1
0
mirror of https://github.com/golang/go synced 2024-10-01 09:38:36 -06:00
go/oracle/oracle.go

559 lines
17 KiB
Go
Raw Normal View History

// Copyright 2014 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Package oracle contains the implementation of the oracle tool whose
// command-line is provided by golang.org/x/tools/cmd/oracle.
//
// http://golang.org/s/oracle-design
// http://golang.org/s/oracle-user-manual
//
package oracle // import "golang.org/x/tools/oracle"
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
// This file defines oracle.Query, the entry point for the oracle tool.
// The actual executable is defined in cmd/oracle.
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// TODO(adonovan): new queries
// - show all statements that may update the selected lvalue
// (local, global, field, etc).
// - show all places where an object of type T is created
// (&T{}, var t T, new(T), new(struct{array [3]T}), etc.
// ORACLE CONTROL FLOW
//
// The Oracle is somewhat convoluted due to the need to support two
// very different use-cases, "one-shot" and "long running", and to do
// so quickly.
//
// The cmd/oracle tool issues "one-shot" queries via the exported
// Query function, which creates an Oracle to answer a single query.
// newOracle consults the 'needs' flags of the query mode and the
// package containing the query to avoid doing more work than it needs
// (loading, parsing, type checking, SSA construction).
//
// The Pythia tool (github.com/fzipp/pythia) is an example of a "long
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// running" tool. It calls New() and then loops, calling
// ParseQueryPos and (*Oracle).Query to handle each incoming HTTP
// query. Since New cannot see which queries will follow, it must
// load, parse, type-check and SSA-build the entire transitive closure
// of the analysis scope, retaining full debug information and all
// typed ASTs.
//
// TODO(adonovan): experiment with inverting the control flow by
// making each mode consist of two functions: a "one-shot setup"
// function and the existing "impl" function. The one-shot setup
// function would do all of the work of Query and newOracle,
// specialized to each mode, calling library utilities for the common
// things. This would give it more control over "scope reduction".
// Long running tools would not call the one-shot setup function but
// would have their own setup function equivalent to the existing
// 'needsAll' flow path.
import (
"fmt"
"go/ast"
"go/build"
"go/token"
"io"
"golang.org/x/tools/astutil"
"golang.org/x/tools/go/loader"
"golang.org/x/tools/go/pointer"
"golang.org/x/tools/go/ssa"
"golang.org/x/tools/go/types"
"golang.org/x/tools/oracle/serial"
)
// An Oracle holds the program state required for one or more queries.
type Oracle struct {
fset *token.FileSet // file set [all queries]
prog *ssa.Program // the SSA program [needSSA]
ptaConfig pointer.Config // pointer analysis configuration [needPTA]
typeInfo map[*types.Package]*loader.PackageInfo // type info for all ASTs in the program [needRetainTypeInfo]
}
// A set of bits indicating the analytical requirements of each mode.
//
// Typed ASTs for the whole program are always constructed
// transiently; they are retained only for the queried package unless
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// needRetainTypeInfo is set.
const (
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
needPos = 1 << iota // needs a position
needExactPos // needs an exact AST selection; implies needPos
needRetainTypeInfo // needs to retain type info for all ASTs in the program
needSSA // needs ssa.Packages for whole program
needSSADebug // needs debug info for ssa.Packages
needPTA = needSSA // needs pointer analysis
needAll = -1 // needs everything (e.g. a sequence of queries)
)
type modeInfo struct {
name string
needs int
impl func(*Oracle, *QueryPos) (queryResult, error)
}
var modes = []*modeInfo{
// Pointer analyses, whole program:
{"callees", needPTA | needExactPos, callees},
{"callers", needPTA | needPos, callers},
{"callgraph", needPTA, doCallgraph},
{"callstack", needPTA | needPos, callstack},
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
{"peers", needPTA | needSSADebug | needPos, peers},
{"pointsto", needPTA | needSSADebug | needExactPos, pointsto},
{"whicherrs", needPTA | needSSADebug | needExactPos, whicherrs},
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// Type-based, modular analyses:
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
{"definition", needPos, definition},
{"describe", needExactPos, describe},
{"freevars", needPos, freevars},
// Type-based, whole-program analyses:
{"implements", needRetainTypeInfo | needPos, implements},
{"referrers", needRetainTypeInfo | needPos, referrers},
}
func findMode(mode string) *modeInfo {
for _, m := range modes {
if m.name == mode {
return m
}
}
return nil
}
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
type printfFunc func(pos interface{}, format string, args ...interface{})
// queryResult is the interface of each query-specific result type.
type queryResult interface {
toSerial(res *serial.Result, fset *token.FileSet)
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
display(printf printfFunc)
}
// A QueryPos represents the position provided as input to a query:
// a textual extent in the program's source code, the AST node it
// corresponds to, and the package to which it belongs.
// Instances are created by ParseQueryPos.
//
type QueryPos struct {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
fset *token.FileSet
start, end token.Pos // source extent of query
path []ast.Node // AST path from query node to root of ast.File
exact bool // 2nd result of PathEnclosingInterval
info *loader.PackageInfo // type info for the queried package (nil for fastQueryPos)
}
// TypeString prints type T relative to the query position.
func (qpos *QueryPos) TypeString(T types.Type) string {
return types.TypeString(qpos.info.Pkg, T)
}
// ObjectString prints object obj relative to the query position.
func (qpos *QueryPos) ObjectString(obj types.Object) string {
return types.ObjectString(qpos.info.Pkg, obj)
}
// SelectionString prints selection sel relative to the query position.
func (qpos *QueryPos) SelectionString(sel *types.Selection) string {
return types.SelectionString(qpos.info.Pkg, sel)
}
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
// A Result encapsulates the result of an oracle.Query.
type Result struct {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
fset *token.FileSet
q queryResult // the query-specific result
mode string // query mode
warnings []pointer.Warning // pointer analysis warnings (TODO(adonovan): fix: never populated!)
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
}
// Serial returns an instance of serial.Result, which implements the
// {xml,json}.Marshaler interfaces so that query results can be
// serialized as JSON or XML.
//
func (res *Result) Serial() *serial.Result {
resj := &serial.Result{Mode: res.mode}
res.q.toSerial(resj, res.fset)
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
for _, w := range res.warnings {
resj.Warnings = append(resj.Warnings, serial.PTAWarning{
Pos: res.fset.Position(w.Pos).String(),
Message: w.Message,
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
})
}
return resj
}
// Query runs a single oracle query.
//
// args specify the main package in (*loader.Config).FromArgs syntax.
// mode is the query mode ("callers", etc).
// ptalog is the (optional) pointer-analysis log file.
// buildContext is the go/build configuration for locating packages.
// reflection determines whether to model reflection soundly (currently slow).
//
// Clients that intend to perform multiple queries against the same
// analysis scope should use this pattern instead:
//
// conf := loader.Config{Build: buildContext, SourceImports: true}
// ... populate config, e.g. conf.FromArgs(args) ...
// iprog, err := conf.Load()
// if err != nil { ... }
// o, err := oracle.New(iprog, nil, false)
// if err != nil { ... }
// for ... {
// qpos, err := oracle.ParseQueryPos(imp, pos, needExact)
// if err != nil { ... }
//
// res, err := o.Query(mode, qpos)
// if err != nil { ... }
//
// // use res
// }
//
// TODO(adonovan): the ideal 'needsExact' parameter for ParseQueryPos
// depends on the query mode; how should we expose this?
//
func Query(args []string, mode, pos string, ptalog io.Writer, buildContext *build.Context, reflection bool) (*Result, error) {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if mode == "what" {
// Bypass package loading, type checking, SSA construction.
return what(pos, buildContext)
}
minfo := findMode(mode)
if minfo == nil {
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
return nil, fmt.Errorf("invalid mode type: %q", mode)
}
conf := loader.Config{Build: buildContext, SourceImports: true}
// Determine initial packages.
args, err := conf.FromArgs(args, true)
if err != nil {
return nil, err
}
if len(args) > 0 {
return nil, fmt.Errorf("surplus arguments: %q", args)
}
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// For queries needing only a single typed package,
// reduce the analysis scope to that package.
if minfo.needs&(needSSA|needRetainTypeInfo) == 0 {
reduceScope(pos, &conf)
}
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// TODO(adonovan): report type errors to the user via Serial
// types, not stderr?
// conf.TypeChecker.Error = func(err error) {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// E := err.(types.Error)
// fmt.Fprintf(os.Stderr, "%s: %s\n", E.Fset.Position(E.Pos), E.Msg)
// }
// Load/parse/type-check the program.
iprog, err := conf.Load()
if err != nil {
return nil, err
}
o, err := newOracle(iprog, ptalog, minfo.needs, reflection)
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if err != nil {
return nil, err
}
qpos, err := ParseQueryPos(iprog, pos, minfo.needs&needExactPos != 0)
if err != nil && minfo.needs&(needPos|needExactPos) != 0 {
return nil, err
}
// SSA is built and we have the QueryPos.
// Release the other ASTs and type info to the GC.
iprog = nil
return o.query(minfo, qpos)
}
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// reduceScope is called for one-shot queries that need only a single
// typed package. It attempts to guess the query package from pos and
// reduce the analysis scope (set of loaded packages) to just that one
// plus (the exported parts of) its dependencies. It leaves its
// arguments unchanged on failure.
//
// TODO(adonovan): this is a real mess... but it's fast.
//
func reduceScope(pos string, conf *loader.Config) {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
fqpos, err := fastQueryPos(pos)
if err != nil {
return // bad query
}
// TODO(adonovan): fix: this gives the wrong results for files
// in non-importable packages such as tests and ad-hoc packages
// specified as a list of files (incl. the oracle's tests).
_, importPath, err := guessImportPath(fqpos.fset.File(fqpos.start).Name(), conf.Build)
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if err != nil {
return // can't find GOPATH dir
}
if importPath == "" {
return
}
// Check that it's possible to load the queried package.
// (e.g. oracle tests contain different 'package' decls in same dir.)
// Keep consistent with logic in loader/util.go!
cfg2 := *conf.Build
cfg2.CgoEnabled = false
bp, err := cfg2.Import(importPath, "", 0)
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if err != nil {
return // no files for package
}
// Check that the queried file appears in the package:
// it might be a '// +build ignore' from an ad-hoc main
// package, e.g. $GOROOT/src/net/http/triv.go.
if !pkgContainsFile(bp, fqpos.fset.File(fqpos.start).Name()) {
return // not found
}
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
conf.TypeCheckFuncBodies = func(p string) bool { return p == importPath }
// Ignore packages specified on command line.
conf.CreatePkgs = nil
conf.ImportPkgs = nil
// Instead load just the one containing the query position
// (and possibly its corresponding tests/production code).
// TODO(adonovan): set 'augment' based on which file list
// contains
_ = conf.ImportWithTests(importPath) // ignore error
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
}
func pkgContainsFile(bp *build.Package, filename string) bool {
for _, files := range [][]string{bp.GoFiles, bp.TestGoFiles, bp.XTestGoFiles} {
for _, file := range files {
if sameFile(file, filename) {
return true
}
}
}
return false
}
// New constructs a new Oracle that can be used for a sequence of queries.
//
// iprog specifies the program to analyze.
// ptalog is the (optional) pointer-analysis log file.
// reflection determines whether to model reflection soundly (currently slow).
//
func New(iprog *loader.Program, ptalog io.Writer, reflection bool) (*Oracle, error) {
return newOracle(iprog, ptalog, needAll, reflection)
}
func newOracle(iprog *loader.Program, ptalog io.Writer, needs int, reflection bool) (*Oracle, error) {
o := &Oracle{fset: iprog.Fset}
// Retain type info for all ASTs in the program.
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if needs&needRetainTypeInfo != 0 {
o.typeInfo = iprog.AllPackages
}
// Create SSA package for the initial packages and their dependencies.
if needs&needSSA != 0 {
var mode ssa.BuilderMode
if needs&needSSADebug != 0 {
mode |= ssa.GlobalDebug
}
prog := ssa.Create(iprog, mode)
// For each initial package (specified on the command line),
// if it has a main function, analyze that,
// otherwise analyze its tests, if any.
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
var testPkgs, mains []*ssa.Package
for _, info := range iprog.InitialPackages() {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
initialPkg := prog.Package(info.Pkg)
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
// Add package to the pointer analysis scope.
if initialPkg.Func("main") != nil {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
mains = append(mains, initialPkg)
} else {
testPkgs = append(testPkgs, initialPkg)
}
}
if testPkgs != nil {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if p := prog.CreateTestMainPackage(testPkgs...); p != nil {
mains = append(mains, p)
}
}
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if mains == nil {
return nil, fmt.Errorf("analysis scope has no main and no tests")
}
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
o.ptaConfig.Log = ptalog
o.ptaConfig.Reflection = reflection
o.ptaConfig.Mains = mains
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
o.prog = prog
}
return o, nil
}
// Query runs the query of the specified mode and selection.
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
//
// TODO(adonovan): fix: this function does not currently support the
// "what" query, which needs to access the go/build.Context.
//
func (o *Oracle) Query(mode string, qpos *QueryPos) (*Result, error) {
minfo := findMode(mode)
if minfo == nil {
return nil, fmt.Errorf("invalid mode type: %q", mode)
}
return o.query(minfo, qpos)
}
func (o *Oracle) query(minfo *modeInfo, qpos *QueryPos) (*Result, error) {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// Clear out residue of previous query (for long-running clients).
o.ptaConfig.Queries = nil
o.ptaConfig.IndirectQueries = nil
res := &Result{
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
mode: minfo.name,
fset: o.fset,
}
var err error
res.q, err = minfo.impl(o, qpos)
if err != nil {
return nil, err
}
return res, nil
}
// ParseQueryPos parses the source query position pos.
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
// If needExact, it must identify a single AST subtree;
// this is appropriate for queries that allow fairly arbitrary syntax,
// e.g. "describe".
//
func ParseQueryPos(iprog *loader.Program, posFlag string, needExact bool) (*QueryPos, error) {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
filename, startOffset, endOffset, err := parsePosFlag(posFlag)
if err != nil {
return nil, err
}
start, end, err := findQueryPos(iprog.Fset, filename, startOffset, endOffset)
if err != nil {
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
return nil, err
}
info, path, exact := iprog.PathEnclosingInterval(start, end)
if path == nil {
return nil, fmt.Errorf("no syntax here")
}
if needExact && !exact {
return nil, fmt.Errorf("ambiguous selection within %s", astutil.NodeDescription(path[0]))
}
return &QueryPos{iprog.Fset, start, end, path, exact, info}, nil
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
}
// WriteTo writes the oracle query result res to out in a compiler diagnostic format.
func (res *Result) WriteTo(out io.Writer) {
printf := func(pos interface{}, format string, args ...interface{}) {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
fprintf(out, res.fset, pos, format, args...)
}
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
res.q.display(printf)
// Print warnings after the main output.
go.tools/oracle: add option to output results in JSON syntax. See json.go for interface specification. Example usage: % oracle -format=json -mode=callgraph code.google.com/p/go.tools/cmd/oracle + Tests, based on (small) golden files. Overview: Each <query>Result structure has been "lowered" so that all but the most trivial logic in each display() function has been moved to the main query. Each one now has a toJSON method that populates a json.Result struct. Though the <query>Result structs are similar to the correponding JSON protocol, they're not close enough to be used directly; for example, the former contain richer semantic entities (token.Pos, ast.Expr, ssa.Value, pointer.Pointer, etc) whereas JSON contains only their printed forms using Go basic types. The choices of what levels of abstractions the two sets of structs should have is somewhat arbitrary. We may want richer information in the JSON output in future. Details: - oracle.Main has been split into oracle.Query() and the printing of the oracle.Result. - the display() method no longer needs an *oracle param, only a print function. - callees: sort the result for determinism. - callees: compute the union across all contexts. - callers: sort the results for determinism. - describe(package): fixed a bug in the predicate for method accessibility: an unexported method defined in pkg A may belong to a type defined in package B (via embedding/promotion) and may thus be accessible to A. New accessibleMethods() utility fixes this. - describe(type): filter methods by accessibility. - added tests of 'callgraph'. - pointer: eliminated the 'caller CallGraphNode' parameter from pointer.Context.Call callback since it was redundant w.r.t site.Caller(). - added warning if CGO_ENABLED is unset. R=crawshaw CC=golang-dev https://golang.org/cl/13270045
2013-09-03 13:29:02 -06:00
if res.warnings != nil {
fmt.Fprintln(out, "\nPointer analysis warnings:")
for _, w := range res.warnings {
printf(w.Pos, "warning: "+w.Message)
}
}
}
// ---------- Utilities ----------
// buildSSA constructs the SSA representation of Go-source function bodies.
// Not needed in simpler modes, e.g. freevars.
//
func buildSSA(o *Oracle) {
o.prog.BuildAll()
}
// ptrAnalysis runs the pointer analysis and returns its result.
func ptrAnalysis(o *Oracle) *pointer.Result {
result, err := pointer.Analyze(&o.ptaConfig)
if err != nil {
panic(err) // pointer analysis internal error
}
return result
}
// unparen returns e with any enclosing parentheses stripped.
func unparen(e ast.Expr) ast.Expr {
for {
p, ok := e.(*ast.ParenExpr)
if !ok {
break
}
e = p.X
}
return e
}
// deref returns a pointer's element type; otherwise it returns typ.
func deref(typ types.Type) types.Type {
if p, ok := typ.Underlying().(*types.Pointer); ok {
return p.Elem()
}
return typ
}
// fprintf prints to w a message of the form "location: message\n"
// where location is derived from pos.
//
// pos must be one of:
// - a token.Pos, denoting a position
// - an ast.Node, denoting an interval
// - anything with a Pos() method:
// ssa.Member, ssa.Value, ssa.Instruction, types.Object, pointer.Label, etc.
// - a QueryPos, denoting the extent of the user's query.
// - nil, meaning no position at all.
//
// The output format is is compatible with the 'gnu'
// compilation-error-regexp in Emacs' compilation mode.
// TODO(adonovan): support other editors.
//
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
func fprintf(w io.Writer, fset *token.FileSet, pos interface{}, format string, args ...interface{}) {
var start, end token.Pos
switch pos := pos.(type) {
case ast.Node:
start = pos.Pos()
end = pos.End()
case token.Pos:
start = pos
end = start
case interface {
Pos() token.Pos
}:
start = pos.Pos()
end = start
case *QueryPos:
start = pos.start
end = pos.end
case nil:
// no-op
default:
panic(fmt.Sprintf("invalid pos: %T", pos))
}
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
if sp := fset.Position(start); start == end {
// (prints "-: " for token.NoPos)
fmt.Fprintf(w, "%s: ", sp)
} else {
go.tools/oracle: improvements to command set and performance. Command set: - what: an extremely fast query that parses a single file and returns the AST stack, package name and the set of query modes that apply to the current selection. Intended for GUI tools that need to grey out UI elements. - definition: shows the definition of an identifier. - pointsto: the PTA features of 'describe' have been split out into their own command. - describe: with PTA stripped out, the cost is now bounded by type checking. Performance: - The importer.Config.TypeCheckFuncBodies predicate supports setting the 'IgnoreFuncBodies' typechecker flag on a per-package basis. This means we can load dependencies from source more quickly if we only need exported types. (We avoid gcimport data because it may be absent or stale.) This also means we can run type-based queries on packages that aren't part of the pointer analysis scope. (Yay.) - Modes that require only type analysis of the query package run a "what" query first, and restrict their analysis scope to just that package and its dependencies (sans func bodies), making them much faster. - We call newOracle not oracle.New in Query, so that the 'needs' bitset isn't ignored (oops!). This makes the non-PTA queries faster. Also: - removed vestigial timers junk. - pos.go: existing position utilties split out into own file. Added parsePosFlag utility. - numerous cosmetic tweaks. + very basic tests. To do in follow-ups: - sophisticated editor integration of "what". - better tests. - refactoring of control flow as described in comment. - changes to "implements", "describe" commands. - update design doc + user manual. R=crawshaw, dominik.honnef CC=golang-dev, gri https://golang.org/cl/40630043
2013-12-13 08:04:55 -07:00
ep := fset.Position(end)
// The -1 below is a concession to Emacs's broken use of
// inclusive (not half-open) intervals.
// Other editors may not want it.
// TODO(adonovan): add an -editor=vim|emacs|acme|auto
// flag; auto uses EMACS=t / VIM=... / etc env vars.
fmt.Fprintf(w, "%s:%d.%d-%d.%d: ",
sp.Filename, sp.Line, sp.Column, ep.Line, ep.Column-1)
}
fmt.Fprintf(w, format, args...)
io.WriteString(w, "\n")
}