1
0
mirror of https://github.com/golang/go synced 2024-11-19 06:14:39 -07:00
go/pointer/analysis.go

331 lines
10 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
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
// This file defines the main datatypes and Analyze function of the pointer analysis.
import (
"fmt"
"go/token"
"io"
"os"
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
"reflect"
"code.google.com/p/go.tools/go/types"
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
"code.google.com/p/go.tools/go/types/typemap"
"code.google.com/p/go.tools/ssa"
)
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
// object.flags bitmask values.
const (
otTagged = 1 << iota // type-tagged object
otIndirect // type-tagged object with indirect payload
otFunction // function object
)
// An object represents a contiguous block of memory to which some
// (generalized) pointer may point.
//
// (Note: most variables called 'obj' are not *objects but nodeids
// such that a.nodes[obj].obj != nil.)
//
type object struct {
// flags is a bitset of the node type (ot*) flags defined above.
flags uint32
// Number of following nodes belonging to the same "object"
// allocation. Zero for all other nodes.
size uint32
// The SSA operation that caused this object to be allocated.
// May be nil for (e.g.) intrinsic allocations.
val ssa.Value
// The call-graph node (=context) in which this object was allocated.
// May be nil for global objects: Global, Const, some Functions.
cgn *cgnode
// If this is an rtype instance object, or a *rtype-tagged
// object, this is its type.
rtype types.Type
}
// nodeid denotes a node.
// It is an index within analysis.nodes.
// We use small integers, not *node pointers, for many reasons:
// - they are smaller on 64-bit systems.
// - sets of them can be represented compactly in bitvectors or BDDs.
// - order matters; a field offset can be computed by simple addition.
type nodeid uint32
// A node is an equivalence class of memory locations.
// Nodes may be pointers, pointed-to locations, neither, or both.
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
//
// Nodes that are pointed-to locations ("labels") have an enclosing
// object (see analysis.enclosingObject).
//
type node struct {
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 non-nil, this node is the start of an object
// (addressable memory location).
// The following obj.size words implicitly belong to the object;
// they locate their object by scanning back.
obj *object
// The type of the field denoted by this node. Non-aggregate,
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
// unless this is an tagged.T node (i.e. the thing
// pointed to by an interface) in which case typ is that type.
typ types.Type
// subelement indicates which directly embedded subelement of
// an object of aggregate type (struct, tuple, array) this is.
subelement *fieldInfo // e.g. ".a.b[*].c"
// Points-to sets.
pts nodeset // points-to set of this node
prevPts nodeset // pts(n) in previous iteration (for difference propagation)
// Graph edges
copyTo nodeset // simple copy constraint edges
// Complex constraints attached to this node (x).
// - *loadConstraint y=*x
// - *offsetAddrConstraint y=&x.f or y=&x[0]
// - *storeConstraint *x=z
// - *typeAssertConstraint y=x.(T)
// - *invokeConstraint y=x.f(params...)
complex constraintset
}
type constraint interface {
String() string
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
// For a complex constraint, returns the nodeid of the pointer
// to which it is attached.
ptr() nodeid
// solve is called for complex constraints when the pts for
// the node to which they are attached has changed.
solve(a *analysis, n *node, delta nodeset)
}
// dst = &src
// pts(dst) ⊇ {src}
// A base constraint used to initialize the solver's pt sets
type addrConstraint struct {
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
dst nodeid // (ptr)
src nodeid
}
// dst = src
// A simple constraint represented directly as a copyTo graph edge.
type copyConstraint struct {
dst nodeid
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
src nodeid // (ptr)
}
// dst = src[offset]
// A complex constraint attached to src (the pointer)
type loadConstraint struct {
offset uint32
dst nodeid
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
src nodeid // (ptr)
}
// dst[offset] = src
// A complex constraint attached to dst (the pointer)
type storeConstraint struct {
offset uint32
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
dst nodeid // (ptr)
src nodeid
}
// dst = &src.f or dst = &src[0]
// A complex constraint attached to dst (the pointer)
type offsetAddrConstraint struct {
offset uint32
dst nodeid
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
src nodeid // (ptr)
}
// dst = src.(typ)
// A complex constraint attached to src (the interface).
type typeAssertConstraint struct {
typ types.Type
dst nodeid
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
src nodeid // (ptr)
}
// src.method(params...)
// A complex constraint attached to iface.
type invokeConstraint struct {
method *types.Func // the abstract method
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
iface nodeid // (ptr) the interface
params nodeid // the first parameter in the params/results block
}
// An analysis instance holds the state of a single pointer analysis problem.
type analysis struct {
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
config *Config // the client's control/observer interface
prog *ssa.Program // the program being analyzed
log io.Writer // log stream; nil to disable
panicNode nodeid // sink for panic, source for recover
nodes []*node // indexed by nodeid
flattenMemo map[types.Type][]*fieldInfo // memoization of flatten()
constraints []constraint // set of constraints
cgnodes []*cgnode // all cgnodes
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
genq []*cgnode // queue of functions to generate constraints for
intrinsics map[*ssa.Function]intrinsic // non-nil values are summaries for intrinsic fns
probes map[*ssa.CallCommon]nodeid // maps call to print() to argument variable
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
globalval map[ssa.Value]nodeid // node for each global ssa.Value
globalobj map[ssa.Value]nodeid // maps v to sole member of pts(v), if singleton
localval map[ssa.Value]nodeid // node for each local ssa.Value
localobj map[ssa.Value]nodeid // maps v to sole member of pts(v), if singleton
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
work worklist // solver's worklist
queries map[ssa.Value][]Pointer // same as Results.Queries
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
// Reflection:
hasher typemap.Hasher // cache of type hashes
reflectValueObj types.Object // type symbol for reflect.Value (if present)
reflectRtypeObj types.Object // *types.TypeName for reflect.rtype (if present)
reflectRtypePtr *types.Pointer // *reflect.rtype
reflectType *types.Named // reflect.Type
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
rtypes typemap.M // nodeid of canonical *rtype-tagged object for type T
reflectZeros typemap.M // nodeid of canonical T-tagged object for zero value
}
// enclosingObj returns the object (addressible memory object) that encloses node id.
// Panic ensues if that node does not belong to any object.
func (a *analysis) enclosingObj(id nodeid) *object {
// Find previous node with obj != nil.
for i := id; i >= 0; i-- {
n := a.nodes[i]
if obj := n.obj; obj != nil {
if i+nodeid(obj.size) <= id {
break // out of bounds
}
return obj
}
}
panic("node has no enclosing object")
}
// labelFor returns the Label for node id.
// Panic ensues if that node is not addressable.
func (a *analysis) labelFor(id nodeid) *Label {
return &Label{
obj: a.enclosingObj(id),
subelement: a.nodes[id].subelement,
}
}
func (a *analysis) warnf(pos token.Pos, format string, args ...interface{}) {
if Warn := a.config.Warn; Warn != nil {
Warn(pos, format, args...)
} else {
fmt.Fprintf(os.Stderr, "%s: warning: ", a.prog.Fset.Position(pos))
fmt.Fprintf(os.Stderr, format, args...)
fmt.Fprintln(os.Stderr)
}
}
// Analyze runs the pointer analysis with the scope and options
// specified by config, and returns the (synthetic) root of the callgraph.
//
func Analyze(config *Config) *Result {
a := &analysis{
config: config,
log: config.Log,
prog: config.prog(),
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
globalval: make(map[ssa.Value]nodeid),
globalobj: make(map[ssa.Value]nodeid),
flattenMemo: make(map[types.Type][]*fieldInfo),
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
hasher: typemap.MakeHasher(),
intrinsics: make(map[*ssa.Function]intrinsic),
probes: make(map[*ssa.CallCommon]nodeid),
work: makeMapWorklist(),
queries: make(map[ssa.Value][]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
if false {
a.log = os.Stderr // for debugging crashes; extremely verbose
}
if a.log != nil {
fmt.Fprintln(a.log, "======== NEW ANALYSIS ========")
}
go.tools/importer: generalize command-line syntax. Motivation: pointer analysis tools (like the oracle) want the user to specify a set of initial packages, like 'go test'. This change enables the user to specify a set of packages on the command line using importer.LoadInitialPackages(args). Each argument is interpreted as either: - a comma-separated list of *.go source files together comprising one non-importable ad-hoc package. e.g. "src/pkg/net/http/triv.go" gives us [main]. - an import path, denoting both the imported package and its non-importable external test package, if any. e.g. "fmt" gives us [fmt, fmt_test]. Current type-checker limitations mean that only the first import path may contribute tests: multiple packages augmented by *_test.go files could create import cycles, which 'go test' avoids by building a separate executable for each one. That approach is less attractive for static analysis. Details: (many files touched, but importer.go is the crux) importer: - PackageInfo.Importable boolean indicates whether package is importable. - un-expose Importer.Packages; expose AllPackages() instead. - CreatePackageFromArgs has become LoadInitialPackages. - imports() moved to util.go, renamed importsOf(). - InitialPackagesUsage usage message exported to clients. - the package name for ad-hoc packages now comes from the 'package' decl, not "main". ssa.Program: - added CreatePackages() method - PackagesByPath un-exposed, renamed 'imported'. - expose AllPackages and ImportedPackage accessors. oracle: - describe: explain and workaround a go/types bug. Misc: - Removed various unnecessary error.Error() calls in Printf args. R=crawshaw CC=golang-dev https://golang.org/cl/13579043
2013-09-06 16:13:57 -06:00
if reflect := a.prog.ImportedPackage("reflect"); reflect != nil {
a.reflectValueObj = reflect.Object.Scope().Lookup("Value")
a.reflectType = reflect.Object.Scope().Lookup("Type").Type().(*types.Named)
a.reflectRtypeObj = reflect.Object.Scope().Lookup("rtype")
a.reflectRtypePtr = types.NewPointer(a.reflectRtypeObj.Type())
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
// Override flattening of reflect.Value, treating it like a basic type.
tReflectValue := a.reflectValueObj.Type()
a.flattenMemo[tReflectValue] = []*fieldInfo{{typ: tReflectValue}}
a.rtypes.SetHasher(a.hasher)
a.reflectZeros.SetHasher(a.hasher)
}
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
root := a.generate()
if a.log != nil {
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
// Show size of constraint system.
counts := make(map[reflect.Type]int)
for _, c := range a.constraints {
counts[reflect.TypeOf(c)]++
}
fmt.Fprintf(a.log, "# constraints:\t%d\n", len(a.constraints))
for t, n := range counts {
fmt.Fprintf(a.log, "\t%s:\t%d\n", t, n)
}
fmt.Fprintf(a.log, "# nodes:\t%d\n", len(a.nodes))
}
//a.optimize()
a.solve()
if a.log != nil {
// Dump solution.
for i, n := range a.nodes {
if n.pts != nil {
fmt.Fprintf(a.log, "pts(n%d) = %s : %s\n", i, n.pts, n.typ)
}
}
}
// Visit discovered call graph.
for _, caller := range a.cgnodes {
for _, site := range caller.sites {
for nid := range a.nodes[site.targets].pts {
callee := a.nodes[nid].obj.cgn
if a.config.BuildCallGraph {
site.callees = append(site.callees, callee)
}
// TODO(adonovan): de-dup these messages.
// Warn about calls to non-intrinsic external functions.
if fn := callee.fn; fn.Blocks == nil && a.findIntrinsic(fn) == nil {
a.warnf(site.pos(), "unsound call to unknown intrinsic: %s", fn)
a.warnf(fn.Pos(), " (declared here)")
}
}
}
}
var callgraph *cgraph
if a.config.BuildCallGraph {
callgraph = &cgraph{root, a.cgnodes}
}
return &Result{
CallGraph: callgraph,
Queries: a.queries,
}
}