Thanks Richard, that allowed me to replace a hand rolled universe scope 👍
My importer varies from yours in that for correct lookups for versioned
packages or those with '-' in I had to copy ImportPathToAssumedName from
x/tools/internal/imports/fix.go.
func simpleImporter(imports map[string]*ast.Object, path string)
(*ast.Object, error) {
pkg := imports[path]
if pkg == nil {
pkg = ast.NewObj(ast.Pkg, ImportPathToAssumedName(path))
pkg.Data = ast.NewScope(nil) // required by ast.NewPackage
for dot-import
imports[path] = pkg
}
return pkg, nil
}
This now works for all cases which don't import external packages. So now I
just need to do the on demand load of packages, which I suspect will lead
me right back to packages.Load.
On Sat, 16 Oct 2021 at 15:59, 'Richard Oudkerk' via golang-nuts <
golang-nuts@googlegroups.com> wrote:
> You could try building the universe scope for ast.NewPackage from
> types.Universe. For example
>
> https://play.golang.org/p/1E5Iu4vW3g9
>
> func NewPackage(fset *token.FileSet, files map[string]*ast.File)
> (*ast.Package, error) {
> univ, err := universe()
> if err != nil {
> return nil, err
> }
> return ast.NewPackage(fset, files, dummyImporter, univ)
> }
>
> func dummyImporter(imports map[string]*ast.Object, importPath string)
> (*ast.Object, error) {
> pkg := imports[importPath]
> if pkg == nil {
> pkg = ast.NewObj(ast.Pkg, path.Base(importPath))
> pkg.Data = ast.NewScope(nil)
> imports[importPath] = pkg
> }
> return pkg, nil
> }
>
> func universe() (*ast.Scope, error) {
> u := ast.NewScope(nil)
> for _, name := range types.Universe.Names() {
> o := types.Universe.Lookup(name)
> if o == nil {
> return nil, fmt.Errorf("failed to lookup %s in universe scope", name)
> }
> var objKind ast.ObjKind
> switch o.(type) {
> case *types.Const, *types.Nil:
> objKind = ast.Con
> case *types.TypeName:
> objKind = ast.Typ
> case *types.Builtin:
> objKind = ast.Fun
> default:
> return nil, fmt.Errorf("unexpected builtin %s of type %T", o.Name(), o)
> }
> obj := ast.NewObj(objKind, name)
> if u.Insert(obj) != nil {
> return nil, fmt.Errorf("types internal error: double declaration")
> }
> obj.Decl = u
> }
> return u, nil
> }
>
> On Saturday, 16 October 2021 at 14:38:43 UTC+1 eli...@gmail.com wrote:
>
>> On Fri, Oct 15, 2021 at 2:13 PM Steven Hartland <ste...@multiplay.co.uk>
>> wrote:
>>
>>> I converted my code to x/tools/go/packages
>>> <https://pkg.go.dev/golang.org/x/tools@v0.1.7/go/packages> and while it
>>> did solve the problem it's VERY slow in comparison.
>>>
>>> I have a set of 21 tests operating on a single package which has at most
>>> two very basic types, no imports and using go/parser
>>> <https://pkg.go.dev/go/parser> they take 0.011s but with go/packages
>>> <https://pkg.go.dev/golang.org/x/tools@v0.1.7/go/packages> that
>>> increases to 3.548s a 300x slow down.
>>>
>>> I'm setting a basic mode: packages.NeedName | packages.NeedSyntax
>>>
>>> The package.Load call takes ~220ms whereas ast.NewPackage only
>>> takes 2.7µs.
>>>
>>
>> Could you post a reproducer of your target package and analysis
>> somewhere? 220ms for packages.Load sounds like a lot. It's true that
>> packages does a lot more work than just the parser (*), but it's not
>> supposed to be that slow. In my tests a simple Load with more functionality
>> takes 60-70ms
>>
>> (*) The type checking takes a bit of time over just parsing to AST, but
>> the biggest difference is loading multiple files from imports. For type
>> checking you need to know, when you see:
>>
>> import foo
>>
>> x := foo.Foo()
>>
>> What the type of `x` is, so go/packages has to analyze the `foo` package
>> as well.
>>
>>
>>
>>>
>>> As the resulting ast.File's are pretty much the same, I'm wondering if
>>> for my use case packages.Load is doing way more than I need?
>>>
>>> Another downside is for tests run in a temporary directory outside of
>>> the package space package.Load fails with:
>>> directory /tmp/tests76985775 outside available modules
>>>
>>> I fixed it by calling ioutil.TempDir with "." but that's not ideal.
>>>
>>> Thoughts?
>>>
>>> On Tue, 12 Oct 2021 at 13:42, Steven Hartland <ste...@multiplay.co.uk>
>>> wrote:
>>>
>>>> Thanks David, much appreciated, I will have a look at both.
>>>>
>>>> When migrating from go/ast to go/types did you hit anything of note I
>>>> should look out for?
>>>>
>>>> On Mon, 11 Oct 2021 at 17:06, David Finkel <david....@gmail.com> wrote:
>>>>
>>>>>
>>>>>
>>>>> On Mon, Oct 11, 2021 at 5:48 AM Steven Hartland <
>>>>> ste...@multiplay.co.uk> wrote:
>>>>>
>>>>>> If the ast.Files passed to ast.NewPackage includes built in types
>>>>>> such as int it returns an error e.g.
>>>>>> file1.go:5:6: undeclared name: int
>>>>>>
>>>>>> Is there a way to prevent that?
>>>>>>
>>>>>
>>>>> Generally, I always add the `builtin` package to the list of packages
>>>>> I'm parsing.
>>>>> I wrote a little library for exactly this kind of package loading a
>>>>> few years ago:
>>>>> https://gitlab.com/dfinkel/goastpkg/-/blob/master/go_ast_parser.go
>>>>> (https://pkg.go.dev/golang.spin-2.net/astpkg)
>>>>>
>>>>>>
>>>>>> Playground example: https://play.golang.org/p/Yg30TTzoLHP
>>>>>>
>>>>>> My goal is to take multiple files, resolve inter file dependencies
>>>>>> e.g. a type referencing another type in a different file and process the
>>>>>> resulting ast.Files. So if there is a better way to achieve this I'm all
>>>>>> ears.
>>>>>>
>>>>>
>>>>> In general, I've stopped using the `go/ast` internal references as
>>>>> much and have started using resolved `go/types` references as they're more
>>>>> reliable and better-specified.
>>>>> (golang.org/x/tools/go/packages
>>>>> <https://pkg.go.dev/golang.org/x/tools@v0.1.7/go/packages> has a
>>>>> LoadMode flag for generating `go/types.Info` (NeedTypesInfo
>>>>> <https://pkg.go.dev/golang.org/x/tools@v0.1.7/go/packages#NeedTypesInfo>
>>>>> ))
>>>>>
>>>>>>
>>>>>> Regards
>>>>>> Steve
>>>>>>
>>>>>> --
>>>>>> You received this message because you are subscribed to the Google
>>>>>> Groups "golang-nuts" group.
>>>>>> To unsubscribe from this group and stop receiving emails from it,
>>>>>> send an email to golang-nuts...@googlegroups.com.
>>>>>> To view this discussion on the web visit
>>>>>> https://groups.google.com/d/msgid/golang-nuts/CAHEMsqbJoJxuo3c-mofMtzXXJhYCzV2skW2ZB3ZPY6WtA8%2BxHw%40mail.gmail.com
>>>>>> <https://groups.google.com/d/msgid/golang-nuts/CAHEMsqbJoJxuo3c-mofMtzXXJhYCzV2skW2ZB3ZPY6WtA8%2BxHw%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>>>>> .
>>>>>>
>>>>> --
>>> You received this message because you are subscribed to the Google
>>> Groups "golang-nuts" group.
>>> To unsubscribe from this group and stop receiving emails from it, send
>>> an email to golang-nuts...@googlegroups.com.
>>>
>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/golang-nuts/CAHEMsqYMSBUfuOUvptv6UrvBFTwFxjOhJZ5sMN-omOx5ESL5hw%40mail.gmail.com
>>> <https://groups.google.com/d/msgid/golang-nuts/CAHEMsqYMSBUfuOUvptv6UrvBFTwFxjOhJZ5sMN-omOx5ESL5hw%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>> .
>>>
>> --
> You received this message because you are subscribed to the Google Groups
> "golang-nuts" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to golang-nuts+unsubscr...@googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/golang-nuts/d570a7ce-a780-46d8-a323-f9c26a6c2561n%40googlegroups.com
> <https://groups.google.com/d/msgid/golang-nuts/d570a7ce-a780-46d8-a323-f9c26a6c2561n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/golang-nuts/CAHEMsqbR9PfZxyAEvqD9ttcC3XWpihW%3DW85utLW7pvg9ormkhw%40mail.gmail.com.
