key], l.numX)
+			}
+		}
+		if len(labels) > 0 {
+			s.Label = labels
+		}
+		if len(numLabels) > 0 {
+			s.NumLabel = numLabels
+		}
+		s.Location = nil
+		for _, lid := range s.locationIDX {
+			s.Location = append(s.Location, locations[lid])
+		}
+		s.locationIDX = nil
+	}
+
+	p.DropFrames, err = getString(p.stringTable, &p.dropFramesX, err)
+	p.KeepFrames, err = getString(p.stringTable, &p.keepFramesX, err)
+
+	if pt := p.PeriodType; pt == nil {
+		p.PeriodType = &ValueType{}
+	}
+
+	if pt := p.PeriodType; pt != nil {
+		pt.Type, err = getString(p.stringTable, &pt.typeX, err)
+		pt.Unit, err = getString(p.stringTable, &pt.unitX, err)
+	}
+	for _, i := range p.commentX {
+		var c string
+		c, err = getString(p.stringTable, &i, err)
+		p.Comments = append(p.Comments, c)
+	}
+
+	p.commentX = nil
+	p.DefaultSampleType, err = getString(p.stringTable, &p.defaultSampleTypeX, err)
+	p.stringTable = nil
+	return err
+}
+
+func (p *ValueType) decoder() []decoder {
+	return valueTypeDecoder
+}
+
+func (p *ValueType) encode(b *buffer) {
+	encodeInt64Opt(b, 1, p.typeX)
+	encodeInt64Opt(b, 2, p.unitX)
+}
+
+var valueTypeDecoder = []decoder{
+	nil, // 0
+	// optional int64 type = 1
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*ValueType).typeX) },
+	// optional int64 unit = 2
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*ValueType).unitX) },
+}
+
+func (p *Sample) decoder() []decoder {
+	return sampleDecoder
+}
+
+func (p *Sample) encode(b *buffer) {
+	encodeUint64s(b, 1, p.locationIDX)
+	for _, x := range p.Value {
+		encodeInt64(b, 2, x)
+	}
+	for _, x := range p.labelX {
+		encodeMessage(b, 3, x)
+	}
+}
+
+var sampleDecoder = []decoder{
+	nil, // 0
+	// repeated uint64 location = 1
+	func(b *buffer, m message) error { return decodeUint64s(b, &m.(*Sample).locationIDX) },
+	// repeated int64 value = 2
+	func(b *buffer, m message) error { return decodeInt64s(b, &m.(*Sample).Value) },
+	// repeated Label label = 3
+	func(b *buffer, m message) error {
+		s := m.(*Sample)
+		n := len(s.labelX)
+		s.labelX = append(s.labelX, Label{})
+		return decodeMessage(b, &s.labelX[n])
+	},
+}
+
+func (p Label) decoder() []decoder {
+	return labelDecoder
+}
+
+func (p Label) encode(b *buffer) {
+	encodeInt64Opt(b, 1, p.keyX)
+	encodeInt64Opt(b, 2, p.strX)
+	encodeInt64Opt(b, 3, p.numX)
+}
+
+var labelDecoder = []decoder{
+	nil, // 0
+	// optional int64 key = 1
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Label).keyX) },
+	// optional int64 str = 2
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Label).strX) },
+	// optional int64 num = 3
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Label).numX) },
+}
+
+func (p *Mapping) decoder() []decoder {
+	return mappingDecoder
+}
+
+func (p *Mapping) encode(b *buffer) {
+	encodeUint64Opt(b, 1, p.ID)
+	encodeUint64Opt(b, 2, p.Start)
+	encodeUint64Opt(b, 3, p.Limit)
+	encodeUint64Opt(b, 4, p.Offset)
+	encodeInt64Opt(b, 5, p.fileX)
+	encodeInt64Opt(b, 6, p.buildIDX)
+	encodeBoolOpt(b, 7, p.HasFunctions)
+	encodeBoolOpt(b, 8, p.HasFilenames)
+	encodeBoolOpt(b, 9, p.HasLineNumbers)
+	encodeBoolOpt(b, 10, p.HasInlineFrames)
+}
+
+var mappingDecoder = []decoder{
+	nil, // 0
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).ID) },            // optional uint64 id = 1
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).Start) },         // optional uint64 memory_offset = 2
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).Limit) },         // optional uint64 memory_limit = 3
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).Offset) },        // optional uint64 file_offset = 4
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Mapping).fileX) },          // optional int64 filename = 5
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Mapping).buildIDX) },       // optional int64 build_id = 6
+	func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasFunctions) },    // optional bool has_functions = 7
+	func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasFilenames) },    // optional bool has_filenames = 8
+	func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasLineNumbers) },  // optional bool has_line_numbers = 9
+	func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasInlineFrames) }, // optional bool has_inline_frames = 10
+}
+
+func (p *Location) decoder() []decoder {
+	return locationDecoder
+}
+
+func (p *Location) encode(b *buffer) {
+	encodeUint64Opt(b, 1, p.ID)
+	encodeUint64Opt(b, 2, p.mappingIDX)
+	encodeUint64Opt(b, 3, p.Address)
+	for i := range p.Line {
+		encodeMessage(b, 4, &p.Line[i])
+	}
+}
+
+var locationDecoder = []decoder{
+	nil, // 0
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Location).ID) },         // optional uint64 id = 1;
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Location).mappingIDX) }, // optional uint64 mapping_id = 2;
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Location).Address) },    // optional uint64 address = 3;
+	func(b *buffer, m message) error { // repeated Line line = 4
+		pp := m.(*Location)
+		n := len(pp.Line)
+		pp.Line = append(pp.Line, Line{})
+		return decodeMessage(b, &pp.Line[n])
+	},
+}
+
+func (p *Line) decoder() []decoder {
+	return lineDecoder
+}
+
+func (p *Line) encode(b *buffer) {
+	encodeUint64Opt(b, 1, p.functionIDX)
+	encodeInt64Opt(b, 2, p.Line)
+}
+
+var lineDecoder = []decoder{
+	nil, // 0
+	// optional uint64 function_id = 1
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Line).functionIDX) },
+	// optional int64 line = 2
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Line).Line) },
+}
+
+func (p *Function) decoder() []decoder {
+	return functionDecoder
+}
+
+func (p *Function) encode(b *buffer) {
+	encodeUint64Opt(b, 1, p.ID)
+	encodeInt64Opt(b, 2, p.nameX)
+	encodeInt64Opt(b, 3, p.systemNameX)
+	encodeInt64Opt(b, 4, p.filenameX)
+	encodeInt64Opt(b, 5, p.StartLine)
+}
+
+var functionDecoder = []decoder{
+	nil, // 0
+	// optional uint64 id = 1
+	func(b *buffer, m message) error { return decodeUint64(b, &m.(*Function).ID) },
+	// optional int64 function_name = 2
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).nameX) },
+	// optional int64 function_system_name = 3
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).systemNameX) },
+	// repeated int64 filename = 4
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).filenameX) },
+	// optional int64 start_line = 5
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).StartLine) },
+}
+
+func addString(strings map[string]int, s string) int64 {
+	i, ok := strings[s]
+	if !ok {
+		i = len(strings)
+		strings[s] = i
+	}
+	return int64(i)
+}
+
+func getString(strings []string, strng *int64, err error) (string, error) {
+	if err != nil {
+		return "", err
+	}
+	s := int(*strng)
+	if s < 0 || s >= len(strings) {
+		return "", errMalformed
+	}
+	*strng = 0
+	return strings[s], nil
+}

src/internal/profile/filter.go

@@ -0,0 +1,54 @@
+// Copyright 2014 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Implements methods to filter samples from profiles.
+
+package profile
+
+// TagMatch selects tags for filtering
+type TagMatch func(key, val string, nval int64) bool
+
+// FilterSamplesByTag removes all samples from the profile, except
+// those that match focus and do not match the ignore regular
+// expression.
+func (p *Profile) FilterSamplesByTag(focus, ignore TagMatch) (fm, im bool) {
+	samples := make([]*Sample, 0, len(p.Sample))
+	for _, s := range p.Sample {
+		focused, ignored := focusedSample(s, focus, ignore)
+		fm = fm || focused
+		im = im || ignored
+		if focused && !ignored {
+			samples = append(samples, s)
+		}
+	}
+	p.Sample = samples
+	return
+}
+
+// focusedSample checks a sample against focus and ignore regexps.
+// Returns whether the focus/ignore regexps match any tags.
+func focusedSample(s *Sample, focus, ignore TagMatch) (fm, im bool) {
+	fm = focus == nil
+	for key, vals := range s.Label {
+		for _, val := range vals {
+			if ignore != nil && ignore(key, val, 0) {
+				im = true
+			}
+			if !fm && focus(key, val, 0) {
+				fm = true
+			}
+		}
+	}
+	for key, vals := range s.NumLabel {
+		for _, val := range vals {
+			if ignore != nil && ignore(key, "", val) {
+				im = true
+			}
+			if !fm && focus(key, "", val) {
+				fm = true
+			}
+		}
+	}
+	return fm, im
+}

src/internal/profile/graph.go

@@ -0,0 +1,519 @@
+// Copyright 2014 Google Inc. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package profile represents a pprof profile as a directed graph.
+//
+// This package is a simplified fork of github.com/google/pprof/internal/graph.
+package profile
+
+import (
+	"fmt"
+	"sort"
+	"strings"
+)
+
+// Options encodes the options for constructing a graph
+type Options struct {
+	SampleValue       func(s []int64) int64 // Function to compute the value of a sample
+	SampleMeanDivisor func(s []int64) int64 // Function to compute the divisor for mean graphs, or nil
+
+	DropNegative bool // Drop nodes with overall negative values
+
+	KeptNodes NodeSet // If non-nil, only use nodes in this set
+}
+
+// Nodes is an ordered collection of graph nodes.
+type Nodes []*Node
+
+// Node is an entry on a profiling report. It represents a unique
+// program location.
+type Node struct {
+	// Info describes the source location associated to this node.
+	Info NodeInfo
+
+	// Function represents the function that this node belongs to. On
+	// graphs with sub-function resolution (eg line number or
+	// addresses), two nodes in a NodeMap that are part of the same
+	// function have the same value of Node.Function. If the Node
+	// represents the whole function, it points back to itself.
+	Function *Node
+
+	// Values associated to this node. Flat is exclusive to this node,
+	// Cum includes all descendents.
+	Flat, FlatDiv, Cum, CumDiv int64
+
+	// In and out Contains the nodes immediately reaching or reached by
+	// this node.
+	In, Out EdgeMap
+}
+
+// Graph summarizes a performance profile into a format that is
+// suitable for visualization.
+type Graph struct {
+	Nodes Nodes
+}
+
+// FlatValue returns the exclusive value for this node, computing the
+// mean if a divisor is available.
+func (n *Node) FlatValue() int64 {
+	if n.FlatDiv == 0 {
+		return n.Flat
+	}
+	return n.Flat / n.FlatDiv
+}
+
+// CumValue returns the inclusive value for this node, computing the
+// mean if a divisor is available.
+func (n *Node) CumValue() int64 {
+	if n.CumDiv == 0 {
+		return n.Cum
+	}
+	return n.Cum / n.CumDiv
+}
+
+// AddToEdge increases the weight of an edge between two nodes. If
+// there isn't such an edge one is created.
+func (n *Node) AddToEdge(to *Node, v int64, residual, inline bool) {
+	n.AddToEdgeDiv(to, 0, v, residual, inline)
+}
+
+// AddToEdgeDiv increases the weight of an edge between two nodes. If
+// there isn't such an edge one is created.
+func (n *Node) AddToEdgeDiv(to *Node, dv, v int64, residual, inline bool) {
+	if e := n.Out.FindTo(to); e != nil {
+		e.WeightDiv += dv
+		e.Weight += v
+		if residual {
+			e.Residual = true
+		}
+		if !inline {
+			e.Inline = false
+		}
+		return
+	}
+
+	info := &Edge{Src: n, Dest: to, WeightDiv: dv, Weight: v, Residual: residual, Inline: inline}
+	n.Out.Add(info)
+	to.In.Add(info)
+}
+
+// NodeInfo contains the attributes for a node.
+type NodeInfo struct {
+	Name              string
+	Address           uint64
+	StartLine, Lineno int
+}
+
+// PrintableName calls the Node's Formatter function with a single space separator.
+func (i *NodeInfo) PrintableName() string {
+	return strings.Join(i.NameComponents(), " ")
+}
+
+// NameComponents returns the components of the printable name to be used for a node.
+func (i *NodeInfo) NameComponents() []string {
+	var name []string
+	if i.Address != 0 {
+		name = append(name, fmt.Sprintf("%016x", i.Address))
+	}
+	if fun := i.Name; fun != "" {
+		name = append(name, fun)
+	}
+
+	switch {
+	case i.Lineno != 0:
+		// User requested line numbers, provide what we have.
+		name = append(name, fmt.Sprintf(":%d", i.Lineno))
+	case i.Name != "":
+		// User requested function name. It was already included.
+	default:
+		// Do not leave it empty if there is no information at all.
+		name = append(name, "<unknown>")
+	}
+	return name
+}
+
+// NodeMap maps from a node info struct to a node. It is used to merge
+// report entries with the same info.
+type NodeMap map[NodeInfo]*Node
+
+// NodeSet is a collection of node info structs.
+type NodeSet map[NodeInfo]bool
+
+// NodePtrSet is a collection of nodes. Trimming a graph or tree requires a set
+// of objects which uniquely identify the nodes to keep. In a graph, NodeInfo
+// works as a unique identifier; however, in a tree multiple nodes may share
+// identical NodeInfos. A *Node does uniquely identify a node so we can use that
+// instead. Though a *Node also uniquely identifies a node in a graph,
+// currently, during trimming, graphs are rebuilt from scratch using only the
+// NodeSet, so there would not be the required context of the initial graph to
+// allow for the use of *Node.
+type NodePtrSet map[*Node]bool
+
+// FindOrInsertNode takes the info for a node and either returns a matching node
+// from the node map if one exists, or adds one to the map if one does not.
+// If kept is non-nil, nodes are only added if they can be located on it.
+func (nm NodeMap) FindOrInsertNode(info NodeInfo, kept NodeSet) *Node {
+	if kept != nil {
+		if _, ok := kept[info]; !ok {
+			return nil
+		}
+	}
+
+	if n, ok := nm[info]; ok {
+		return n
+	}
+
+	n := &Node{
+		Info: info,
+	}
+	nm[info] = n
+	if info.Address == 0 && info.Lineno == 0 {
+		// This node represents the whole function, so point Function
+		// back to itself.
+		n.Function = n
+		return n
+	}
+	// Find a node that represents the whole function.
+	info.Address = 0
+	info.Lineno = 0
+	n.Function = nm.FindOrInsertNode(info, nil)
+	return n
+}
+
+// EdgeMap is used to represent the incoming/outgoing edges from a node.
+type EdgeMap []*Edge
+
+func (em EdgeMap) FindTo(n *Node) *Edge {
+	for _, e := range em {
+		if e.Dest == n {
+			return e
+		}
+	}
+	return nil
+}
+
+func (em *EdgeMap) Add(e *Edge) {
+	*em = append(*em, e)
+}
+
+func (em *EdgeMap) Delete(e *Edge) {
+	for i, edge := range *em {
+		if edge == e {
+			(*em)[i] = (*em)[len(*em)-1]
+			*em = (*em)[:len(*em)-1]
+			return
+		}
+	}
+}
+
+// Edge contains any attributes to be represented about edges in a graph.
+type Edge struct {
+	Src, Dest *Node
+	// The summary weight of the edge
+	Weight, WeightDiv int64
+
+	// residual edges connect nodes that were connected through a
+	// separate node, which has been removed from the report.
+	Residual bool
+	// An inline edge represents a call that was inlined into the caller.
+	Inline bool
+}
+
+// WeightValue returns the weight value for this edge, normalizing if a
+// divisor is available.
+func (e *Edge) WeightValue() int64 {
+	if e.WeightDiv == 0 {
+		return e.Weight
+	}
+	return e.Weight / e.WeightDiv
+}
+
+// NewGraph computes a graph from a profile.
+func NewGraph(prof *Profile, o *Options) *Graph {
+	nodes, locationMap := CreateNodes(prof, o)
+	seenNode := make(map[*Node]bool)
+	seenEdge := make(map[nodePair]bool)
+	for _, sample := range prof.Sample {
+		var w, dw int64
+		w = o.SampleValue(sample.Value)
+		if o.SampleMeanDivisor != nil {
+			dw = o.SampleMeanDivisor(sample.Value)
+		}
+		if dw == 0 && w == 0 {
+			continue
+		}
+		for k := range seenNode {
+			delete(seenNode, k)
+		}
+		for k := range seenEdge {
+			delete(seenEdge, k)
+		}
+		var parent *Node
+		// A residual edge goes over one or more nodes that were not kept.
+		residual := false
+
+		// Group the sample frames, based on a global map.
+		// Count only the last two frames as a call edge. Frames higher up
+		// the stack are unlikely to be repeated calls (e.g. runtime.main
+		// calling main.main). So adding weights to call edges higher up
+		// the stack may be not reflecting the actual call edge weights
+		// in the program. Without a branch profile this is just an
+		// approximation.
+		i := 1
+		if last := len(sample.Location) - 1; last < i {
+			i = last
+		}
+		for ; i >= 0; i-- {
+			l := sample.Location[i]
+			locNodes := locationMap.get(l.ID)
+			for ni := len(locNodes) - 1; ni >= 0; ni-- {
+				n := locNodes[ni]
+				if n == nil {
+					residual = true
+					continue
+				}
+				// Add cum weight to all nodes in stack, avoiding double counting.
+				_, sawNode := seenNode[n]
+				if !sawNode {
+					seenNode[n] = true
+					n.addSample(dw, w, false)
+				}
+				// Update edge weights for all edges in stack, avoiding double counting.
+				if (!sawNode || !seenEdge[nodePair{n, parent}]) && parent != nil && n != parent {
+					seenEdge[nodePair{n, parent}] = true
+					parent.AddToEdgeDiv(n, dw, w, residual, ni != len(locNodes)-1)
+				}
+
+				parent = n
+				residual = false
+			}
+		}
+		if parent != nil && !residual {
+			// Add flat weight to leaf node.
+			parent.addSample(dw, w, true)
+		}
+	}
+
+	return selectNodesForGraph(nodes, o.DropNegative)
+}
+
+func selectNodesForGraph(nodes Nodes, dropNegative bool) *Graph {
+	// Collect nodes into a graph.
+	gNodes := make(Nodes, 0, len(nodes))
+	for _, n := range nodes {
+		if n == nil {
+			continue
+		}
+		if n.Cum == 0 && n.Flat == 0 {
+			continue
+		}
+		if dropNegative && isNegative(n) {
+			continue
+		}
+		gNodes = append(gNodes, n)
+	}
+	return &Graph{gNodes}
+}
+
+type nodePair struct {
+	src, dest *Node
+}
+
+// isNegative returns true if the node is considered as "negative" for the
+// purposes of drop_negative.
+func isNegative(n *Node) bool {
+	switch {
+	case n.Flat < 0:
+		return true
+	case n.Flat == 0 && n.Cum < 0:
+		return true
+	default:
+		return false
+	}
+}
+
+type locationMap struct {
+	s []Nodes          // a slice for small sequential IDs
+	m map[uint64]Nodes // fallback for large IDs (unlikely)
+}
+
+func (l *locationMap) add(id uint64, n Nodes) {
+	if id < uint64(len(l.s)) {
+		l.s[id] = n
+	} else {
+		l.m[id] = n
+	}
+}
+
+func (l locationMap) get(id uint64) Nodes {
+	if id < uint64(len(l.s)) {
+		return l.s[id]
+	} else {
+		return l.m[id]
+	}
+}
+
+// CreateNodes creates graph nodes for all locations in a profile. It
+// returns set of all nodes, plus a mapping of each location to the
+// set of corresponding nodes (one per location.Line).
+func CreateNodes(prof *Profile, o *Options) (Nodes, locationMap) {
+	locations := locationMap{make([]Nodes, len(prof.Location)+1), make(map[uint64]Nodes)}
+	nm := make(NodeMap, len(prof.Location))
+	for _, l := range prof.Location {
+		lines := l.Line
+		if len(lines) == 0 {
+			lines = []Line{{}} // Create empty line to include location info.
+		}
+		nodes := make(Nodes, len(lines))
+		for ln := range lines {
+			nodes[ln] = nm.findOrInsertLine(l, lines[ln], o)
+		}
+		locations.add(l.ID, nodes)
+	}
+	return nm.nodes(), locations
+}
+
+func (nm NodeMap) nodes() Nodes {
+	nodes := make(Nodes, 0, len(nm))
+	for _, n := range nm {
+		nodes = append(nodes, n)
+	}
+	return nodes
+}
+
+func (nm NodeMap) findOrInsertLine(l *Location, li Line, o *Options) *Node {
+	var objfile string
+	if m := l.Mapping; m != nil && m.File != "" {
+		objfile = m.File
+	}
+
+	if ni := nodeInfo(l, li, objfile, o); ni != nil {
+		return nm.FindOrInsertNode(*ni, o.KeptNodes)
+	}
+	return nil
+}
+
+func nodeInfo(l *Location, line Line, objfile string, o *Options) *NodeInfo {
+	if line.Function == nil {
+		return &NodeInfo{Address: l.Address}
+	}
+	ni := &NodeInfo{
+		Address: l.Address,
+		Lineno:  int(line.Line),
+		Name:    line.Function.Name,
+	}
+	ni.StartLine = int(line.Function.StartLine)
+	return ni
+}
+
+// Sum adds the flat and cum values of a set of nodes.
+func (ns Nodes) Sum() (flat int64, cum int64) {
+	for _, n := range ns {
+		flat += n.Flat
+		cum += n.Cum
+	}
+	return
+}
+
+func (n *Node) addSample(dw, w int64, flat bool) {
+	// Update sample value
+	if flat {
+		n.FlatDiv += dw
+		n.Flat += w
+	} else {
+		n.CumDiv += dw
+		n.Cum += w
+	}
+}
+
+// String returns a text representation of a graph, for debugging purposes.
+func (g *Graph) String() string {
+	var s []string
+
+	nodeIndex := make(map[*Node]int, len(g.Nodes))
+
+	for i, n := range g.Nodes {
+		nodeIndex[n] = i + 1
+	}
+
+	for i, n := range g.Nodes {
+		name := n.Info.PrintableName()
+		var in, out []int
+
+		for _, from := range n.In {
+			in = append(in, nodeIndex[from.Src])
+		}
+		for _, to := range n.Out {
+			out = append(out, nodeIndex[to.Dest])
+		}
+		s = append(s, fmt.Sprintf("%d: %s[flat=%d cum=%d] %x -> %v ", i+1, name, n.Flat, n.Cum, in, out))
+	}
+	return strings.Join(s, "\n")
+}
+
+// Sort returns a slice of the edges in the map, in a consistent
+// order. The sort order is first based on the edge weight
+// (higher-to-lower) and then by the node names to avoid flakiness.
+func (em EdgeMap) Sort() []*Edge {
+	el := make(edgeList, 0, len(em))
+	for _, w := range em {
+		el = append(el, w)
+	}
+
+	sort.Sort(el)
+	return el
+}
+
+// Sum returns the total weight for a set of nodes.
+func (em EdgeMap) Sum() int64 {
+	var ret int64
+	for _, edge := range em {
+		ret += edge.Weight
+	}
+	return ret
+}
+
+type edgeList []*Edge
+
+func (el edgeList) Len() int {
+	return len(el)
+}
+
+func (el edgeList) Less(i, j int) bool {
+	if el[i].Weight != el[j].Weight {
+		return abs64(el[i].Weight) > abs64(el[j].Weight)
+	}
+
+	from1 := el[i].Src.Info.PrintableName()
+	from2 := el[j].Src.Info.PrintableName()
+	if from1 != from2 {
+		return from1 < from2
+	}
+
+	to1 := el[i].Dest.Info.PrintableName()
+	to2 := el[j].Dest.Info.PrintableName()
+
+	return to1 < to2
+}
+
+func (el edgeList) Swap(i, j int) {
+	el[i], el[j] = el[j], el[i]
+}
+
+func abs64(i int64) int64 {
+	if i < 0 {
+		return -i
+	}
+	return i
+}

src/internal/profile/merge.go

@@ -0,0 +1,461 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package profile
+
+import (
+	"fmt"
+	"sort"
+	"strconv"
+	"strings"
+)
+
+// Merge merges all the profiles in profs into a single Profile.
+// Returns a new profile independent of the input profiles. The merged
+// profile is compacted to eliminate unused samples, locations,
+// functions and mappings. Profiles must have identical profile sample
+// and period types or the merge will fail. profile.Period of the
+// resulting profile will be the maximum of all profiles, and
+// profile.TimeNanos will be the earliest nonzero one.
+func Merge(srcs []*Profile) (*Profile, error) {
+	if len(srcs) == 0 {
+		return nil, fmt.Errorf("no profiles to merge")
+	}
+	p, err := combineHeaders(srcs)
+	if err != nil {
+		return nil, err
+	}
+
+	pm := &profileMerger{
+		p:         p,
+		samples:   make(map[sampleKey]*Sample, len(srcs[0].Sample)),
+		locations: make(map[locationKey]*Location, len(srcs[0].Location)),
+		functions: make(map[functionKey]*Function, len(srcs[0].Function)),
+		mappings:  make(map[mappingKey]*Mapping, len(srcs[0].Mapping)),
+	}
+
+	for _, src := range srcs {
+		// Clear the profile-specific hash tables
+		pm.locationsByID = make(map[uint64]*Location, len(src.Location))
+		pm.functionsByID = make(map[uint64]*Function, len(src.Function))
+		pm.mappingsByID = make(map[uint64]mapInfo, len(src.Mapping))
+
+		if len(pm.mappings) == 0 && len(src.Mapping) > 0 {
+			// The Mapping list has the property that the first mapping
+			// represents the main binary. Take the first Mapping we see,
+			// otherwise the operations below will add mappings in an
+			// arbitrary order.
+			pm.mapMapping(src.Mapping[0])
+		}
+
+		for _, s := range src.Sample {
+			if !isZeroSample(s) {
+				pm.mapSample(s)
+			}
+		}
+	}
+
+	for _, s := range p.Sample {
+		if isZeroSample(s) {
+			// If there are any zero samples, re-merge the profile to GC
+			// them.
+			return Merge([]*Profile{p})
+		}
+	}
+
+	return p, nil
+}
+
+// Normalize normalizes the source profile by multiplying each value in profile by the
+// ratio of the sum of the base profile's values of that sample type to the sum of the
+// source profile's value of that sample type.
+func (p *Profile) Normalize(pb *Profile) error {
+
+	if err := p.compatible(pb); err != nil {
+		return err
+	}
+
+	baseVals := make([]int64, len(p.SampleType))
+	for _, s := range pb.Sample {
+		for i, v := range s.Value {
+			baseVals[i] += v
+		}
+	}
+
+	srcVals := make([]int64, len(p.SampleType))
+	for _, s := range p.Sample {
+		for i, v := range s.Value {
+			srcVals[i] += v
+		}
+	}
+
+	normScale := make([]float64, len(baseVals))
+	for i := range baseVals {
+		if srcVals[i] == 0 {
+			normScale[i] = 0.0
+		} else {
+			normScale[i] = float64(baseVals[i]) / float64(srcVals[i])
+		}
+	}
+	p.ScaleN(normScale)
+	return nil
+}
+
+func isZeroSample(s *Sample) bool {
+	for _, v := range s.Value {
+		if v != 0 {
+			return false
+		}
+	}
+	return true
+}
+
+type profileMerger struct {
+	p *Profile
+
+	// Memoization tables within a profile.
+	locationsByID map[uint64]*Location
+	functionsByID map[uint64]*Function
+	mappingsByID  map[uint64]mapInfo
+
+	// Memoization tables for profile entities.
+	samples   map[sampleKey]*Sample
+	locations map[locationKey]*Location
+	functions map[functionKey]*Function
+	mappings  map[mappingKey]*Mapping
+}
+
+type mapInfo struct {
+	m      *Mapping
+	offset int64
+}
+
+func (pm *profileMerger) mapSample(src *Sample) *Sample {
+	s := &Sample{
+		Location: make([]*Location, len(src.Location)),
+		Value:    make([]int64, len(src.Value)),
+		Label:    make(map[string][]string, len(src.Label)),
+		NumLabel: make(map[string][]int64, len(src.NumLabel)),
+		NumUnit:  make(map[string][]string, len(src.NumLabel)),
+	}
+	for i, l := range src.Location {
+		s.Location[i] = pm.mapLocation(l)
+	}
+	for k, v := range src.Label {
+		vv := make([]string, len(v))
+		copy(vv, v)
+		s.Label[k] = vv
+	}
+	for k, v := range src.NumLabel {
+		u := src.NumUnit[k]
+		vv := make([]int64, len(v))
+		uu := make([]string, len(u))
+		copy(vv, v)
+		copy(uu, u)
+		s.NumLabel[k] = vv
+		s.NumUnit[k] = uu
+	}
+	// Check memoization table. Must be done on the remapped location to
+	// account for the remapped mapping. Add current values to the
+	// existing sample.
+	k := s.key()
+	if ss, ok := pm.samples[k]; ok {
+		for i, v := range src.Value {
+			ss.Value[i] += v
+		}
+		return ss
+	}
+	copy(s.Value, src.Value)
+	pm.samples[k] = s
+	pm.p.Sample = append(pm.p.Sample, s)
+	return s
+}
+
+// key generates sampleKey to be used as a key for maps.
+func (sample *Sample) key() sampleKey {
+	ids := make([]string, len(sample.Location))
+	for i, l := range sample.Location {
+		ids[i] = strconv.FormatUint(l.ID, 16)
+	}
+
+	labels := make([]string, 0, len(sample.Label))
+	for k, v := range sample.Label {
+		labels = append(labels, fmt.Sprintf("%q%q", k, v))
+	}
+	sort.Strings(labels)
+
+	numlabels := make([]string, 0, len(sample.NumLabel))
+	for k, v := range sample.NumLabel {
+		numlabels = append(numlabels, fmt.Sprintf("%q%x%x", k, v, sample.NumUnit[k]))
+	}
+	sort.Strings(numlabels)
+
+	return sampleKey{
+		strings.Join(ids, "|"),
+		strings.Join(labels, ""),
+		strings.Join(numlabels, ""),
+	}
+}
+
+type sampleKey struct {
+	locations string
+	labels    string
+	numlabels string
+}
+
+func (pm *profileMerger) mapLocation(src *Location) *Location {
+	if src == nil {
+		return nil
+	}
+
+	if l, ok := pm.locationsByID[src.ID]; ok {
+		pm.locationsByID[src.ID] = l
+		return l
+	}
+
+	mi := pm.mapMapping(src.Mapping)
+	l := &Location{
+		ID:       uint64(len(pm.p.Location) + 1),
+		Mapping:  mi.m,
+		Address:  uint64(int64(src.Address) + mi.offset),
+		Line:     make([]Line, len(src.Line)),
+		IsFolded: src.IsFolded,
+	}
+	for i, ln := range src.Line {
+		l.Line[i] = pm.mapLine(ln)
+	}
+	// Check memoization table. Must be done on the remapped location to
+	// account for the remapped mapping ID.
+	k := l.key()
+	if ll, ok := pm.locations[k]; ok {
+		pm.locationsByID[src.ID] = ll
+		return ll
+	}
+	pm.locationsByID[src.ID] = l
+	pm.locations[k] = l
+	pm.p.Location = append(pm.p.Location, l)
+	return l
+}
+
+// key generates locationKey to be used as a key for maps.
+func (l *Location) key() locationKey {
+	key := locationKey{
+		addr:     l.Address,
+		isFolded: l.IsFolded,
+	}
+	if l.Mapping != nil {
+		// Normalizes address to handle address space randomization.
+		key.addr -= l.Mapping.Start
+		key.mappingID = l.Mapping.ID
+	}
+	lines := make([]string, len(l.Line)*2)
+	for i, line := range l.Line {
+		if line.Function != nil {
+			lines[i*2] = strconv.FormatUint(line.Function.ID, 16)
+		}
+		lines[i*2+1] = strconv.FormatInt(line.Line, 16)
+	}
+	key.lines = strings.Join(lines, "|")
+	return key
+}
+
+type locationKey struct {
+	addr, mappingID uint64
+	lines           string
+	isFolded        bool
+}
+
+func (pm *profileMerger) mapMapping(src *Mapping) mapInfo {
+	if src == nil {
+		return mapInfo{}
+	}
+
+	if mi, ok := pm.mappingsByID[src.ID]; ok {
+		return mi
+	}
+
+	// Check memoization tables.
+	mk := src.key()
+	if m, ok := pm.mappings[mk]; ok {
+		mi := mapInfo{m, int64(m.Start) - int64(src.Start)}
+		pm.mappingsByID[src.ID] = mi
+		return mi
+	}
+	m := &Mapping{
+		ID:              uint64(len(pm.p.Mapping) + 1),
+		Start:           src.Start,
+		Limit:           src.Limit,
+		Offset:          src.Offset,
+		File:            src.File,
+		BuildID:         src.BuildID,
+		HasFunctions:    src.HasFunctions,
+		HasFilenames:    src.HasFilenames,
+		HasLineNumbers:  src.HasLineNumbers,
+		HasInlineFrames: src.HasInlineFrames,
+	}
+	pm.p.Mapping = append(pm.p.Mapping, m)
+
+	// Update memoization tables.
+	pm.mappings[mk] = m
+	mi := mapInfo{m, 0}
+	pm.mappingsByID[src.ID] = mi
+	return mi
+}
+
+// key generates encoded strings of Mapping to be used as a key for
+// maps.
+func (m *Mapping) key() mappingKey {
+	// Normalize addresses to handle address space randomization.
+	// Round up to next 4K boundary to avoid minor discrepancies.
+	const mapsizeRounding = 0x1000
+
+	size := m.Limit - m.Start
+	size = size + mapsizeRounding - 1
+	size = size - (size % mapsizeRounding)
+	key := mappingKey{
+		size:   size,
+		offset: m.Offset,
+	}
+
+	switch {
+	case m.BuildID != "":
+		key.buildIDOrFile = m.BuildID
+	case m.File != "":
+		key.buildIDOrFile = m.File
+	default:
+		// A mapping containing neither build ID nor file name is a fake mapping. A
+		// key with empty buildIDOrFile is used for fake mappings so that they are
+		// treated as the same mapping during merging.
+	}
+	return key
+}
+
+type mappingKey struct {
+	size, offset  uint64
+	buildIDOrFile string
+}
+
+func (pm *profileMerger) mapLine(src Line) Line {
+	ln := Line{
+		Function: pm.mapFunction(src.Function),
+		Line:     src.Line,
+	}
+	return ln
+}
+
+func (pm *profileMerger) mapFunction(src *Function) *Function {
+	if src == nil {
+		return nil
+	}
+	if f, ok := pm.functionsByID[src.ID]; ok {
+		return f
+	}
+	k := src.key()
+	if f, ok := pm.functions[k]; ok {
+		pm.functionsByID[src.ID] = f
+		return f
+	}
+	f := &Function{
+		ID:         uint64(len(pm.p.Function) + 1),
+		Name:       src.Name,
+		SystemName: src.SystemName,
+		Filename:   src.Filename,
+		StartLine:  src.StartLine,
+	}
+	pm.functions[k] = f
+	pm.functionsByID[src.ID] = f
+	pm.p.Function = append(pm.p.Function, f)
+	return f
+}
+
+// key generates a struct to be used as a key for maps.
+func (f *Function) key() functionKey {
+	return functionKey{
+		f.StartLine,
+		f.Name,
+		f.SystemName,
+		f.Filename,
+	}
+}
+
+type functionKey struct {
+	startLine                  int64
+	name, systemName, fileName string
+}
+
+// combineHeaders checks that all profiles can be merged and returns
+// their combined profile.
+func combineHeaders(srcs []*Profile) (*Profile, error) {
+	for _, s := range srcs[1:] {
+		if err := srcs[0].compatible(s); err != nil {
+			return nil, err
+		}
+	}
+
+	var timeNanos, durationNanos, period int64
+	var comments []string
+	seenComments := map[string]bool{}
+	var defaultSampleType string
+	for _, s := range srcs {
+		if timeNanos == 0 || s.TimeNanos < timeNanos {
+			timeNanos = s.TimeNanos
+		}
+		durationNanos += s.DurationNanos
+		if period == 0 || period < s.Period {
+			period = s.Period
+		}
+		for _, c := range s.Comments {
+			if seen := seenComments[c]; !seen {
+				comments = append(comments, c)
+				seenComments[c] = true
+			}
+		}
+		if defaultSampleType == "" {
+			defaultSampleType = s.DefaultSampleType
+		}
+	}
+
+	p := &Profile{
+		SampleType: make([]*ValueType, len(srcs[0].SampleType)),
+
+		DropFrames: srcs[0].DropFrames,
+		KeepFrames: srcs[0].KeepFrames,
+
+		TimeNanos:     timeNanos,
+		DurationNanos: durationNanos,
+		PeriodType:    srcs[0].PeriodType,
+		Period:        period,
+
+		Comments:          comments,
+		DefaultSampleType: defaultSampleType,
+	}
+	copy(p.SampleType, srcs[0].SampleType)
+	return p, nil
+}
+
+// compatible determines if two profiles can be compared/merged.
+// returns nil if the profiles are compatible; otherwise an error with
+// details on the incompatibility.
+func (p *Profile) compatible(pb *Profile) error {
+	if !equalValueType(p.PeriodType, pb.PeriodType) {
+		return fmt.Errorf("incompatible period types %v and %v", p.PeriodType, pb.PeriodType)
+	}
+
+	if len(p.SampleType) != len(pb.SampleType) {
+		return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
+	}
+
+	for i := range p.SampleType {
+		if !equalValueType(p.SampleType[i], pb.SampleType[i]) {
+			return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
+		}
+	}
+	return nil
+}
+
+// equalValueType returns true if the two value types are semantically
+// equal. It ignores the internal fields used during encode/decode.
+func equalValueType(st1, st2 *ValueType) bool {
+	return st1.Type == st2.Type && st1.Unit == st2.Unit
+}

src/internal/profile/profile.go

@@ -0,0 +1,568 @@
+// Copyright 2014 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package profile provides a representation of
+// github.com/google/pprof/proto/profile.proto and
+// methods to encode/decode/merge profiles in this format.
+package profile
+
+import (
+	"bytes"
+	"compress/gzip"
+	"fmt"
+	"io"
+	"strings"
+	"time"
+)
+
+// Profile is an in-memory representation of profile.proto.
+type Profile struct {
+	SampleType        []*ValueType
+	DefaultSampleType string
+	Sample            []*Sample
+	Mapping           []*Mapping
+	Location          []*Location
+	Function          []*Function
+	Comments          []string
+
+	DropFrames string
+	KeepFrames string
+
+	TimeNanos     int64
+	DurationNanos int64
+	PeriodType    *ValueType
+	Period        int64
+
+	commentX           []int64
+	dropFramesX        int64
+	keepFramesX        int64
+	stringTable        []string
+	defaultSampleTypeX int64
+}
+
+// ValueType corresponds to Profile.ValueType
+type ValueType struct {
+	Type string // cpu, wall, inuse_space, etc
+	Unit string // seconds, nanoseconds, bytes, etc
+
+	typeX int64
+	unitX int64
+}
+
+// Sample corresponds to Profile.Sample
+type Sample struct {
+	Location []*Location
+	Value    []int64
+	Label    map[string][]string
+	NumLabel map[string][]int64
+	NumUnit  map[string][]string
+
+	locationIDX []uint64
+	labelX      []Label
+}
+
+// Label corresponds to Profile.Label
+type Label struct {
+	keyX int64
+	// Exactly one of the two following values must be set
+	strX int64
+	numX int64 // Integer value for this label
+}
+
+// Mapping corresponds to Profile.Mapping
+type Mapping struct {
+	ID              uint64
+	Start           uint64
+	Limit           uint64
+	Offset          uint64
+	File            string
+	BuildID         string
+	HasFunctions    bool
+	HasFilenames    bool
+	HasLineNumbers  bool
+	HasInlineFrames bool
+
+	fileX    int64
+	buildIDX int64
+}
+
+// Location corresponds to Profile.Location
+type Location struct {
+	ID       uint64
+	Mapping  *Mapping
+	Address  uint64
+	Line     []Line
+	IsFolded bool
+
+	mappingIDX uint64
+}
+
+// Line corresponds to Profile.Line
+type Line struct {
+	Function *Function
+	Line     int64
+
+	functionIDX uint64
+}
+
+// Function corresponds to Profile.Function
+type Function struct {
+	ID         uint64
+	Name       string
+	SystemName string
+	Filename   string
+	StartLine  int64
+
+	nameX       int64
+	systemNameX int64
+	filenameX   int64
+}
+
+// Parse parses a profile and checks for its validity. The input must be an
+// encoded pprof protobuf, which may optionally be gzip-compressed.
+func Parse(r io.Reader) (*Profile, error) {
+	orig, err := io.ReadAll(r)
+	if err != nil {
+		return nil, err
+	}
+
+	if len(orig) >= 2 && orig[0] == 0x1f && orig[1] == 0x8b {
+		gz, err := gzip.NewReader(bytes.NewBuffer(orig))
+		if err != nil {
+			return nil, fmt.Errorf("decompressing profile: %v", err)
+		}
+		data, err := io.ReadAll(gz)
+		if err != nil {
+			return nil, fmt.Errorf("decompressing profile: %v", err)
+		}
+		orig = data
+	}
+
+	p, err := parseUncompressed(orig)
+	if err != nil {
+		return nil, fmt.Errorf("parsing profile: %w", err)
+	}
+
+	if err := p.CheckValid(); err != nil {
+		return nil, fmt.Errorf("malformed profile: %v", err)
+	}
+	return p, nil
+}
+
+var errMalformed = fmt.Errorf("malformed profile format")
+var ErrNoData = fmt.Errorf("empty input file")
+
+func parseUncompressed(data []byte) (*Profile, error) {
+	if len(data) == 0 {
+		return nil, ErrNoData
+	}
+
+	p := &Profile{}
+	if err := unmarshal(data, p); err != nil {
+		return nil, err
+	}
+
+	if err := p.postDecode(); err != nil {
+		return nil, err
+	}
+
+	return p, nil
+}
+
+// Write writes the profile as a gzip-compressed marshaled protobuf.
+func (p *Profile) Write(w io.Writer) error {
+	p.preEncode()
+	b := marshal(p)
+	zw := gzip.NewWriter(w)
+	defer zw.Close()
+	_, err := zw.Write(b)
+	return err
+}
+
+// CheckValid tests whether the profile is valid. Checks include, but are
+// not limited to:
+//   - len(Profile.Sample[n].value) == len(Profile.value_unit)
+//   - Sample.id has a corresponding Profile.Location
+func (p *Profile) CheckValid() error {
+	// Check that sample values are consistent
+	sampleLen := len(p.SampleType)
+	if sampleLen == 0 && len(p.Sample) != 0 {
+		return fmt.Errorf("missing sample type information")
+	}
+	for _, s := range p.Sample {
+		if len(s.Value) != sampleLen {
+			return fmt.Errorf("mismatch: sample has: %d values vs. %d types", len(s.Value), len(p.SampleType))
+		}
+	}
+
+	// Check that all mappings/locations/functions are in the tables
+	// Check that there are no duplicate ids
+	mappings := make(map[uint64]*Mapping, len(p.Mapping))
+	for _, m := range p.Mapping {
+		if m.ID == 0 {
+			return fmt.Errorf("found mapping with reserved ID=0")
+		}
+		if mappings[m.ID] != nil {
+			return fmt.Errorf("multiple mappings with same id: %d", m.ID)
+		}
+		mappings[m.ID] = m
+	}
+	functions := make(map[uint64]*Function, len(p.Function))
+	for _, f := range p.Function {
+		if f.ID == 0 {
+			return fmt.Errorf("found function with reserved ID=0")
+		}
+		if functions[f.ID] != nil {
+			return fmt.Errorf("multiple functions with same id: %d", f.ID)
+		}
+		functions[f.ID] = f
+	}
+	locations := make(map[uint64]*Location, len(p.Location))
+	for _, l := range p.Location {
+		if l.ID == 0 {
+			return fmt.Errorf("found location with reserved id=0")
+		}
+		if locations[l.ID] != nil {
+			return fmt.Errorf("multiple locations with same id: %d", l.ID)
+		}
+		locations[l.ID] = l
+		if m := l.Mapping; m != nil {
+			if m.ID == 0 || mappings[m.ID] != m {
+				return fmt.Errorf("inconsistent mapping %p: %d", m, m.ID)
+			}
+		}
+		for _, ln := range l.Line {
+			if f := ln.Function; f != nil {
+				if f.ID == 0 || functions[f.ID] != f {
+					return fmt.Errorf("inconsistent function %p: %d", f, f.ID)
+				}
+			}
+		}
+	}
+	return nil
+}
+
+// Aggregate merges the locations in the profile into equivalence
+// classes preserving the request attributes. It also updates the
+// samples to point to the merged locations.
+func (p *Profile) Aggregate(inlineFrame, function, filename, linenumber, address bool) error {
+	for _, m := range p.Mapping {
+		m.HasInlineFrames = m.HasInlineFrames && inlineFrame
+		m.HasFunctions = m.HasFunctions && function
+		m.HasFilenames = m.HasFilenames && filename
+		m.HasLineNumbers = m.HasLineNumbers && linenumber
+	}
+
+	// Aggregate functions
+	if !function || !filename {
+		for _, f := range p.Function {
+			if !function {
+				f.Name = ""
+				f.SystemName = ""
+			}
+			if !filename {
+				f.Filename = ""
+			}
+		}
+	}
+
+	// Aggregate locations
+	if !inlineFrame || !address || !linenumber {
+		for _, l := range p.Location {
+			if !inlineFrame && len(l.Line) > 1 {
+				l.Line = l.Line[len(l.Line)-1:]
+			}
+			if !linenumber {
+				for i := range l.Line {
+					l.Line[i].Line = 0
+				}
+			}
+			if !address {
+				l.Address = 0
+			}
+		}
+	}
+
+	return p.CheckValid()
+}
+
+// Print dumps a text representation of a profile. Intended mainly
+// for debugging purposes.
+func (p *Profile) String() string {
+
+	ss := make([]string, 0, len(p.Sample)+len(p.Mapping)+len(p.Location))
+	if pt := p.PeriodType; pt != nil {
+		ss = append(ss, fmt.Sprintf("PeriodType: %s %s", pt.Type, pt.Unit))
+	}
+	ss = append(ss, fmt.Sprintf("Period: %d", p.Period))
+	if p.TimeNanos != 0 {
+		ss = append(ss, fmt.Sprintf("Time: %v", time.Unix(0, p.TimeNanos)))
+	}
+	if p.DurationNanos != 0 {
+		ss = append(ss, fmt.Sprintf("Duration: %v", time.Duration(p.DurationNanos)))
+	}
+
+	ss = append(ss, "Samples:")
+	var sh1 string
+	for _, s := range p.SampleType {
+		sh1 = sh1 + fmt.Sprintf("%s/%s ", s.Type, s.Unit)
+	}
+	ss = append(ss, strings.TrimSpace(sh1))
+	for _, s := range p.Sample {
+		var sv string
+		for _, v := range s.Value {
+			sv = fmt.Sprintf("%s %10d", sv, v)
+		}
+		sv = sv + ": "
+		for _, l := range s.Location {
+			sv = sv + fmt.Sprintf("%d ", l.ID)
+		}
+		ss = append(ss, sv)
+		const labelHeader = "                "
+		if len(s.Label) > 0 {
+			ls := labelHeader
+			for k, v := range s.Label {
+				ls = ls + fmt.Sprintf("%s:%v ", k, v)
+			}
+			ss = append(ss, ls)
+		}
+		if len(s.NumLabel) > 0 {
+			ls := labelHeader
+			for k, v := range s.NumLabel {
+				ls = ls + fmt.Sprintf("%s:%v ", k, v)
+			}
+			ss = append(ss, ls)
+		}
+	}
+
+	ss = append(ss, "Locations")
+	for _, l := range p.Location {
+		locStr := fmt.Sprintf("%6d: %#x ", l.ID, l.Address)
+		if m := l.Mapping; m != nil {
+			locStr = locStr + fmt.Sprintf("M=%d ", m.ID)
+		}
+		if len(l.Line) == 0 {
+			ss = append(ss, locStr)
+		}
+		for li := range l.Line {
+			lnStr := "??"
+			if fn := l.Line[li].Function; fn != nil {
+				lnStr = fmt.Sprintf("%s %s:%d s=%d",
+					fn.Name,
+					fn.Filename,
+					l.Line[li].Line,
+					fn.StartLine)
+				if fn.Name != fn.SystemName {
+					lnStr = lnStr + "(" + fn.SystemName + ")"
+				}
+			}
+			ss = append(ss, locStr+lnStr)
+			// Do not print location details past the first line
+			locStr = "             "
+		}
+	}
+
+	ss = append(ss, "Mappings")
+	for _, m := range p.Mapping {
+		bits := ""
+		if m.HasFunctions {
+			bits += "[FN]"
+		}
+		if m.HasFilenames {
+			bits += "[FL]"
+		}
+		if m.HasLineNumbers {
+			bits += "[LN]"
+		}
+		if m.HasInlineFrames {
+			bits += "[IN]"
+		}
+		ss = append(ss, fmt.Sprintf("%d: %#x/%#x/%#x %s %s %s",
+			m.ID,
+			m.Start, m.Limit, m.Offset,
+			m.File,
+			m.BuildID,
+			bits))
+	}
+
+	return strings.Join(ss, "\n") + "\n"
+}
+
+// Merge adds profile p adjusted by ratio r into profile p. Profiles
+// must be compatible (same Type and SampleType).
+// TODO(rsilvera): consider normalizing the profiles based on the
+// total samples collected.
+func (p *Profile) Merge(pb *Profile, r float64) error {
+	if err := p.Compatible(pb); err != nil {
+		return err
+	}
+
+	pb = pb.Copy()
+
+	// Keep the largest of the two periods.
+	if pb.Period > p.Period {
+		p.Period = pb.Period
+	}
+
+	p.DurationNanos += pb.DurationNanos
+
+	p.Mapping = append(p.Mapping, pb.Mapping...)
+	for i, m := range p.Mapping {
+		m.ID = uint64(i + 1)
+	}
+	p.Location = append(p.Location, pb.Location...)
+	for i, l := range p.Location {
+		l.ID = uint64(i + 1)
+	}
+	p.Function = append(p.Function, pb.Function...)
+	for i, f := range p.Function {
+		f.ID = uint64(i + 1)
+	}
+
+	if r != 1.0 {
+		for _, s := range pb.Sample {
+			for i, v := range s.Value {
+				s.Value[i] = int64((float64(v) * r))
+			}
+		}
+	}
+	p.Sample = append(p.Sample, pb.Sample...)
+	return p.CheckValid()
+}
+
+// Compatible determines if two profiles can be compared/merged.
+// returns nil if the profiles are compatible; otherwise an error with
+// details on the incompatibility.
+func (p *Profile) Compatible(pb *Profile) error {
+	if !compatibleValueTypes(p.PeriodType, pb.PeriodType) {
+		return fmt.Errorf("incompatible period types %v and %v", p.PeriodType, pb.PeriodType)
+	}
+
+	if len(p.SampleType) != len(pb.SampleType) {
+		return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
+	}
+
+	for i := range p.SampleType {
+		if !compatibleValueTypes(p.SampleType[i], pb.SampleType[i]) {
+			return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
+		}
+	}
+
+	return nil
+}
+
+// HasFunctions determines if all locations in this profile have
+// symbolized function information.
+func (p *Profile) HasFunctions() bool {
+	for _, l := range p.Location {
+		if l.Mapping == nil || !l.Mapping.HasFunctions {
+			return false
+		}
+	}
+	return true
+}
+
+// HasFileLines determines if all locations in this profile have
+// symbolized file and line number information.
+func (p *Profile) HasFileLines() bool {
+	for _, l := range p.Location {
+		if l.Mapping == nil || (!l.Mapping.HasFilenames || !l.Mapping.HasLineNumbers) {
+			return false
+		}
+	}
+	return true
+}
+
+func compatibleValueTypes(v1, v2 *ValueType) bool {
+	if v1 == nil || v2 == nil {
+		return true // No grounds to disqualify.
+	}
+	return v1.Type == v2.Type && v1.Unit == v2.Unit
+}
+
+// Copy makes a fully independent copy of a profile.
+func (p *Profile) Copy() *Profile {
+	p.preEncode()
+	b := marshal(p)
+
+	pp := &Profile{}
+	if err := unmarshal(b, pp); err != nil {
+		panic(err)
+	}
+	if err := pp.postDecode(); err != nil {
+		panic(err)
+	}
+
+	return pp
+}
+
+// Demangler maps symbol names to a human-readable form. This may
+// include C++ demangling and additional simplification. Names that
+// are not demangled may be missing from the resulting map.
+type Demangler func(name []string) (map[string]string, error)
+
+// Demangle attempts to demangle and optionally simplify any function
+// names referenced in the profile. It works on a best-effort basis:
+// it will silently preserve the original names in case of any errors.
+func (p *Profile) Demangle(d Demangler) error {
+	// Collect names to demangle.
+	var names []string
+	for _, fn := range p.Function {
+		names = append(names, fn.SystemName)
+	}
+
+	// Update profile with demangled names.
+	demangled, err := d(names)
+	if err != nil {
+		return err
+	}
+	for _, fn := range p.Function {
+		if dd, ok := demangled[fn.SystemName]; ok {
+			fn.Name = dd
+		}
+	}
+	return nil
+}
+
+// Empty reports whether the profile contains no samples.
+func (p *Profile) Empty() bool {
+	return len(p.Sample) == 0
+}
+
+// Scale multiplies all sample values in a profile by a constant.
+func (p *Profile) Scale(ratio float64) {
+	if ratio == 1 {
+		return
+	}
+	ratios := make([]float64, len(p.SampleType))
+	for i := range p.SampleType {
+		ratios[i] = ratio
+	}
+	p.ScaleN(ratios)
+}
+
+// ScaleN multiplies each sample values in a sample by a different amount.
+func (p *Profile) ScaleN(ratios []float64) error {
+	if len(p.SampleType) != len(ratios) {
+		return fmt.Errorf("mismatched scale ratios, got %d, want %d", len(ratios), len(p.SampleType))
+	}
+	allOnes := true
+	for _, r := range ratios {
+		if r != 1 {
+			allOnes = false
+			break
+		}
+	}
+	if allOnes {
+		return nil
+	}
+	for _, s := range p.Sample {
+		for i, v := range s.Value {
+			if ratios[i] != 1 {
+				s.Value[i] = int64(float64(v) * ratios[i])
+			}
+		}
+	}
+	return nil
+}

src/internal/profile/proto.go

@@ -0,0 +1,356 @@
+// Copyright 2014 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This file is a simple protocol buffer encoder and decoder.
+//
+// A protocol message must implement the message interface:
+//   decoder() []decoder
+//   encode(*buffer)
+//
+// The decode method returns a slice indexed by field number that gives the
+// function to decode that field.
+// The encode method encodes its receiver into the given buffer.
+//
+// The two methods are simple enough to be implemented by hand rather than
+// by using a protocol compiler.
+//
+// See profile.go for examples of messages implementing this interface.
+//
+// There is no support for groups, message sets, or "has" bits.
+
+package profile
+
+import (
+	"errors"
+	"fmt"
+)
+
+type buffer struct {
+	field int
+	typ   int
+	u64   uint64
+	data  []byte
+	tmp   [16]byte
+}
+
+type decoder func(*buffer, message) error
+
+type message interface {
+	decoder() []decoder
+	encode(*buffer)
+}
+
+func marshal(m message) []byte {
+	var b buffer
+	m.encode(&b)
+	return b.data
+}
+
+func encodeVarint(b *buffer, x uint64) {
+	for x >= 128 {
+		b.data = append(b.data, byte(x)|0x80)
+		x >>= 7
+	}
+	b.data = append(b.data, byte(x))
+}
+
+func encodeLength(b *buffer, tag int, len int) {
+	encodeVarint(b, uint64(tag)<<3|2)
+	encodeVarint(b, uint64(len))
+}
+
+func encodeUint64(b *buffer, tag int, x uint64) {
+	// append varint to b.data
+	encodeVarint(b, uint64(tag)<<3|0)
+	encodeVarint(b, x)
+}
+
+func encodeUint64s(b *buffer, tag int, x []uint64) {
+	if len(x) > 2 {
+		// Use packed encoding
+		n1 := len(b.data)
+		for _, u := range x {
+			encodeVarint(b, u)
+		}
+		n2 := len(b.data)
+		encodeLength(b, tag, n2-n1)
+		n3 := len(b.data)
+		copy(b.tmp[:], b.data[n2:n3])
+		copy(b.data[n1+(n3-n2):], b.data[n1:n2])
+		copy(b.data[n1:], b.tmp[:n3-n2])
+		return
+	}
+	for _, u := range x {
+		encodeUint64(b, tag, u)
+	}
+}
+
+func encodeUint64Opt(b *buffer, tag int, x uint64) {
+	if x == 0 {
+		return
+	}
+	encodeUint64(b, tag, x)
+}
+
+func encodeInt64(b *buffer, tag int, x int64) {
+	u := uint64(x)
+	encodeUint64(b, tag, u)
+}
+
+func encodeInt64Opt(b *buffer, tag int, x int64) {
+	if x == 0 {
+		return
+	}
+	encodeInt64(b, tag, x)
+}
+
+func encodeInt64s(b *buffer, tag int, x []int64) {
+	if len(x) > 2 {
+		// Use packed encoding
+		n1 := len(b.data)
+		for _, u := range x {
+			encodeVarint(b, uint64(u))
+		}
+		n2 := len(b.data)
+		encodeLength(b, tag, n2-n1)
+		n3 := len(b.data)
+		copy(b.tmp[:], b.data[n2:n3])
+		copy(b.data[n1+(n3-n2):], b.data[n1:n2])
+		copy(b.data[n1:], b.tmp[:n3-n2])
+		return
+	}
+	for _, u := range x {
+		encodeInt64(b, tag, u)
+	}
+}
+
+func encodeString(b *buffer, tag int, x string) {
+	encodeLength(b, tag, len(x))
+	b.data = append(b.data, x...)
+}
+
+func encodeStrings(b *buffer, tag int, x []string) {
+	for _, s := range x {
+		encodeString(b, tag, s)
+	}
+}
+
+func encodeBool(b *buffer, tag int, x bool) {
+	if x {
+		encodeUint64(b, tag, 1)
+	} else {
+		encodeUint64(b, tag, 0)
+	}
+}
+
+func encodeBoolOpt(b *buffer, tag int, x bool) {
+	if !x {
+		return
+	}
+	encodeBool(b, tag, x)
+}
+
+func encodeMessage(b *buffer, tag int, m message) {
+	n1 := len(b.data)
+	m.encode(b)
+	n2 := len(b.data)
+	encodeLength(b, tag, n2-n1)
+	n3 := len(b.data)
+	copy(b.tmp[:], b.data[n2:n3])
+	copy(b.data[n1+(n3-n2):], b.data[n1:n2])
+	copy(b.data[n1:], b.tmp[:n3-n2])
+}
+
+func unmarshal(data []byte, m message) (err error) {
+	b := buffer{data: data, typ: 2}
+	return decodeMessage(&b, m)
+}
+
+func le64(p []byte) uint64 {
+	return uint64(p[0]) | uint64(p[1])<<8 | uint64(p[2])<<16 | uint64(p[3])<<24 | uint64(p[4])<<32 | uint64(p[5])<<40 | uint64(p[6])<<48 | uint64(p[7])<<56
+}
+
+func le32(p []byte) uint32 {
+	return uint32(p[0]) | uint32(p[1])<<8 | uint32(p[2])<<16 | uint32(p[3])<<24
+}
+
+func decodeVarint(data []byte) (uint64, []byte, error) {
+	var i int
+	var u uint64
+	for i = 0; ; i++ {
+		if i >= 10 || i >= len(data) {
+			return 0, nil, errors.New("bad varint")
+		}
+		u |= uint64(data[i]&0x7F) << uint(7*i)
+		if data[i]&0x80 == 0 {
+			return u, data[i+1:], nil
+		}
+	}
+}
+
+func decodeField(b *buffer, data []byte) ([]byte, error) {
+	x, data, err := decodeVarint(data)
+	if err != nil {
+		return nil, err
+	}
+	b.field = int(x >> 3)
+	b.typ = int(x & 7)
+	b.data = nil
+	b.u64 = 0
+	switch b.typ {
+	case 0:
+		b.u64, data, err = decodeVarint(data)
+		if err != nil {
+			return nil, err
+		}
+	case 1:
+		if len(data) < 8 {
+			return nil, errors.New("not enough data")
+		}
+		b.u64 = le64(data[:8])
+		data = data[8:]
+	case 2:
+		var n uint64
+		n, data, err = decodeVarint(data)
+		if err != nil {
+			return nil, err
+		}
+		if n > uint64(len(data)) {
+			return nil, errors.New("too much data")
+		}
+		b.data = data[:n]
+		data = data[n:]
+	case 5:
+		if len(data) < 4 {
+			return nil, errors.New("not enough data")
+		}
+		b.u64 = uint64(le32(data[:4]))
+		data = data[4:]
+	default:
+		return nil, fmt.Errorf("unknown wire type: %d", b.typ)
+	}
+
+	return data, nil
+}
+
+func checkType(b *buffer, typ int) error {
+	if b.typ != typ {
+		return errors.New("type mismatch")
+	}
+	return nil
+}
+
+func decodeMessage(b *buffer, m message) error {
+	if err := checkType(b, 2); err != nil {
+		return err
+	}
+	dec := m.decoder()
+	data := b.data
+	for len(data) > 0 {
+		// pull varint field# + type
+		var err error
+		data, err = decodeField(b, data)
+		if err != nil {
+			return err
+		}
+		if b.field >= len(dec) || dec[b.field] == nil {
+			continue
+		}
+		if err := dec[b.field](b, m); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func decodeInt64(b *buffer, x *int64) error {
+	if err := checkType(b, 0); err != nil {
+		return err
+	}
+	*x = int64(b.u64)
+	return nil
+}
+
+func decodeInt64s(b *buffer, x *[]int64) error {
+	if b.typ == 2 {
+		// Packed encoding
+		data := b.data
+		for len(data) > 0 {
+			var u uint64
+			var err error
+
+			if u, data, err = decodeVarint(data); err != nil {
+				return err
+			}
+			*x = append(*x, int64(u))
+		}
+		return nil
+	}
+	var i int64
+	if err := decodeInt64(b, &i); err != nil {
+		return err
+	}
+	*x = append(*x, i)
+	return nil
+}
+
+func decodeUint64(b *buffer, x *uint64) error {
+	if err := checkType(b, 0); err != nil {
+		return err
+	}
+	*x = b.u64
+	return nil
+}
+
+func decodeUint64s(b *buffer, x *[]uint64) error {
+	if b.typ == 2 {
+		data := b.data
+		// Packed encoding
+		for len(data) > 0 {
+			var u uint64
+			var err error
+
+			if u, data, err = decodeVarint(data); err != nil {
+				return err
+			}
+			*x = append(*x, u)
+		}
+		return nil
+	}
+	var u uint64
+	if err := decodeUint64(b, &u); err != nil {
+		return err
+	}
+	*x = append(*x, u)
+	return nil
+}
+
+func decodeString(b *buffer, x *string) error {
+	if err := checkType(b, 2); err != nil {
+		return err
+	}
+	*x = string(b.data)
+	return nil
+}
+
+func decodeStrings(b *buffer, x *[]string) error {
+	var s string
+	if err := decodeString(b, &s); err != nil {
+		return err
+	}
+	*x = append(*x, s)
+	return nil
+}
+
+func decodeBool(b *buffer, x *bool) error {
+	if err := checkType(b, 0); err != nil {
+		return err
+	}
+	if int64(b.u64) == 0 {
+		*x = false
+	} else {
+		*x = true
+	}
+	return nil
+}

src/internal/profile/proto_test.go

@@ -0,0 +1,71 @@
+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package profile
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestPackedEncoding(t *testing.T) {
+
+	type testcase struct {
+		uint64s []uint64
+		int64s  []int64
+		encoded []byte
+	}
+	for i, tc := range []testcase{
+		{
+			[]uint64{0, 1, 10, 100, 1000, 10000},
+			[]int64{1000, 0, 1000},
+			[]byte{10, 8, 0, 1, 10, 100, 232, 7, 144, 78, 18, 5, 232, 7, 0, 232, 7},
+		},
+		{
+			[]uint64{10000},
+			nil,
+			[]byte{8, 144, 78},
+		},
+		{
+			nil,
+			[]int64{-10000},
+			[]byte{16, 240, 177, 255, 255, 255, 255, 255, 255, 255, 1},
+		},
+	} {
+		source := &packedInts{tc.uint64s, tc.int64s}
+		if got, want := marshal(source), tc.encoded; !reflect.DeepEqual(got, want) {
+			t.Errorf("failed encode %d, got %v, want %v", i, got, want)
+		}
+
+		dest := new(packedInts)
+		if err := unmarshal(tc.encoded, dest); err != nil {
+			t.Errorf("failed decode %d: %v", i, err)
+			continue
+		}
+		if got, want := dest.uint64s, tc.uint64s; !reflect.DeepEqual(got, want) {
+			t.Errorf("failed decode uint64s %d, got %v, want %v", i, got, want)
+		}
+		if got, want := dest.int64s, tc.int64s; !reflect.DeepEqual(got, want) {
+			t.Errorf("failed decode int64s %d, got %v, want %v", i, got, want)
+		}
+	}
+}
+
+type packedInts struct {
+	uint64s []uint64
+	int64s  []int64
+}
+
+func (u *packedInts) decoder() []decoder {
+	return []decoder{
+		nil,
+		func(b *buffer, m message) error { return decodeUint64s(b, &m.(*packedInts).uint64s) },
+		func(b *buffer, m message) error { return decodeInt64s(b, &m.(*packedInts).int64s) },
+	}
+}
+
+func (u *packedInts) encode(b *buffer) {
+	encodeUint64s(b, 1, u.uint64s)
+	encodeInt64s(b, 2, u.int64s)
+}

src/internal/profile/prune.go

@@ -0,0 +1,97 @@
+// Copyright 2014 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Implements methods to remove frames from profiles.
+
+package profile
+
+import (
+	"fmt"
+	"regexp"
+)
+
+// Prune removes all nodes beneath a node matching dropRx, and not
+// matching keepRx. If the root node of a Sample matches, the sample
+// will have an empty stack.
+func (p *Profile) Prune(dropRx, keepRx *regexp.Regexp) {
+	prune := make(map[uint64]bool)
+	pruneBeneath := make(map[uint64]bool)
+
+	for _, loc := range p.Location {
+		var i int
+		for i = len(loc.Line) - 1; i >= 0; i-- {
+			if fn := loc.Line[i].Function; fn != nil && fn.Name != "" {
+				funcName := fn.Name
+				// Account for leading '.' on the PPC ELF v1 ABI.
+				if funcName[0] == '.' {
+					funcName = funcName[1:]
+				}
+				if dropRx.MatchString(funcName) {
+					if keepRx == nil || !keepRx.MatchString(funcName) {
+						break
+					}
+				}
+			}
+		}
+
+		if i >= 0 {
+			// Found matching entry to prune.
+			pruneBeneath[loc.ID] = true
+
+			// Remove the matching location.
+			if i == len(loc.Line)-1 {
+				// Matched the top entry: prune the whole location.
+				prune[loc.ID] = true
+			} else {
+				loc.Line = loc.Line[i+1:]
+			}
+		}
+	}
+
+	// Prune locs from each Sample
+	for _, sample := range p.Sample {
+		// Scan from the root to the leaves to find the prune location.
+		// Do not prune frames before the first user frame, to avoid
+		// pruning everything.
+		foundUser := false
+		for i := len(sample.Location) - 1; i >= 0; i-- {
+			id := sample.Location[i].ID
+			if !prune[id] && !pruneBeneath[id] {
+				foundUser = true
+				continue
+			}
+			if !foundUser {
+				continue
+			}
+			if prune[id] {
+				sample.Location = sample.Location[i+1:]
+				break
+			}
+			if pruneBeneath[id] {
+				sample.Location = sample.Location[i:]
+				break
+			}
+		}
+	}
+}
+
+// RemoveUninteresting prunes and elides profiles using built-in
+// tables of uninteresting function names.
+func (p *Profile) RemoveUninteresting() error {
+	var keep, drop *regexp.Regexp
+	var err error
+
+	if p.DropFrames != "" {
+		if drop, err = regexp.Compile("^(" + p.DropFrames + ")$"); err != nil {
+			return fmt.Errorf("failed to compile regexp %s: %v", p.DropFrames, err)
+		}
+		if p.KeepFrames != "" {
+			if keep, err = regexp.Compile("^(" + p.KeepFrames + ")$"); err != nil {
+				return fmt.Errorf("failed to compile regexp %s: %v", p.KeepFrames, err)
+			}
+		}
+		p.Prune(drop, keep)
+	}
+	return nil
+}