[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
// Copyright 2021 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 noder
|
|
|
|
|
|
|
|
|
|
import (
|
2021-01-19 13:54:33 -08:00
|
|
|
"fmt"
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
"os"
|
|
|
|
|
|
|
|
|
|
"cmd/compile/internal/base"
|
|
|
|
|
"cmd/compile/internal/dwarfgen"
|
|
|
|
|
"cmd/compile/internal/ir"
|
|
|
|
|
"cmd/compile/internal/syntax"
|
|
|
|
|
"cmd/compile/internal/typecheck"
|
|
|
|
|
"cmd/compile/internal/types"
|
|
|
|
|
"cmd/compile/internal/types2"
|
|
|
|
|
"cmd/internal/src"
|
|
|
|
|
)
|
|
|
|
|
|
2021-05-27 02:47:25 -07:00
|
|
|
// checkFiles configures and runs the types2 checker on the given
|
|
|
|
|
// parsed source files and then returns the result.
|
2021-06-11 01:45:24 -07:00
|
|
|
func checkFiles(noders []*noder) (posMap, *types2.Package, *types2.Info) {
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
if base.SyntaxErrors() != 0 {
|
|
|
|
|
base.ErrorExit()
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// setup and syntax error reporting
|
|
|
|
|
var m posMap
|
|
|
|
|
files := make([]*syntax.File, len(noders))
|
|
|
|
|
for i, p := range noders {
|
|
|
|
|
m.join(&p.posMap)
|
|
|
|
|
files[i] = p.file
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// typechecking
|
2021-09-29 13:05:37 -04:00
|
|
|
ctxt := types2.NewContext()
|
2021-06-11 01:45:24 -07:00
|
|
|
importer := gcimports{
|
2021-09-29 13:05:37 -04:00
|
|
|
ctxt: ctxt,
|
2021-09-02 10:42:43 +07:00
|
|
|
packages: map[string]*types2.Package{"unsafe": types2.Unsafe},
|
2021-06-11 01:45:24 -07:00
|
|
|
}
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
conf := types2.Config{
|
2021-09-29 13:05:37 -04:00
|
|
|
Context: ctxt,
|
2021-02-03 22:14:04 -08:00
|
|
|
GoVersion: base.Flag.Lang,
|
2021-01-21 20:20:22 -08:00
|
|
|
IgnoreLabels: true, // parser already checked via syntax.CheckBranches mode
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
CompilerErrorMessages: true, // use error strings matching existing compiler errors
|
|
|
|
|
Error: func(err error) {
|
|
|
|
|
terr := err.(types2.Error)
|
|
|
|
|
base.ErrorfAt(m.makeXPos(terr.Pos), "%s", terr.Msg)
|
|
|
|
|
},
|
2021-06-11 01:45:24 -07:00
|
|
|
Importer: &importer,
|
2021-05-27 02:47:25 -07:00
|
|
|
Sizes: &gcSizes{},
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
}
|
2021-05-27 02:47:25 -07:00
|
|
|
info := &types2.Info{
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
Types: make(map[syntax.Expr]types2.TypeAndValue),
|
|
|
|
|
Defs: make(map[*syntax.Name]types2.Object),
|
|
|
|
|
Uses: make(map[*syntax.Name]types2.Object),
|
|
|
|
|
Selections: make(map[*syntax.SelectorExpr]*types2.Selection),
|
|
|
|
|
Implicits: make(map[syntax.Node]types2.Object),
|
|
|
|
|
Scopes: make(map[syntax.Node]*types2.Scope),
|
2021-09-22 09:55:10 -07:00
|
|
|
Instances: make(map[*syntax.Name]types2.Instance),
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
// expand as needed
|
|
|
|
|
}
|
2021-05-27 02:47:25 -07:00
|
|
|
|
2021-09-08 16:03:57 -07:00
|
|
|
pkg, err := conf.Check(base.Ctxt.Pkgpath, files, info)
|
2021-06-11 01:45:24 -07:00
|
|
|
|
2021-02-21 22:27:19 +07:00
|
|
|
base.ExitIfErrors()
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
if err != nil {
|
|
|
|
|
base.FatalfAt(src.NoXPos, "conf.Check error: %v", err)
|
|
|
|
|
}
|
2021-05-27 02:47:25 -07:00
|
|
|
|
|
|
|
|
return m, pkg, info
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// check2 type checks a Go package using types2, and then generates IR
|
|
|
|
|
// using the results.
|
|
|
|
|
func check2(noders []*noder) {
|
2021-06-11 01:45:24 -07:00
|
|
|
m, pkg, info := checkFiles(noders)
|
2021-05-27 02:47:25 -07:00
|
|
|
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
if base.Flag.G < 2 {
|
|
|
|
|
os.Exit(0)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
g := irgen{
|
|
|
|
|
target: typecheck.Target,
|
|
|
|
|
self: pkg,
|
2021-05-27 02:47:25 -07:00
|
|
|
info: info,
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
posMap: m,
|
|
|
|
|
objs: make(map[types2.Object]*ir.Name),
|
2021-01-19 10:38:33 -08:00
|
|
|
typs: make(map[types2.Type]*types.Type),
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
}
|
|
|
|
|
g.generate(noders)
|
|
|
|
|
|
|
|
|
|
if base.Flag.G < 3 {
|
|
|
|
|
os.Exit(0)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2021-12-21 07:59:16 -08:00
|
|
|
// Information about sub-dictionary entries in a dictionary
|
|
|
|
|
type subDictInfo struct {
|
|
|
|
|
// Call or XDOT node that requires a dictionary.
|
|
|
|
|
callNode ir.Node
|
|
|
|
|
// Saved CallExpr.X node (*ir.SelectorExpr or *InstExpr node) for a generic
|
|
|
|
|
// method or function call, since this node will get dropped when the generic
|
|
|
|
|
// method/function call is transformed to a call on the instantiated shape
|
|
|
|
|
// function. Nil for other kinds of calls or XDOTs.
|
|
|
|
|
savedXNode ir.Node
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
// dictInfo is the dictionary format for an instantiation of a generic function with
|
|
|
|
|
// particular shapes. shapeParams, derivedTypes, subDictCalls, and itabConvs describe
|
|
|
|
|
// the actual dictionary entries in order, and the remaining fields are other info
|
|
|
|
|
// needed in doing dictionary processing during compilation.
|
|
|
|
|
type dictInfo struct {
|
|
|
|
|
// Types substituted for the type parameters, which are shape types.
|
|
|
|
|
shapeParams []*types.Type
|
|
|
|
|
// All types derived from those typeparams used in the instantiation.
|
2021-06-07 18:13:15 -07:00
|
|
|
derivedTypes []*types.Type
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
// Nodes in the instantiation that requires a subdictionary. Includes
|
2021-06-28 18:04:58 -07:00
|
|
|
// method and function calls (OCALL), function values (OFUNCINST), method
|
|
|
|
|
// values/expressions (OXDOT).
|
2021-12-21 07:59:16 -08:00
|
|
|
subDictCalls []subDictInfo
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
// Nodes in the instantiation that are a conversion from a typeparam/derived
|
[dev.typeparams] cmd/compile: add dictionary entries for itab conversion
This fix the case where a type param or derived type is converted to a
non-empty interface. Previously, we were converting to an empty
interface and then using DOTTYPE to convert to the correct non-empty
interface. In that case, we can get the needed itab directly from the
dictionary. This is needed for correctness from shapes, if the
destination interface is parameterized, else we will incorrectly convert
to the shape version of the interface.
Creating/writing an itab can involve generating wrappers for a bunch of
methods, which may use dictionaries. So, all the
dictionaries/instantiations are being generated on the fly and have
recursive relationships, it is simplest to finish creating/writing the
itabs at the end of the stenciling phase. So, we create a list of the
dictionaries which need to be completed by writing out their itab
entries.
The existing tests ordered.go, ifaceconv.go, and issue44688.go make use
of this optimization.
Got itab conversions for bound calls working, except for 13.go.
Also, want to get rid of the concretify, but I think we need more info
on the Bound from types2.
Change-Id: If552958a7b8a435500d6cc42c401572c367b30d1
Reviewed-on: https://go-review.googlesource.com/c/go/+/336993
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-07-12 19:34:15 -07:00
|
|
|
// type to a specific interface.
|
|
|
|
|
itabConvs []ir.Node
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
|
|
|
|
|
// Mapping from each shape type that substitutes a type param, to its
|
2021-10-13 01:39:13 +00:00
|
|
|
// type bound (which is also substituted with shapes if it is parameterized)
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
shapeToBound map[*types.Type]*types.Type
|
|
|
|
|
|
2021-08-03 08:10:17 -07:00
|
|
|
// For type switches on nonempty interfaces, a map from OTYPE entries of
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
// HasShape type, to the interface type we're switching from.
|
2021-08-03 08:10:17 -07:00
|
|
|
type2switchType map[ir.Node]*types.Type
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
|
|
|
|
|
startSubDict int // Start of dict entries for subdictionaries
|
|
|
|
|
startItabConv int // Start of dict entries for itab conversions
|
|
|
|
|
dictLen int // Total number of entries in dictionary
|
2021-06-07 18:13:15 -07:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
// instInfo is information gathered on an shape instantiation of a function.
|
2021-06-28 18:04:58 -07:00
|
|
|
type instInfo struct {
|
|
|
|
|
fun *ir.Func // The instantiated function (with body)
|
|
|
|
|
dictParam *ir.Name // The node inside fun that refers to the dictionary param
|
|
|
|
|
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
dictInfo *dictInfo
|
2021-06-28 18:04:58 -07:00
|
|
|
}
|
|
|
|
|
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
type irgen struct {
|
|
|
|
|
target *ir.Package
|
|
|
|
|
self *types2.Package
|
|
|
|
|
info *types2.Info
|
|
|
|
|
|
|
|
|
|
posMap
|
|
|
|
|
objs map[types2.Object]*ir.Name
|
2021-01-19 10:38:33 -08:00
|
|
|
typs map[types2.Type]*types.Type
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
marker dwarfgen.ScopeMarker
|
cmd/compile: get instantiated generic types working with interfaces
Get instantiatiated generic types working with interfaces, including
typechecking assignments to interfaces and instantiating all the methods
properly. To get it all working, this change includes:
- Add support for substituting in interfaces in subster.typ()
- Fill in the info for the methods for all instantiated generic types,
so those methods will be available for later typechecking (by the old
typechecker) when assigning an instantiated generic type to an
interface. We also want those methods available so we have the list
when we want to instantiate all methods of an instantiated type. We
have both for instantiated types encountered during the initial noder
phase, and for instantiated types created during stenciling of a
function/method.
- When we first create a fully-instantiated generic type (whether
during initial noder2 pass or while instantiating a method/function),
add it to a list so that all of its methods will also be
instantiated. This is needed so that an instantiated type can be
assigned to an interface.
- Properly substitute type names in the names of instantiated methods.
- New accessor methods for types.Type.RParam.
- To deal with generic types which are empty structs (or just don't use
their type params anywhere), we want to set HasTParam if a named type
has any type params that are not fully instantiated, even if the
type param is not used in the type.
- In subst.typ() and elsewhere, always set sym.Def for a new forwarding
type we are creating, so we always create a single unique type for
each generic type instantiation. This handles recursion within a
type, and also recursive relationships across many types or methods.
We remove the seen[] hashtable, which was serving the same purpose,
but for subst.typ() only. We now handle all kinds of recursive types.
- We don't seem to need to force types.CheckSize() on
created/substituted generic types anymore, so commented out for now.
- Add an RParams accessor to types2.Signature, and also a new
exported types2.AsSignature() function.
Change-Id: If6c5dd98427b20bfe9de3379cc16f83df9c9b632
Reviewed-on: https://go-review.googlesource.com/c/go/+/298449
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
2021-03-03 13:33:27 -08:00
|
|
|
|
2021-08-24 00:32:30 -07:00
|
|
|
// laterFuncs records tasks that need to run after all declarations
|
|
|
|
|
// are processed.
|
|
|
|
|
laterFuncs []func()
|
2021-10-13 17:11:16 +07:00
|
|
|
// haveEmbed indicates whether the current node belongs to file that
|
|
|
|
|
// imports "embed" package.
|
|
|
|
|
haveEmbed bool
|
2021-08-24 00:32:30 -07:00
|
|
|
|
|
|
|
|
// exprStmtOK indicates whether it's safe to generate expressions or
|
|
|
|
|
// statements yet.
|
|
|
|
|
exprStmtOK bool
|
|
|
|
|
|
2021-08-18 06:28:40 -07:00
|
|
|
// types which we need to finish, by doing g.fillinMethods.
|
|
|
|
|
typesToFinalize []*typeDelayInfo
|
|
|
|
|
|
2021-08-05 23:26:21 -07:00
|
|
|
// True when we are compiling a top-level generic function or method. Use to
|
|
|
|
|
// avoid adding closures of generic functions/methods to the target.Decls
|
|
|
|
|
// list.
|
|
|
|
|
topFuncIsGeneric bool
|
cmd/compile: remove typeparam subscripts, use func/type prefix for uniqueness
In types1 and for the export format, we were using type param subscripts
coming from types2 (originally for debugging) to provide unique names.
We need unique full-names for type params in types1 to ensure consistent
references to type params in function/method bodies and type params
derived from translation from types2. We also currently need unique
names for type params in importer/iimport.go and gcimporter/iimport.go,
because there are no levels of scoping in the package symbol lookup and
pkgIndex table.
As a step to eliminate the typeparam subscripts (which have no
relation to the source code), we change so that the typeparams' unique
name is just prefixing the type param name with the name of the
enclosing generic function, type, or method.
We now no longer use types2.TypeString in types2-to-types1 translation,
so Typestring can be changed to eliminate the subscript, as needed.
Also, types2.TypeParam.SetId() is no longer needed and is eliminated.
We can decide later if we want to do the further step of adding scoping
to the importer/iimport.go and gcimporter/iimport.go, which could be
used to eliminate the type param "path" prefix from the export format.
Change-Id: I0e37795664be2c2e1869b8f9e93393b83fc56409
Reviewed-on: https://go-review.googlesource.com/c/go/+/353135
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
Reviewed-by: Robert Findley <rfindley@google.com>
2021-09-29 10:54:27 -07:00
|
|
|
|
|
|
|
|
// The context during type/function/method declarations that is used to
|
|
|
|
|
// uniquely name type parameters. We need unique names for type params so we
|
|
|
|
|
// can be sure they match up correctly between types2-to-types1 translation
|
|
|
|
|
// and types1 importing.
|
|
|
|
|
curDecl string
|
[dev.typeparams] cmd/compile: add dictionary entries for itab conversion
This fix the case where a type param or derived type is converted to a
non-empty interface. Previously, we were converting to an empty
interface and then using DOTTYPE to convert to the correct non-empty
interface. In that case, we can get the needed itab directly from the
dictionary. This is needed for correctness from shapes, if the
destination interface is parameterized, else we will incorrectly convert
to the shape version of the interface.
Creating/writing an itab can involve generating wrappers for a bunch of
methods, which may use dictionaries. So, all the
dictionaries/instantiations are being generated on the fly and have
recursive relationships, it is simplest to finish creating/writing the
itabs at the end of the stenciling phase. So, we create a list of the
dictionaries which need to be completed by writing out their itab
entries.
The existing tests ordered.go, ifaceconv.go, and issue44688.go make use
of this optimization.
Got itab conversions for bound calls working, except for 13.go.
Also, want to get rid of the concretify, but I think we need more info
on the Bound from types2.
Change-Id: If552958a7b8a435500d6cc42c401572c367b30d1
Reviewed-on: https://go-review.googlesource.com/c/go/+/336993
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-07-12 19:34:15 -07:00
|
|
|
}
|
|
|
|
|
|
2021-09-19 09:13:47 -07:00
|
|
|
// genInst has the information for creating needed instantiations and modifying
|
|
|
|
|
// functions to use instantiations.
|
|
|
|
|
type genInst struct {
|
|
|
|
|
dnum int // for generating unique dictionary variables
|
|
|
|
|
|
|
|
|
|
// Map from the names of all instantiations to information about the
|
|
|
|
|
// instantiations.
|
|
|
|
|
instInfoMap map[*types.Sym]*instInfo
|
|
|
|
|
|
|
|
|
|
// Dictionary syms which we need to finish, by writing out any itabconv
|
|
|
|
|
// entries.
|
|
|
|
|
dictSymsToFinalize []*delayInfo
|
|
|
|
|
|
|
|
|
|
// New instantiations created during this round of buildInstantiations().
|
|
|
|
|
newInsts []ir.Node
|
|
|
|
|
}
|
|
|
|
|
|
2021-08-24 00:32:30 -07:00
|
|
|
func (g *irgen) later(fn func()) {
|
|
|
|
|
g.laterFuncs = append(g.laterFuncs, fn)
|
|
|
|
|
}
|
|
|
|
|
|
[dev.typeparams] cmd/compile: add dictionary entries for itab conversion
This fix the case where a type param or derived type is converted to a
non-empty interface. Previously, we were converting to an empty
interface and then using DOTTYPE to convert to the correct non-empty
interface. In that case, we can get the needed itab directly from the
dictionary. This is needed for correctness from shapes, if the
destination interface is parameterized, else we will incorrectly convert
to the shape version of the interface.
Creating/writing an itab can involve generating wrappers for a bunch of
methods, which may use dictionaries. So, all the
dictionaries/instantiations are being generated on the fly and have
recursive relationships, it is simplest to finish creating/writing the
itabs at the end of the stenciling phase. So, we create a list of the
dictionaries which need to be completed by writing out their itab
entries.
The existing tests ordered.go, ifaceconv.go, and issue44688.go make use
of this optimization.
Got itab conversions for bound calls working, except for 13.go.
Also, want to get rid of the concretify, but I think we need more info
on the Bound from types2.
Change-Id: If552958a7b8a435500d6cc42c401572c367b30d1
Reviewed-on: https://go-review.googlesource.com/c/go/+/336993
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-07-12 19:34:15 -07:00
|
|
|
type delayInfo struct {
|
cmd/compile: switch to computing dict format on instantiated functions
Change to computing the dictionary form on each shape-instantiated
function, rather than once on the underlying generic function/method.
The problem with computing the dictionary format on the generic function
is that we had to force early transformations to create all the
needed/implicit CONVIFACE nodes, since many of these nodes cause the
need for a dictionary entry. Also, the dictionary entries needed can
different with different instantiations of the same generic function,
especially depending on whether a type argument is a non-interface or
interface type, or a instantiated type vs. a non-instantiated type.
By computing the dictionary format on the instantiated function, we are
scanning a function where all the transformations have been done to
create implicit CONVFIFACE nodes, and we know the above relevant
information about the type params (which are shapes).
Much of the change is more mechanical changes from typeparams to shapes,
and generic functions/info to instantiated functions/info. Some of the
most important non-mechanical changes are:
- Separated out the dictionary transformations to nodes into a separate
dictPass, since we need to analyze instantiated functions after
stenciling, but before the dictionary transformations.
- Added type param index to shape types, since we need to be able
distinguish type params of an instantiation which are different but
happen to have the same shape.
- Allow the type substituter to work with shapes again (since for the
dictionary entries we need to substitute shape params to the concrete
type args).
- Support types.IdentityStrict() that does strict type comparison (no
special case for shapes). This needed for type substitution,
formatting and creating dictionaries, etc. We can maybe create better
names for this function.
- Add new information to instInfo to contain a mapping from the shape
type params to their instantiated type bound. This is needed when
doing the dictionary transformations related to type bounds.
Change-Id: I1c3ca312c5384f318c4dd7d0858dba9766396ff6
Reviewed-on: https://go-review.googlesource.com/c/go/+/349613
Trust: Dan Scales <danscales@google.com>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
2021-09-13 13:53:18 -07:00
|
|
|
gf *ir.Name
|
|
|
|
|
targs []*types.Type
|
|
|
|
|
sym *types.Sym
|
|
|
|
|
off int
|
|
|
|
|
isMeth bool
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
}
|
|
|
|
|
|
2021-08-18 06:28:40 -07:00
|
|
|
type typeDelayInfo struct {
|
|
|
|
|
typ *types2.Named
|
|
|
|
|
ntyp *types.Type
|
|
|
|
|
}
|
|
|
|
|
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
func (g *irgen) generate(noders []*noder) {
|
|
|
|
|
types.LocalPkg.Name = g.self.Name()
|
2021-05-27 02:48:33 -07:00
|
|
|
types.LocalPkg.Height = g.self.Height()
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
typecheck.TypecheckAllowed = true
|
|
|
|
|
|
2021-01-20 12:54:23 -08:00
|
|
|
// Prevent size calculations until we set the underlying type
|
|
|
|
|
// for all package-block defined types.
|
|
|
|
|
types.DeferCheckSize()
|
|
|
|
|
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
// At this point, types2 has already handled name resolution and
|
|
|
|
|
// type checking. We just need to map from its object and type
|
|
|
|
|
// representations to those currently used by the rest of the
|
2021-08-24 00:32:30 -07:00
|
|
|
// compiler. This happens in a few passes.
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
|
|
|
|
|
// 1. Process all import declarations. We use the compiler's own
|
|
|
|
|
// importer for this, rather than types2's gcimporter-derived one,
|
|
|
|
|
// to handle extensions and inline function bodies correctly.
|
|
|
|
|
//
|
|
|
|
|
// Also, we need to do this in a separate pass, because mappings are
|
|
|
|
|
// instantiated on demand. If we interleaved processing import
|
|
|
|
|
// declarations with other declarations, it's likely we'd end up
|
|
|
|
|
// wanting to map an object/type from another source file, but not
|
|
|
|
|
// yet have the import data it relies on.
|
|
|
|
|
declLists := make([][]syntax.Decl, len(noders))
|
|
|
|
|
Outer:
|
|
|
|
|
for i, p := range noders {
|
|
|
|
|
g.pragmaFlags(p.file.Pragma, ir.GoBuildPragma)
|
|
|
|
|
for j, decl := range p.file.DeclList {
|
|
|
|
|
switch decl := decl.(type) {
|
|
|
|
|
case *syntax.ImportDecl:
|
|
|
|
|
g.importDecl(p, decl)
|
|
|
|
|
default:
|
|
|
|
|
declLists[i] = p.file.DeclList[j:]
|
|
|
|
|
continue Outer // no more ImportDecls
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// 2. Process all package-block type declarations. As with imports,
|
|
|
|
|
// we need to make sure all types are properly instantiated before
|
|
|
|
|
// trying to map any expressions that utilize them. In particular,
|
|
|
|
|
// we need to make sure type pragmas are already known (see comment
|
|
|
|
|
// in irgen.typeDecl).
|
|
|
|
|
//
|
|
|
|
|
// We could perhaps instead defer processing of package-block
|
|
|
|
|
// variable initializers and function bodies, like noder does, but
|
|
|
|
|
// special-casing just package-block type declarations minimizes the
|
|
|
|
|
// differences between processing package-block and function-scoped
|
|
|
|
|
// declarations.
|
|
|
|
|
for _, declList := range declLists {
|
|
|
|
|
for _, decl := range declList {
|
|
|
|
|
switch decl := decl.(type) {
|
|
|
|
|
case *syntax.TypeDecl:
|
|
|
|
|
g.typeDecl((*ir.Nodes)(&g.target.Decls), decl)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
2021-01-20 12:54:23 -08:00
|
|
|
types.ResumeCheckSize()
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
|
|
|
|
|
// 3. Process all remaining declarations.
|
2021-10-13 17:11:16 +07:00
|
|
|
for i, declList := range declLists {
|
|
|
|
|
old := g.haveEmbed
|
|
|
|
|
g.haveEmbed = noders[i].importedEmbed
|
2021-08-24 00:32:30 -07:00
|
|
|
g.decls((*ir.Nodes)(&g.target.Decls), declList)
|
2021-10-13 17:11:16 +07:00
|
|
|
g.haveEmbed = old
|
2021-08-24 00:32:30 -07:00
|
|
|
}
|
|
|
|
|
g.exprStmtOK = true
|
|
|
|
|
|
|
|
|
|
// 4. Run any "later" tasks. Avoid using 'range' so that tasks can
|
|
|
|
|
// recursively queue further tasks. (Not currently utilized though.)
|
|
|
|
|
for len(g.laterFuncs) > 0 {
|
|
|
|
|
fn := g.laterFuncs[0]
|
|
|
|
|
g.laterFuncs = g.laterFuncs[1:]
|
|
|
|
|
fn()
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
}
|
2021-01-19 13:54:33 -08:00
|
|
|
|
|
|
|
|
if base.Flag.W > 1 {
|
|
|
|
|
for _, n := range g.target.Decls {
|
|
|
|
|
s := fmt.Sprintf("\nafter noder2 %v", n)
|
|
|
|
|
ir.Dump(s, n)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
for _, p := range noders {
|
|
|
|
|
// Process linkname and cgo pragmas.
|
|
|
|
|
p.processPragmas()
|
|
|
|
|
|
|
|
|
|
// Double check for any type-checking inconsistencies. This can be
|
|
|
|
|
// removed once we're confident in IR generation results.
|
2021-06-23 12:08:42 -07:00
|
|
|
syntax.Crawl(p.file, func(n syntax.Node) bool {
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
g.validate(n)
|
|
|
|
|
return false
|
|
|
|
|
})
|
|
|
|
|
}
|
2021-02-03 15:45:26 -08:00
|
|
|
|
2021-09-01 10:13:41 +07:00
|
|
|
if base.Flag.Complete {
|
|
|
|
|
for _, n := range g.target.Decls {
|
|
|
|
|
if fn, ok := n.(*ir.Func); ok {
|
|
|
|
|
if fn.Body == nil && fn.Nname.Sym().Linkname == "" {
|
|
|
|
|
base.ErrorfAt(fn.Pos(), "missing function body")
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Check for unusual case where noder2 encounters a type error that types2
|
|
|
|
|
// doesn't check for (e.g. notinheap incompatibility).
|
|
|
|
|
base.ExitIfErrors()
|
|
|
|
|
|
|
|
|
|
typecheck.DeclareUniverse()
|
|
|
|
|
|
2021-09-19 09:13:47 -07:00
|
|
|
// Create any needed instantiations of generic functions and transform
|
|
|
|
|
// existing and new functions to use those instantiations.
|
|
|
|
|
BuildInstantiations(true)
|
2021-02-03 15:45:26 -08:00
|
|
|
|
[dev.typeparams] cmd/compile: get export/import of generic types & functions working
The general idea is that we now export/import typeparams, typeparam
lists for generic types and functions, and instantiated types
(instantiations of generic types with either new typeparams or concrete
types).
This changes the export format -- the next CL in the stack adds the
export versions and checks for it in the appropriate places.
We always export/import generic function bodies, using the same code
that we use for exporting/importing the bodies of inlineable functions.
To avoid complicated scoping, we consider all type params as unique and
give them unique names for types1. We therefore include the types2 ids
(subscripts) in the export format and re-create on import. We always
access the same unique types1 typeParam type for the same typeparam
name.
We create fully-instantiated generic types and functions in the original
source package. We do an extra NeedRuntimeType() call to make sure that
the correct DWARF information is written out. We call SetDupOK(true) for
the functions/methods to have the linker automatically drop duplicate
instantiations.
Other miscellaneous details:
- Export/import of typeparam bounds works for methods (but not
typelists) for now, but will change with the typeset changes.
- Added a new types.Instantiate function roughly analogous to the
types2.Instantiate function recently added.
- Always access methods info from the original/base generic type, since
the methods of an instantiated type are not filled in (in types2 or
types1).
- New field OrigSym in types.Type to keep track of base generic type
that instantiated type was based on. We use the generic type's symbol
(OrigSym) as the link, rather than a Type pointer, since we haven't
always created the base type yet when we want to set the link (during
types2 to types1 conversion).
- Added types2.AsTypeParam(), (*types2.TypeParam).SetId()
- New test minimp.dir, which tests use of generic function Min across
packages. Another test stringimp.dir, which also exports a generic
function Stringify across packages, where the type param has a bound
(Stringer) as well. New test pairimp.dir, which tests use of generic
type Pair (with no methods) across packages.
- New test valimp.dir, which tests use of generic type (with methods
and related functions) across packages.
- Modified several other tests (adder.go, settable.go, smallest.go,
stringable.go, struct.go, sum.go) to export their generic
functions/types to show that generic functions/types can be exported
successfully (but this doesn't test import).
Change-Id: Ie61ce9d54a46d368ddc7a76c41399378963bb57f
Reviewed-on: https://go-review.googlesource.com/c/go/+/319930
Trust: Dan Scales <danscales@google.com>
Trust: Robert Griesemer <gri@golang.org>
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
2021-04-13 15:37:36 -07:00
|
|
|
// Remove all generic functions from g.target.Decl, since they have been
|
|
|
|
|
// used for stenciling, but don't compile. Generic functions will already
|
|
|
|
|
// have been marked for export as appropriate.
|
2021-02-03 15:45:26 -08:00
|
|
|
j := 0
|
|
|
|
|
for i, decl := range g.target.Decls {
|
2021-02-11 10:50:20 -08:00
|
|
|
if decl.Op() != ir.ODCLFUNC || !decl.Type().HasTParam() {
|
2021-02-03 15:45:26 -08:00
|
|
|
g.target.Decls[j] = g.target.Decls[i]
|
|
|
|
|
j++
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
g.target.Decls = g.target.Decls[:j]
|
2021-08-24 00:32:30 -07:00
|
|
|
|
|
|
|
|
base.Assertf(len(g.laterFuncs) == 0, "still have %d later funcs", len(g.laterFuncs))
|
[dev.typeparams] cmd/compile: add types2-based noder
This CL adds "irgen", a new noding implementation that utilizes types2
to guide IR construction. Notably, it completely skips dealing with
constant and type expressions (aside from using ir.TypeNode to
interoperate with the types1 typechecker), because types2 already
handled those. It also omits any syntax checking, trusting that types2
already rejected any errors.
It currently still utilizes the types1 typechecker for the desugaring
operations it handles (e.g., turning OAS2 into OAS2FUNC/etc, inserting
implicit conversions, rewriting f(g()) functions, and so on). However,
the IR is constructed in a fully incremental fashion, so it should be
easy to now piecemeal replace those dependencies as needed.
Nearly all of "go test std cmd" passes with -G=3 enabled by
default. The main remaining blocker is the number of test/run.go
failures. There also appear to be cases where types2 does not provide
us with position information. These will be iterated upon.
Portions and ideas from Dan Scales's CL 276653.
Change-Id: Ic99e8f2d0267b0312d30c10d5d043f5817a59c9d
Reviewed-on: https://go-review.googlesource.com/c/go/+/281932
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Dan Scales <danscales@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Trust: Matthew Dempsky <mdempsky@google.com>
Trust: Robert Griesemer <gri@golang.org>
2021-01-09 00:57:55 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func (g *irgen) unhandled(what string, p poser) {
|
|
|
|
|
base.FatalfAt(g.pos(p), "unhandled %s: %T", what, p)
|
|
|
|
|
panic("unreachable")
|
|
|
|
|
}
|
2021-09-24 23:11:18 +07:00
|
|
|
|
|
|
|
|
// delayTransform returns true if we should delay all transforms, because we are
|
|
|
|
|
// creating the nodes for a generic function/method.
|
|
|
|
|
func (g *irgen) delayTransform() bool {
|
|
|
|
|
return g.topFuncIsGeneric
|
|
|
|
|
}
|