1
0
mirror of https://github.com/golang/go synced 2024-11-19 06:04:39 -07:00
go/pointer/api.go

229 lines
6.1 KiB
Go
Raw Normal View History

// Copyright 2013 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 pointer
import (
"fmt"
"go/token"
"io"
"code.google.com/p/go.tools/call"
"code.google.com/p/go.tools/go/types/typemap"
"code.google.com/p/go.tools/ssa"
)
// A Config formulates a pointer analysis problem for Analyze().
type Config struct {
// Mains contains the set of 'main' packages to analyze
// Clients must provide the analysis with at least one
// package defining a main() function.
Mains []*ssa.Package
// Reflection determines whether to handle reflection
// operators soundly, which is currently rather slow since it
// causes constraint to be generated during solving
// proportional to the number of constraint variables, which
// has not yet been reduced by presolver optimisation.
Reflection bool
// BuildCallGraph determines whether to construct a callgraph.
// If enabled, the graph will be available in Result.CallGraph.
BuildCallGraph bool
// Print is invoked during the analysis for each discovered
// call to the built-in print(x), providing a convenient way
// to identify arbitrary expressions of interest in the tests.
//
// Pointer p may be saved until the analysis is complete, at
// which point its methods provide access to the analysis
// (The result of callings its methods within the Print
// callback is undefined.) p is nil if x is non-pointerlike.
//
Print func(site *ssa.CallCommon, p Pointer)
// The client populates Queries[v] for each ssa.Value v of
// interest.
//
// The boolean (Indirect) indicates whether to compute the
// points-to set for v (false) or *v (true): the latter is
// typically wanted for Values corresponding to source-level
// lvalues, e.g. an *ssa.Global.
//
// The pointer analysis will populate the corresponding
// Results.Queries value when it creates the pointer variable
// for v or *v. Upon completion the client can inspect that
// map for the results.
//
// If a Value belongs to a function that the analysis treats
// context-sensitively, the corresponding Results.Queries slice
// may have multiple Pointers, one per distinct context. Use
// PointsToCombined to merge them.
//
Queries map[ssa.Value]Indirect
// If Log is non-nil, log messages are written to it.
// Logging is extremely verbose.
Log io.Writer
}
type Indirect bool // map[ssa.Value]Indirect is not a set
func (c *Config) prog() *ssa.Program {
for _, main := range c.Mains {
return main.Prog
}
panic("empty scope")
}
type Warning struct {
Pos token.Pos
Message string
}
// A Result contains the results of a pointer analysis.
//
// See Config for how to request the various Result components.
//
type Result struct {
CallGraph call.Graph // discovered call graph
Queries map[ssa.Value][]Pointer // points-to sets for queried ssa.Values
Warnings []Warning // warnings of unsoundness
}
// A Pointer is an equivalence class of pointerlike values.
type Pointer interface {
// PointsTo returns the points-to set of this pointer.
PointsTo() PointsToSet
// MayAlias reports whether the receiver pointer may alias
// the argument pointer.
MayAlias(Pointer) bool
go.tools/pointer: strength reduction during constraint generation. Motivation: simple constraints---copy and addr---are more amenable to pre-solver optimizations (forthcoming) than complex constraints: load, store, and all others. In code such as the following: t0 = new struct { x, y int } t1 = &t0.y t2 = *t1 there's no need for the full generality of a (complex) load constraint for t2=*t1 since t1 can only point to t0.y. All we need is a (simple) copy constraint t2 = (t0.y) where (t0.y) is the object node label for that field. For all "addressable" SSA instructions, we tabulate whether their points-to set is necessarily a singleton. For some (e.g. Alloc, MakeSlice, etc) this is always true by design. For others (e.g. FieldAddr) it depends on their operands. We exploit this information when generating constraints: all load-form and store-form constraints are reduced to copy constraints if the pointer's PTS is a singleton. Similarly all FieldAddr (y=&x.f) and IndexAddr (y=&x[0]) constraints are reduced to offset addition, for singleton operands. Here's the constraint mix when running on the oracle itself. The total number of constraints is unchanged but the fraction that are complex has gone down to 21% from 53%. before after --simple-- addr 20682 46949 copy 61454 91211 --complex-- offsetAddr 41621 15325 load 18769 12925 store 30758 6908 invoke 758 760 typeAssert 1688 1689 total 175832 175869 Also: - Add Pointer.Context() for local variables, since we now plumb cgnodes throughout. Nice. - Refactor all load-form (load, receive, lookup) and store-form (Store, send, MapUpdate) constraints to use genLoad and genStore. - Log counts of constraints by type. - valNodes split into localval and globalval maps; localval is purged after each function. - analogous maps localobj[v] and globalobj[v] hold sole label for pts(v), if singleton. - fnObj map subsumed by globalobj. - make{Function/Global/Constant} inlined into objectValue. Much cleaner. R=crawshaw CC=golang-dev https://golang.org/cl/13979043
2013-09-27 09:33:01 -06:00
// Context returns the context of this pointer,
// if it corresponds to a local variable.
Context() call.GraphNode
String() string
}
// A PointsToSet is a set of labels (locations or allocations).
//
type PointsToSet interface {
// PointsTo returns the set of labels that this points-to set
// contains.
Labels() []*Label
// Intersects reports whether this points-to set and the
// argument points-to set contain common members.
Intersects(PointsToSet) bool
go.tools/pointer: reflection, part 1: maps, and some core features. Core: reflect.TypeOf reflect.ValueOf reflect.Zero reflect.Value.Interface Maps: (reflect.Value).MapIndex (reflect.Value).MapKeys (reflect.Value).SetMapIndex (*reflect.rtype).Elem (*reflect.rtype).Key + tests: pointer/testdata/mapreflect.go. oracle/testdata/src/main/reflection.go. Interface objects (T, V...) have been renamed "tagged objects". Abstraction: we model reflect.Value similar to interface{}---as a pointer that points only to tagged objects---but a reflect.Value may also point to an "indirect tagged object", one in which the payload V is of type *T not T. These are required because reflect.Values can hold lvalues, e.g. when derived via Field() or Elem(), though we won't use them till we get to structs and pointers. Solving: each reflection intrinsic defines a new constraint and resolution rule. Because of the nature of reflection, generalizing across types, the resolution rules dynamically create additional complex constraints during solving, where previously only simple (copy) constraints were created. This requires some solver changes: The work done before the main solver loop (to attach new constraints to the graph) is now done before each iteration, in processNewConstraints. Its loop over constraints is broken into two passes: the first handles base (addr-of) constraints, the second handles simple and complex constraints. constraint.init() has been inlined. The only behaviour that varies across constraints is ptr() Sadly this will pessimize presolver optimisations, when we get there; such is the price of reflection. Objects: reflection intrinsics create objects (i.e. cause memory allocations) with no SSA operation. We will represent them as the cgnode of the instrinsic (e.g. reflect.New), so we extend Labels and node.data to represent objects as a product (not sum) of ssa.Value and cgnode and pull this out into its own type, struct object. This simplifies a number of invariants and saves space. The ntObject flag is now represented by obj!=nil; the other flags are moved into object. cgnodes are now always recorded in objects/Labels for which it is appropriate (all but those for globals, constants and the shared contours for functions). Also: - Prepopulate the flattenMemo cache to consider reflect.Value a fake pointer, not a struct. - Improve accessors and documentation on type Label. - @conctypes assertions renamed @types (since dyn. types needn't be concrete). - add oracle 'describe' test on an interface (missing, an oversight). R=crawshaw CC=golang-dev https://golang.org/cl/13418048
2013-09-16 07:49:10 -06:00
// If this PointsToSet came from a Pointer of interface kind
// or a reflect.Value, DynamicTypes returns the set of dynamic
// types that it may contain. (For an interface, they will
// always be concrete types.)
//
// The result is a mapping whose keys are the dynamic types to
// which it may point. For each pointer-like key type, the
// corresponding map value is a set of pointer abstractions of
// that dynamic type, represented as a []Pointer slice. Use
// PointsToCombined to merge them.
//
// The result is empty unless CanHaveDynamicTypes(T).
//
DynamicTypes() *typemap.M
}
// Union returns the set containing all the elements of each set in sets.
func Union(sets ...PointsToSet) PointsToSet {
var union ptset
for _, set := range sets {
set := set.(ptset)
union.a = set.a
union.pts.addAll(set.pts)
}
return union
}
// PointsToCombined returns the combined points-to set of all the
// specified pointers.
func PointsToCombined(ptrs []Pointer) PointsToSet {
var ptsets []PointsToSet
for _, ptr := range ptrs {
ptsets = append(ptsets, ptr.PointsTo())
}
return Union(ptsets...)
}
// ---- PointsToSet public interface
type ptset struct {
a *analysis // may be nil if pts is nil
pts nodeset
}
func (s ptset) Labels() []*Label {
var labels []*Label
for l := range s.pts {
go.tools/pointer: reflection, part 1: maps, and some core features. Core: reflect.TypeOf reflect.ValueOf reflect.Zero reflect.Value.Interface Maps: (reflect.Value).MapIndex (reflect.Value).MapKeys (reflect.Value).SetMapIndex (*reflect.rtype).Elem (*reflect.rtype).Key + tests: pointer/testdata/mapreflect.go. oracle/testdata/src/main/reflection.go. Interface objects (T, V...) have been renamed "tagged objects". Abstraction: we model reflect.Value similar to interface{}---as a pointer that points only to tagged objects---but a reflect.Value may also point to an "indirect tagged object", one in which the payload V is of type *T not T. These are required because reflect.Values can hold lvalues, e.g. when derived via Field() or Elem(), though we won't use them till we get to structs and pointers. Solving: each reflection intrinsic defines a new constraint and resolution rule. Because of the nature of reflection, generalizing across types, the resolution rules dynamically create additional complex constraints during solving, where previously only simple (copy) constraints were created. This requires some solver changes: The work done before the main solver loop (to attach new constraints to the graph) is now done before each iteration, in processNewConstraints. Its loop over constraints is broken into two passes: the first handles base (addr-of) constraints, the second handles simple and complex constraints. constraint.init() has been inlined. The only behaviour that varies across constraints is ptr() Sadly this will pessimize presolver optimisations, when we get there; such is the price of reflection. Objects: reflection intrinsics create objects (i.e. cause memory allocations) with no SSA operation. We will represent them as the cgnode of the instrinsic (e.g. reflect.New), so we extend Labels and node.data to represent objects as a product (not sum) of ssa.Value and cgnode and pull this out into its own type, struct object. This simplifies a number of invariants and saves space. The ntObject flag is now represented by obj!=nil; the other flags are moved into object. cgnodes are now always recorded in objects/Labels for which it is appropriate (all but those for globals, constants and the shared contours for functions). Also: - Prepopulate the flattenMemo cache to consider reflect.Value a fake pointer, not a struct. - Improve accessors and documentation on type Label. - @conctypes assertions renamed @types (since dyn. types needn't be concrete). - add oracle 'describe' test on an interface (missing, an oversight). R=crawshaw CC=golang-dev https://golang.org/cl/13418048
2013-09-16 07:49:10 -06:00
labels = append(labels, s.a.labelFor(l))
}
return labels
}
go.tools/pointer: reflection, part 1: maps, and some core features. Core: reflect.TypeOf reflect.ValueOf reflect.Zero reflect.Value.Interface Maps: (reflect.Value).MapIndex (reflect.Value).MapKeys (reflect.Value).SetMapIndex (*reflect.rtype).Elem (*reflect.rtype).Key + tests: pointer/testdata/mapreflect.go. oracle/testdata/src/main/reflection.go. Interface objects (T, V...) have been renamed "tagged objects". Abstraction: we model reflect.Value similar to interface{}---as a pointer that points only to tagged objects---but a reflect.Value may also point to an "indirect tagged object", one in which the payload V is of type *T not T. These are required because reflect.Values can hold lvalues, e.g. when derived via Field() or Elem(), though we won't use them till we get to structs and pointers. Solving: each reflection intrinsic defines a new constraint and resolution rule. Because of the nature of reflection, generalizing across types, the resolution rules dynamically create additional complex constraints during solving, where previously only simple (copy) constraints were created. This requires some solver changes: The work done before the main solver loop (to attach new constraints to the graph) is now done before each iteration, in processNewConstraints. Its loop over constraints is broken into two passes: the first handles base (addr-of) constraints, the second handles simple and complex constraints. constraint.init() has been inlined. The only behaviour that varies across constraints is ptr() Sadly this will pessimize presolver optimisations, when we get there; such is the price of reflection. Objects: reflection intrinsics create objects (i.e. cause memory allocations) with no SSA operation. We will represent them as the cgnode of the instrinsic (e.g. reflect.New), so we extend Labels and node.data to represent objects as a product (not sum) of ssa.Value and cgnode and pull this out into its own type, struct object. This simplifies a number of invariants and saves space. The ntObject flag is now represented by obj!=nil; the other flags are moved into object. cgnodes are now always recorded in objects/Labels for which it is appropriate (all but those for globals, constants and the shared contours for functions). Also: - Prepopulate the flattenMemo cache to consider reflect.Value a fake pointer, not a struct. - Improve accessors and documentation on type Label. - @conctypes assertions renamed @types (since dyn. types needn't be concrete). - add oracle 'describe' test on an interface (missing, an oversight). R=crawshaw CC=golang-dev https://golang.org/cl/13418048
2013-09-16 07:49:10 -06:00
func (s ptset) DynamicTypes() *typemap.M {
var tmap typemap.M
tmap.SetHasher(s.a.hasher)
for ifaceObjId := range s.pts {
if !s.a.isTaggedObject(ifaceObjId) {
go.tools/pointer: reflection, part 1: maps, and some core features. Core: reflect.TypeOf reflect.ValueOf reflect.Zero reflect.Value.Interface Maps: (reflect.Value).MapIndex (reflect.Value).MapKeys (reflect.Value).SetMapIndex (*reflect.rtype).Elem (*reflect.rtype).Key + tests: pointer/testdata/mapreflect.go. oracle/testdata/src/main/reflection.go. Interface objects (T, V...) have been renamed "tagged objects". Abstraction: we model reflect.Value similar to interface{}---as a pointer that points only to tagged objects---but a reflect.Value may also point to an "indirect tagged object", one in which the payload V is of type *T not T. These are required because reflect.Values can hold lvalues, e.g. when derived via Field() or Elem(), though we won't use them till we get to structs and pointers. Solving: each reflection intrinsic defines a new constraint and resolution rule. Because of the nature of reflection, generalizing across types, the resolution rules dynamically create additional complex constraints during solving, where previously only simple (copy) constraints were created. This requires some solver changes: The work done before the main solver loop (to attach new constraints to the graph) is now done before each iteration, in processNewConstraints. Its loop over constraints is broken into two passes: the first handles base (addr-of) constraints, the second handles simple and complex constraints. constraint.init() has been inlined. The only behaviour that varies across constraints is ptr() Sadly this will pessimize presolver optimisations, when we get there; such is the price of reflection. Objects: reflection intrinsics create objects (i.e. cause memory allocations) with no SSA operation. We will represent them as the cgnode of the instrinsic (e.g. reflect.New), so we extend Labels and node.data to represent objects as a product (not sum) of ssa.Value and cgnode and pull this out into its own type, struct object. This simplifies a number of invariants and saves space. The ntObject flag is now represented by obj!=nil; the other flags are moved into object. cgnodes are now always recorded in objects/Labels for which it is appropriate (all but those for globals, constants and the shared contours for functions). Also: - Prepopulate the flattenMemo cache to consider reflect.Value a fake pointer, not a struct. - Improve accessors and documentation on type Label. - @conctypes assertions renamed @types (since dyn. types needn't be concrete). - add oracle 'describe' test on an interface (missing, an oversight). R=crawshaw CC=golang-dev https://golang.org/cl/13418048
2013-09-16 07:49:10 -06:00
continue // !CanHaveDynamicTypes(tDyn)
}
tDyn, v, indirect := s.a.taggedValue(ifaceObjId)
go.tools/pointer: reflection, part 1: maps, and some core features. Core: reflect.TypeOf reflect.ValueOf reflect.Zero reflect.Value.Interface Maps: (reflect.Value).MapIndex (reflect.Value).MapKeys (reflect.Value).SetMapIndex (*reflect.rtype).Elem (*reflect.rtype).Key + tests: pointer/testdata/mapreflect.go. oracle/testdata/src/main/reflection.go. Interface objects (T, V...) have been renamed "tagged objects". Abstraction: we model reflect.Value similar to interface{}---as a pointer that points only to tagged objects---but a reflect.Value may also point to an "indirect tagged object", one in which the payload V is of type *T not T. These are required because reflect.Values can hold lvalues, e.g. when derived via Field() or Elem(), though we won't use them till we get to structs and pointers. Solving: each reflection intrinsic defines a new constraint and resolution rule. Because of the nature of reflection, generalizing across types, the resolution rules dynamically create additional complex constraints during solving, where previously only simple (copy) constraints were created. This requires some solver changes: The work done before the main solver loop (to attach new constraints to the graph) is now done before each iteration, in processNewConstraints. Its loop over constraints is broken into two passes: the first handles base (addr-of) constraints, the second handles simple and complex constraints. constraint.init() has been inlined. The only behaviour that varies across constraints is ptr() Sadly this will pessimize presolver optimisations, when we get there; such is the price of reflection. Objects: reflection intrinsics create objects (i.e. cause memory allocations) with no SSA operation. We will represent them as the cgnode of the instrinsic (e.g. reflect.New), so we extend Labels and node.data to represent objects as a product (not sum) of ssa.Value and cgnode and pull this out into its own type, struct object. This simplifies a number of invariants and saves space. The ntObject flag is now represented by obj!=nil; the other flags are moved into object. cgnodes are now always recorded in objects/Labels for which it is appropriate (all but those for globals, constants and the shared contours for functions). Also: - Prepopulate the flattenMemo cache to consider reflect.Value a fake pointer, not a struct. - Improve accessors and documentation on type Label. - @conctypes assertions renamed @types (since dyn. types needn't be concrete). - add oracle 'describe' test on an interface (missing, an oversight). R=crawshaw CC=golang-dev https://golang.org/cl/13418048
2013-09-16 07:49:10 -06:00
if indirect {
panic("indirect tagged object") // implement later
}
go.tools/pointer: reflection, part 1: maps, and some core features. Core: reflect.TypeOf reflect.ValueOf reflect.Zero reflect.Value.Interface Maps: (reflect.Value).MapIndex (reflect.Value).MapKeys (reflect.Value).SetMapIndex (*reflect.rtype).Elem (*reflect.rtype).Key + tests: pointer/testdata/mapreflect.go. oracle/testdata/src/main/reflection.go. Interface objects (T, V...) have been renamed "tagged objects". Abstraction: we model reflect.Value similar to interface{}---as a pointer that points only to tagged objects---but a reflect.Value may also point to an "indirect tagged object", one in which the payload V is of type *T not T. These are required because reflect.Values can hold lvalues, e.g. when derived via Field() or Elem(), though we won't use them till we get to structs and pointers. Solving: each reflection intrinsic defines a new constraint and resolution rule. Because of the nature of reflection, generalizing across types, the resolution rules dynamically create additional complex constraints during solving, where previously only simple (copy) constraints were created. This requires some solver changes: The work done before the main solver loop (to attach new constraints to the graph) is now done before each iteration, in processNewConstraints. Its loop over constraints is broken into two passes: the first handles base (addr-of) constraints, the second handles simple and complex constraints. constraint.init() has been inlined. The only behaviour that varies across constraints is ptr() Sadly this will pessimize presolver optimisations, when we get there; such is the price of reflection. Objects: reflection intrinsics create objects (i.e. cause memory allocations) with no SSA operation. We will represent them as the cgnode of the instrinsic (e.g. reflect.New), so we extend Labels and node.data to represent objects as a product (not sum) of ssa.Value and cgnode and pull this out into its own type, struct object. This simplifies a number of invariants and saves space. The ntObject flag is now represented by obj!=nil; the other flags are moved into object. cgnodes are now always recorded in objects/Labels for which it is appropriate (all but those for globals, constants and the shared contours for functions). Also: - Prepopulate the flattenMemo cache to consider reflect.Value a fake pointer, not a struct. - Improve accessors and documentation on type Label. - @conctypes assertions renamed @types (since dyn. types needn't be concrete). - add oracle 'describe' test on an interface (missing, an oversight). R=crawshaw CC=golang-dev https://golang.org/cl/13418048
2013-09-16 07:49:10 -06:00
prev, _ := tmap.At(tDyn).([]Pointer)
go.tools/pointer: strength reduction during constraint generation. Motivation: simple constraints---copy and addr---are more amenable to pre-solver optimizations (forthcoming) than complex constraints: load, store, and all others. In code such as the following: t0 = new struct { x, y int } t1 = &t0.y t2 = *t1 there's no need for the full generality of a (complex) load constraint for t2=*t1 since t1 can only point to t0.y. All we need is a (simple) copy constraint t2 = (t0.y) where (t0.y) is the object node label for that field. For all "addressable" SSA instructions, we tabulate whether their points-to set is necessarily a singleton. For some (e.g. Alloc, MakeSlice, etc) this is always true by design. For others (e.g. FieldAddr) it depends on their operands. We exploit this information when generating constraints: all load-form and store-form constraints are reduced to copy constraints if the pointer's PTS is a singleton. Similarly all FieldAddr (y=&x.f) and IndexAddr (y=&x[0]) constraints are reduced to offset addition, for singleton operands. Here's the constraint mix when running on the oracle itself. The total number of constraints is unchanged but the fraction that are complex has gone down to 21% from 53%. before after --simple-- addr 20682 46949 copy 61454 91211 --complex-- offsetAddr 41621 15325 load 18769 12925 store 30758 6908 invoke 758 760 typeAssert 1688 1689 total 175832 175869 Also: - Add Pointer.Context() for local variables, since we now plumb cgnodes throughout. Nice. - Refactor all load-form (load, receive, lookup) and store-form (Store, send, MapUpdate) constraints to use genLoad and genStore. - Log counts of constraints by type. - valNodes split into localval and globalval maps; localval is purged after each function. - analogous maps localobj[v] and globalobj[v] hold sole label for pts(v), if singleton. - fnObj map subsumed by globalobj. - make{Function/Global/Constant} inlined into objectValue. Much cleaner. R=crawshaw CC=golang-dev https://golang.org/cl/13979043
2013-09-27 09:33:01 -06:00
tmap.Set(tDyn, append(prev, ptr{s.a, nil, v}))
}
return &tmap
}
func (x ptset) Intersects(y_ PointsToSet) bool {
y := y_.(ptset)
for l := range x.pts {
if _, ok := y.pts[l]; ok {
return true
}
}
return false
}
// ---- Pointer public interface
// ptr adapts a node to the Pointer interface.
type ptr struct {
go.tools/pointer: strength reduction during constraint generation. Motivation: simple constraints---copy and addr---are more amenable to pre-solver optimizations (forthcoming) than complex constraints: load, store, and all others. In code such as the following: t0 = new struct { x, y int } t1 = &t0.y t2 = *t1 there's no need for the full generality of a (complex) load constraint for t2=*t1 since t1 can only point to t0.y. All we need is a (simple) copy constraint t2 = (t0.y) where (t0.y) is the object node label for that field. For all "addressable" SSA instructions, we tabulate whether their points-to set is necessarily a singleton. For some (e.g. Alloc, MakeSlice, etc) this is always true by design. For others (e.g. FieldAddr) it depends on their operands. We exploit this information when generating constraints: all load-form and store-form constraints are reduced to copy constraints if the pointer's PTS is a singleton. Similarly all FieldAddr (y=&x.f) and IndexAddr (y=&x[0]) constraints are reduced to offset addition, for singleton operands. Here's the constraint mix when running on the oracle itself. The total number of constraints is unchanged but the fraction that are complex has gone down to 21% from 53%. before after --simple-- addr 20682 46949 copy 61454 91211 --complex-- offsetAddr 41621 15325 load 18769 12925 store 30758 6908 invoke 758 760 typeAssert 1688 1689 total 175832 175869 Also: - Add Pointer.Context() for local variables, since we now plumb cgnodes throughout. Nice. - Refactor all load-form (load, receive, lookup) and store-form (Store, send, MapUpdate) constraints to use genLoad and genStore. - Log counts of constraints by type. - valNodes split into localval and globalval maps; localval is purged after each function. - analogous maps localobj[v] and globalobj[v] hold sole label for pts(v), if singleton. - fnObj map subsumed by globalobj. - make{Function/Global/Constant} inlined into objectValue. Much cleaner. R=crawshaw CC=golang-dev https://golang.org/cl/13979043
2013-09-27 09:33:01 -06:00
a *analysis
cgn *cgnode
n nodeid // non-zero
}
func (p ptr) String() string {
return fmt.Sprintf("n%d", p.n)
}
go.tools/pointer: strength reduction during constraint generation. Motivation: simple constraints---copy and addr---are more amenable to pre-solver optimizations (forthcoming) than complex constraints: load, store, and all others. In code such as the following: t0 = new struct { x, y int } t1 = &t0.y t2 = *t1 there's no need for the full generality of a (complex) load constraint for t2=*t1 since t1 can only point to t0.y. All we need is a (simple) copy constraint t2 = (t0.y) where (t0.y) is the object node label for that field. For all "addressable" SSA instructions, we tabulate whether their points-to set is necessarily a singleton. For some (e.g. Alloc, MakeSlice, etc) this is always true by design. For others (e.g. FieldAddr) it depends on their operands. We exploit this information when generating constraints: all load-form and store-form constraints are reduced to copy constraints if the pointer's PTS is a singleton. Similarly all FieldAddr (y=&x.f) and IndexAddr (y=&x[0]) constraints are reduced to offset addition, for singleton operands. Here's the constraint mix when running on the oracle itself. The total number of constraints is unchanged but the fraction that are complex has gone down to 21% from 53%. before after --simple-- addr 20682 46949 copy 61454 91211 --complex-- offsetAddr 41621 15325 load 18769 12925 store 30758 6908 invoke 758 760 typeAssert 1688 1689 total 175832 175869 Also: - Add Pointer.Context() for local variables, since we now plumb cgnodes throughout. Nice. - Refactor all load-form (load, receive, lookup) and store-form (Store, send, MapUpdate) constraints to use genLoad and genStore. - Log counts of constraints by type. - valNodes split into localval and globalval maps; localval is purged after each function. - analogous maps localobj[v] and globalobj[v] hold sole label for pts(v), if singleton. - fnObj map subsumed by globalobj. - make{Function/Global/Constant} inlined into objectValue. Much cleaner. R=crawshaw CC=golang-dev https://golang.org/cl/13979043
2013-09-27 09:33:01 -06:00
func (p ptr) Context() call.GraphNode {
return p.cgn
}
func (p ptr) PointsTo() PointsToSet {
return ptset{p.a, p.a.nodes[p.n].pts}
}
func (p ptr) MayAlias(q Pointer) bool {
return p.PointsTo().Intersects(q.PointsTo())
}
go.tools/pointer: reflection, part 1: maps, and some core features. Core: reflect.TypeOf reflect.ValueOf reflect.Zero reflect.Value.Interface Maps: (reflect.Value).MapIndex (reflect.Value).MapKeys (reflect.Value).SetMapIndex (*reflect.rtype).Elem (*reflect.rtype).Key + tests: pointer/testdata/mapreflect.go. oracle/testdata/src/main/reflection.go. Interface objects (T, V...) have been renamed "tagged objects". Abstraction: we model reflect.Value similar to interface{}---as a pointer that points only to tagged objects---but a reflect.Value may also point to an "indirect tagged object", one in which the payload V is of type *T not T. These are required because reflect.Values can hold lvalues, e.g. when derived via Field() or Elem(), though we won't use them till we get to structs and pointers. Solving: each reflection intrinsic defines a new constraint and resolution rule. Because of the nature of reflection, generalizing across types, the resolution rules dynamically create additional complex constraints during solving, where previously only simple (copy) constraints were created. This requires some solver changes: The work done before the main solver loop (to attach new constraints to the graph) is now done before each iteration, in processNewConstraints. Its loop over constraints is broken into two passes: the first handles base (addr-of) constraints, the second handles simple and complex constraints. constraint.init() has been inlined. The only behaviour that varies across constraints is ptr() Sadly this will pessimize presolver optimisations, when we get there; such is the price of reflection. Objects: reflection intrinsics create objects (i.e. cause memory allocations) with no SSA operation. We will represent them as the cgnode of the instrinsic (e.g. reflect.New), so we extend Labels and node.data to represent objects as a product (not sum) of ssa.Value and cgnode and pull this out into its own type, struct object. This simplifies a number of invariants and saves space. The ntObject flag is now represented by obj!=nil; the other flags are moved into object. cgnodes are now always recorded in objects/Labels for which it is appropriate (all but those for globals, constants and the shared contours for functions). Also: - Prepopulate the flattenMemo cache to consider reflect.Value a fake pointer, not a struct. - Improve accessors and documentation on type Label. - @conctypes assertions renamed @types (since dyn. types needn't be concrete). - add oracle 'describe' test on an interface (missing, an oversight). R=crawshaw CC=golang-dev https://golang.org/cl/13418048
2013-09-16 07:49:10 -06:00
func (p ptr) DynamicTypes() *typemap.M {
return p.PointsTo().DynamicTypes()
}