1
0
mirror of https://github.com/golang/go synced 2024-10-01 12:28:37 -06:00
go/oracle/what.go

211 lines
5.1 KiB
Go
Raw Normal View History

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
// Copyright 2013 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package oracle
import (
"fmt"
"go/ast"
"go/build"
"go/token"
"os"
"path/filepath"
"sort"
"strings"
"golang.org/x/tools/go/ast/astutil"
"golang.org/x/tools/oracle/serial"
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
)
// what reports all the information about the query selection that can be
// obtained from parsing only its containing source file.
// It is intended to be a very low-latency query callable from GUI
// tools, e.g. to populate a menu of options of slower queries about
// the selected location.
//
oracle: several major improvements Features: More robust: silently ignore type errors in modes that don't need SSA form: describe, referrers, implements, freevars, description. This makes the tool much more robust for everyday queries. Less configuration: don't require a scope argument for all queries. Only queries that do pointer analysis need it. For the rest, the initial position is enough for importQueryPackage to deduce the scope. It now works for queries in GoFiles, TestGoFiles, or XTestGoFiles. (It no longer works for ad-hoc main packages like $GOROOT/src/net/http/triv.go) More complete: "referrers" computes the scope automatically by scanning the import graph of the entire workspace, using gorename's refactor/importgraph package. This requires two passes at loading. Faster: simplified start-up logic avoids unnecessary package loading and SSA construction (a consequence of bad abstraction) in many cases. "callgraph": remove it. Unlike all the other commands it isn't related to the current selection, and we have golang.org/x/tools/cmdcallgraph now. Internals: Drop support for long-running clients (i.e., Pythia), since godoc -analysis supports all the same features except "pointsto", and precomputes all the results so latency is much lower. Get rid of various unhelpful abstractions introduced to support long-running clients. Expand out the set-up logic for each subcommand. This is simpler, easier to read, and gives us more control, at a small cost in duplication---the familiar story of abstractions. Discard PTA warnings. We weren't showing them (nor should we). Split tests into separate directories (so that importgraph works). Change-Id: I55d46b3ab33cdf7ac22436fcc2148fe04c901237 Reviewed-on: https://go-review.googlesource.com/8243 Reviewed-by: David Crawshaw <crawshaw@golang.org>
2015-03-30 09:21:48 -06:00
func what(q *Query) error {
qpos, err := fastQueryPos(q.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 err != nil {
oracle: several major improvements Features: More robust: silently ignore type errors in modes that don't need SSA form: describe, referrers, implements, freevars, description. This makes the tool much more robust for everyday queries. Less configuration: don't require a scope argument for all queries. Only queries that do pointer analysis need it. For the rest, the initial position is enough for importQueryPackage to deduce the scope. It now works for queries in GoFiles, TestGoFiles, or XTestGoFiles. (It no longer works for ad-hoc main packages like $GOROOT/src/net/http/triv.go) More complete: "referrers" computes the scope automatically by scanning the import graph of the entire workspace, using gorename's refactor/importgraph package. This requires two passes at loading. Faster: simplified start-up logic avoids unnecessary package loading and SSA construction (a consequence of bad abstraction) in many cases. "callgraph": remove it. Unlike all the other commands it isn't related to the current selection, and we have golang.org/x/tools/cmdcallgraph now. Internals: Drop support for long-running clients (i.e., Pythia), since godoc -analysis supports all the same features except "pointsto", and precomputes all the results so latency is much lower. Get rid of various unhelpful abstractions introduced to support long-running clients. Expand out the set-up logic for each subcommand. This is simpler, easier to read, and gives us more control, at a small cost in duplication---the familiar story of abstractions. Discard PTA warnings. We weren't showing them (nor should we). Split tests into separate directories (so that importgraph works). Change-Id: I55d46b3ab33cdf7ac22436fcc2148fe04c901237 Reviewed-on: https://go-review.googlesource.com/8243 Reviewed-by: David Crawshaw <crawshaw@golang.org>
2015-03-30 09:21:48 -06:00
return err
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
}
oracle: several major improvements Features: More robust: silently ignore type errors in modes that don't need SSA form: describe, referrers, implements, freevars, description. This makes the tool much more robust for everyday queries. Less configuration: don't require a scope argument for all queries. Only queries that do pointer analysis need it. For the rest, the initial position is enough for importQueryPackage to deduce the scope. It now works for queries in GoFiles, TestGoFiles, or XTestGoFiles. (It no longer works for ad-hoc main packages like $GOROOT/src/net/http/triv.go) More complete: "referrers" computes the scope automatically by scanning the import graph of the entire workspace, using gorename's refactor/importgraph package. This requires two passes at loading. Faster: simplified start-up logic avoids unnecessary package loading and SSA construction (a consequence of bad abstraction) in many cases. "callgraph": remove it. Unlike all the other commands it isn't related to the current selection, and we have golang.org/x/tools/cmdcallgraph now. Internals: Drop support for long-running clients (i.e., Pythia), since godoc -analysis supports all the same features except "pointsto", and precomputes all the results so latency is much lower. Get rid of various unhelpful abstractions introduced to support long-running clients. Expand out the set-up logic for each subcommand. This is simpler, easier to read, and gives us more control, at a small cost in duplication---the familiar story of abstractions. Discard PTA warnings. We weren't showing them (nor should we). Split tests into separate directories (so that importgraph works). Change-Id: I55d46b3ab33cdf7ac22436fcc2148fe04c901237 Reviewed-on: https://go-review.googlesource.com/8243 Reviewed-by: David Crawshaw <crawshaw@golang.org>
2015-03-30 09:21:48 -06:00
q.Fset = qpos.fset
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
// (ignore errors)
oracle: several major improvements Features: More robust: silently ignore type errors in modes that don't need SSA form: describe, referrers, implements, freevars, description. This makes the tool much more robust for everyday queries. Less configuration: don't require a scope argument for all queries. Only queries that do pointer analysis need it. For the rest, the initial position is enough for importQueryPackage to deduce the scope. It now works for queries in GoFiles, TestGoFiles, or XTestGoFiles. (It no longer works for ad-hoc main packages like $GOROOT/src/net/http/triv.go) More complete: "referrers" computes the scope automatically by scanning the import graph of the entire workspace, using gorename's refactor/importgraph package. This requires two passes at loading. Faster: simplified start-up logic avoids unnecessary package loading and SSA construction (a consequence of bad abstraction) in many cases. "callgraph": remove it. Unlike all the other commands it isn't related to the current selection, and we have golang.org/x/tools/cmdcallgraph now. Internals: Drop support for long-running clients (i.e., Pythia), since godoc -analysis supports all the same features except "pointsto", and precomputes all the results so latency is much lower. Get rid of various unhelpful abstractions introduced to support long-running clients. Expand out the set-up logic for each subcommand. This is simpler, easier to read, and gives us more control, at a small cost in duplication---the familiar story of abstractions. Discard PTA warnings. We weren't showing them (nor should we). Split tests into separate directories (so that importgraph works). Change-Id: I55d46b3ab33cdf7ac22436fcc2148fe04c901237 Reviewed-on: https://go-review.googlesource.com/8243 Reviewed-by: David Crawshaw <crawshaw@golang.org>
2015-03-30 09:21:48 -06:00
srcdir, importPath, _ := guessImportPath(q.Fset.File(qpos.start).Name(), q.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
// Determine which query modes are applicable to the selection.
enable := map[string]bool{
oracle: several major improvements Features: More robust: silently ignore type errors in modes that don't need SSA form: describe, referrers, implements, freevars, description. This makes the tool much more robust for everyday queries. Less configuration: don't require a scope argument for all queries. Only queries that do pointer analysis need it. For the rest, the initial position is enough for importQueryPackage to deduce the scope. It now works for queries in GoFiles, TestGoFiles, or XTestGoFiles. (It no longer works for ad-hoc main packages like $GOROOT/src/net/http/triv.go) More complete: "referrers" computes the scope automatically by scanning the import graph of the entire workspace, using gorename's refactor/importgraph package. This requires two passes at loading. Faster: simplified start-up logic avoids unnecessary package loading and SSA construction (a consequence of bad abstraction) in many cases. "callgraph": remove it. Unlike all the other commands it isn't related to the current selection, and we have golang.org/x/tools/cmdcallgraph now. Internals: Drop support for long-running clients (i.e., Pythia), since godoc -analysis supports all the same features except "pointsto", and precomputes all the results so latency is much lower. Get rid of various unhelpful abstractions introduced to support long-running clients. Expand out the set-up logic for each subcommand. This is simpler, easier to read, and gives us more control, at a small cost in duplication---the familiar story of abstractions. Discard PTA warnings. We weren't showing them (nor should we). Split tests into separate directories (so that importgraph works). Change-Id: I55d46b3ab33cdf7ac22436fcc2148fe04c901237 Reviewed-on: https://go-review.googlesource.com/8243 Reviewed-by: David Crawshaw <crawshaw@golang.org>
2015-03-30 09:21:48 -06:00
"describe": true, // any syntax; always enabled
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 qpos.end > qpos.start {
enable["freevars"] = true // nonempty 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
for _, n := range qpos.path {
switch n := n.(type) {
case *ast.Ident:
enable["definition"] = true
enable["referrers"] = true
enable["implements"] = true
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
case *ast.CallExpr:
enable["callees"] = true
case *ast.FuncDecl:
enable["callers"] = true
enable["callstack"] = true
case *ast.SendStmt:
enable["peers"] = true
case *ast.UnaryExpr:
if n.Op == token.ARROW {
enable["peers"] = true
}
}
// For implements, we approximate findInterestingNode.
if _, ok := enable["implements"]; !ok {
switch n.(type) {
case *ast.ArrayType,
*ast.StructType,
*ast.FuncType,
*ast.InterfaceType,
*ast.MapType,
*ast.ChanType:
enable["implements"] = true
}
}
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 pointsto, we approximate findInterestingNode.
if _, ok := enable["pointsto"]; !ok {
switch n.(type) {
case ast.Stmt,
*ast.ArrayType,
*ast.StructType,
*ast.FuncType,
*ast.InterfaceType,
*ast.MapType,
*ast.ChanType:
enable["pointsto"] = false // not an expr
case ast.Expr, ast.Decl, *ast.ValueSpec:
enable["pointsto"] = true // an expr, maybe
default:
// Comment, Field, KeyValueExpr, etc: ascend.
}
}
}
// If we don't have an exact selection, disable modes that need one.
if !qpos.exact {
oracle: several major improvements Features: More robust: silently ignore type errors in modes that don't need SSA form: describe, referrers, implements, freevars, description. This makes the tool much more robust for everyday queries. Less configuration: don't require a scope argument for all queries. Only queries that do pointer analysis need it. For the rest, the initial position is enough for importQueryPackage to deduce the scope. It now works for queries in GoFiles, TestGoFiles, or XTestGoFiles. (It no longer works for ad-hoc main packages like $GOROOT/src/net/http/triv.go) More complete: "referrers" computes the scope automatically by scanning the import graph of the entire workspace, using gorename's refactor/importgraph package. This requires two passes at loading. Faster: simplified start-up logic avoids unnecessary package loading and SSA construction (a consequence of bad abstraction) in many cases. "callgraph": remove it. Unlike all the other commands it isn't related to the current selection, and we have golang.org/x/tools/cmdcallgraph now. Internals: Drop support for long-running clients (i.e., Pythia), since godoc -analysis supports all the same features except "pointsto", and precomputes all the results so latency is much lower. Get rid of various unhelpful abstractions introduced to support long-running clients. Expand out the set-up logic for each subcommand. This is simpler, easier to read, and gives us more control, at a small cost in duplication---the familiar story of abstractions. Discard PTA warnings. We weren't showing them (nor should we). Split tests into separate directories (so that importgraph works). Change-Id: I55d46b3ab33cdf7ac22436fcc2148fe04c901237 Reviewed-on: https://go-review.googlesource.com/8243 Reviewed-by: David Crawshaw <crawshaw@golang.org>
2015-03-30 09:21:48 -06:00
enable["callees"] = false
enable["pointsto"] = false
enable["whicherrs"] = false
enable["describe"] = false
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 modes []string
for mode := range enable {
modes = append(modes, mode)
}
sort.Strings(modes)
oracle: several major improvements Features: More robust: silently ignore type errors in modes that don't need SSA form: describe, referrers, implements, freevars, description. This makes the tool much more robust for everyday queries. Less configuration: don't require a scope argument for all queries. Only queries that do pointer analysis need it. For the rest, the initial position is enough for importQueryPackage to deduce the scope. It now works for queries in GoFiles, TestGoFiles, or XTestGoFiles. (It no longer works for ad-hoc main packages like $GOROOT/src/net/http/triv.go) More complete: "referrers" computes the scope automatically by scanning the import graph of the entire workspace, using gorename's refactor/importgraph package. This requires two passes at loading. Faster: simplified start-up logic avoids unnecessary package loading and SSA construction (a consequence of bad abstraction) in many cases. "callgraph": remove it. Unlike all the other commands it isn't related to the current selection, and we have golang.org/x/tools/cmdcallgraph now. Internals: Drop support for long-running clients (i.e., Pythia), since godoc -analysis supports all the same features except "pointsto", and precomputes all the results so latency is much lower. Get rid of various unhelpful abstractions introduced to support long-running clients. Expand out the set-up logic for each subcommand. This is simpler, easier to read, and gives us more control, at a small cost in duplication---the familiar story of abstractions. Discard PTA warnings. We weren't showing them (nor should we). Split tests into separate directories (so that importgraph works). Change-Id: I55d46b3ab33cdf7ac22436fcc2148fe04c901237 Reviewed-on: https://go-review.googlesource.com/8243 Reviewed-by: David Crawshaw <crawshaw@golang.org>
2015-03-30 09:21:48 -06:00
q.result = &whatResult{
path: qpos.path,
srcdir: srcdir,
importPath: importPath,
modes: modes,
}
return 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
}
// guessImportPath finds the package containing filename, and returns
// its source directory (an element of $GOPATH) and its import path
// relative to it.
//
// TODO(adonovan): what about _test.go files that are not part of the
// package?
//
func guessImportPath(filename string, buildContext *build.Context) (srcdir, importPath string, err error) {
absFile, err := filepath.Abs(filename)
if err != nil {
err = fmt.Errorf("can't form absolute path of %s", filename)
return
}
absFileDir := segments(filepath.Dir(absFile))
// Find the innermost directory in $GOPATH that encloses filename.
minD := 1024
for _, gopathDir := range buildContext.SrcDirs() {
absDir, err := filepath.Abs(gopathDir)
if err != nil {
continue // e.g. non-existent dir on $GOPATH
}
d := prefixLen(segments(absDir), absFileDir)
// If there are multiple matches,
// prefer the innermost enclosing directory
// (smallest d).
if d >= 0 && d < minD {
minD = d
srcdir = gopathDir
importPath = strings.Join(absFileDir[len(absFileDir)-minD:], string(os.PathSeparator))
}
}
if srcdir == "" {
err = fmt.Errorf("directory %s is not beneath GOROOT or GOPATH: %s",
filepath.Dir(absFile), strings.Join(buildContext.SrcDirs(), ", "))
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
}
return
}
func segments(path string) []string {
return strings.Split(path, string(os.PathSeparator))
}
// prefixLen returns the length of the remainder of y if x is a prefix
// of y, a negative number otherwise.
func prefixLen(x, y []string) int {
d := len(y) - len(x)
if d >= 0 {
for i := range x {
if y[i] != x[i] {
return -1 // not a prefix
}
}
}
return d
}
type whatResult struct {
path []ast.Node
modes []string
srcdir string
importPath string
}
func (r *whatResult) display(printf printfFunc) {
for _, n := range r.path {
printf(n, "%s", astutil.NodeDescription(n))
}
printf(nil, "modes: %s", r.modes)
printf(nil, "srcdir: %s", r.srcdir)
printf(nil, "import path: %s", r.importPath)
}
func (r *whatResult) toSerial(res *serial.Result, fset *token.FileSet) {
var enclosing []serial.SyntaxNode
for _, n := range r.path {
enclosing = append(enclosing, serial.SyntaxNode{
Description: astutil.NodeDescription(n),
Start: fset.Position(n.Pos()).Offset,
End: fset.Position(n.End()).Offset,
})
}
res.What = &serial.What{
Modes: r.modes,
SrcDir: r.srcdir,
ImportPath: r.importPath,
Enclosing: enclosing,
}
}