1
0
mirror of https://github.com/golang/go synced 2024-11-05 22:36:10 -07:00
go/internal/lsp/tests/tests.go
Rohan Challa 88b1720de6 internal/lsp/tests: fix regexp for removing links that contain versions
This change fixes the regex that removes the versions for links so that tests will still run under GOPATH mode. It also removes a link for an import that needed to be downloaded.

Change-Id: I7ed4f500d1bd9d2136188d30952eedb8d8aee6e4
Reviewed-on: https://go-review.googlesource.com/c/tools/+/220140
Run-TryBot: Rohan Challa <rohan@golang.org>
Reviewed-by: Rebecca Stambler <rstambler@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2020-02-19 22:25:53 +00:00

1234 lines
36 KiB
Go

// Copyright 2019 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 tests exports functionality to be used across a variety of gopls tests.
package tests
import (
"bytes"
"context"
"flag"
"fmt"
"go/ast"
"go/token"
"io/ioutil"
"os"
"path/filepath"
"regexp"
"runtime"
"sort"
"strconv"
"strings"
"sync"
"testing"
"time"
"golang.org/x/tools/go/expect"
"golang.org/x/tools/go/packages"
"golang.org/x/tools/go/packages/packagestest"
"golang.org/x/tools/internal/lsp/protocol"
"golang.org/x/tools/internal/lsp/source"
"golang.org/x/tools/internal/span"
"golang.org/x/tools/txtar"
)
const (
overlayFileSuffix = ".overlay"
goldenFileSuffix = ".golden"
inFileSuffix = ".in"
summaryFile = "summary.txt"
testModule = "golang.org/x/tools/internal/lsp"
)
var UpdateGolden = flag.Bool("golden", false, "Update golden files")
type CodeLens map[span.Span][]protocol.CodeLens
type Diagnostics map[span.URI][]source.Diagnostic
type CompletionItems map[token.Pos]*source.CompletionItem
type Completions map[span.Span][]Completion
type CompletionSnippets map[span.Span][]CompletionSnippet
type UnimportedCompletions map[span.Span][]Completion
type DeepCompletions map[span.Span][]Completion
type FuzzyCompletions map[span.Span][]Completion
type CaseSensitiveCompletions map[span.Span][]Completion
type RankCompletions map[span.Span][]Completion
type FoldingRanges []span.Span
type Formats []span.Span
type Imports []span.Span
type SuggestedFixes []span.Span
type Definitions map[span.Span]Definition
type Implementations map[span.Span][]span.Span
type Highlights map[span.Span][]span.Span
type References map[span.Span][]span.Span
type Renames map[span.Span]string
type PrepareRenames map[span.Span]*source.PrepareItem
type Symbols map[span.URI][]protocol.DocumentSymbol
type SymbolsChildren map[string][]protocol.DocumentSymbol
type SymbolInformation map[span.Span]protocol.SymbolInformation
type WorkspaceSymbols map[string][]protocol.SymbolInformation
type Signatures map[span.Span]*protocol.SignatureHelp
type Links map[span.URI][]Link
type Data struct {
Config packages.Config
Exported *packagestest.Exported
CodeLens CodeLens
Diagnostics Diagnostics
CompletionItems CompletionItems
Completions Completions
CompletionSnippets CompletionSnippets
UnimportedCompletions UnimportedCompletions
DeepCompletions DeepCompletions
FuzzyCompletions FuzzyCompletions
CaseSensitiveCompletions CaseSensitiveCompletions
RankCompletions RankCompletions
FoldingRanges FoldingRanges
Formats Formats
Imports Imports
SuggestedFixes SuggestedFixes
Definitions Definitions
Implementations Implementations
Highlights Highlights
References References
Renames Renames
PrepareRenames PrepareRenames
Symbols Symbols
symbolsChildren SymbolsChildren
symbolInformation SymbolInformation
WorkspaceSymbols WorkspaceSymbols
FuzzyWorkspaceSymbols WorkspaceSymbols
CaseSensitiveWorkspaceSymbols WorkspaceSymbols
Signatures Signatures
Links Links
t testing.TB
fragments map[string]string
dir string
Folder string
golden map[string]*Golden
ModfileFlagAvailable bool
mappersMu sync.Mutex
mappers map[span.URI]*protocol.ColumnMapper
}
type Tests interface {
CodeLens(*testing.T, span.Span, []protocol.CodeLens)
Diagnostics(*testing.T, span.URI, []source.Diagnostic)
Completion(*testing.T, span.Span, Completion, CompletionItems)
CompletionSnippet(*testing.T, span.Span, CompletionSnippet, bool, CompletionItems)
UnimportedCompletion(*testing.T, span.Span, Completion, CompletionItems)
DeepCompletion(*testing.T, span.Span, Completion, CompletionItems)
FuzzyCompletion(*testing.T, span.Span, Completion, CompletionItems)
CaseSensitiveCompletion(*testing.T, span.Span, Completion, CompletionItems)
RankCompletion(*testing.T, span.Span, Completion, CompletionItems)
FoldingRanges(*testing.T, span.Span)
Format(*testing.T, span.Span)
Import(*testing.T, span.Span)
SuggestedFix(*testing.T, span.Span)
Definition(*testing.T, span.Span, Definition)
Implementation(*testing.T, span.Span, []span.Span)
Highlight(*testing.T, span.Span, []span.Span)
References(*testing.T, span.Span, []span.Span)
Rename(*testing.T, span.Span, string)
PrepareRename(*testing.T, span.Span, *source.PrepareItem)
Symbols(*testing.T, span.URI, []protocol.DocumentSymbol)
WorkspaceSymbols(*testing.T, string, []protocol.SymbolInformation, map[string]struct{})
FuzzyWorkspaceSymbols(*testing.T, string, []protocol.SymbolInformation, map[string]struct{})
CaseSensitiveWorkspaceSymbols(*testing.T, string, []protocol.SymbolInformation, map[string]struct{})
SignatureHelp(*testing.T, span.Span, *protocol.SignatureHelp)
Link(*testing.T, span.URI, []Link)
}
type Definition struct {
Name string
IsType bool
OnlyHover bool
Src, Def span.Span
}
type CompletionTestType int
const (
// Default runs the standard completion tests.
CompletionDefault = CompletionTestType(iota)
// Unimported tests the autocompletion of unimported packages.
CompletionUnimported
// Deep tests deep completion.
CompletionDeep
// Fuzzy tests deep completion and fuzzy matching.
CompletionFuzzy
// CaseSensitive tests case sensitive completion.
CompletionCaseSensitive
// CompletionRank candidates in test must be valid and in the right relative order.
CompletionRank
)
type WorkspaceSymbolsTestType int
const (
// Default runs the standard workspace symbols tests.
WorkspaceSymbolsDefault = WorkspaceSymbolsTestType(iota)
// Fuzzy tests workspace symbols with fuzzy matching.
WorkspaceSymbolsFuzzy
// CaseSensitive tests workspace symbols with case sensitive.
WorkspaceSymbolsCaseSensitive
)
type Completion struct {
CompletionItems []token.Pos
}
type CompletionSnippet struct {
CompletionItem token.Pos
PlainSnippet string
PlaceholderSnippet string
}
type Link struct {
Src span.Span
Target string
NotePosition token.Position
}
type Golden struct {
Filename string
Archive *txtar.Archive
Modified bool
}
func Context(t testing.TB) context.Context {
return context.Background()
}
func DefaultOptions() source.Options {
o := source.DefaultOptions()
o.SupportedCodeActions = map[source.FileKind]map[protocol.CodeActionKind]bool{
source.Go: {
protocol.SourceOrganizeImports: true,
protocol.QuickFix: true,
},
source.Mod: {
protocol.SourceOrganizeImports: true,
},
source.Sum: {},
}
o.HoverKind = source.SynopsisDocumentation
o.InsertTextFormat = protocol.SnippetTextFormat
o.CompletionBudget = time.Minute
return o
}
var haveCgo = false
// For Load() to properly create the folder structure required when testing with modules.
// The directory structure of a test needs to look like the example below:
//
// - dir
// - primarymod
// - .go files
// - packages
// - go.mod (optional)
// - modules
// - repoa
// - mod1
// - .go files
// - packages
// - go.mod (optional)
// - mod2
// - repob
// - mod1
//
// All the files that are primarily being tested should be in the primarymod folder,
// any auxillary packages should be declared in the modules folder.
// The modules folder requires each module to have the following format: repo/module
// Then inside each repo/module, there can be any number of packages and files that are
// needed to test the primarymod.
func Load(t testing.TB, exporter packagestest.Exporter, dir string) []*Data {
t.Helper()
folders, err := testFolders(dir)
if err != nil {
t.Fatalf("could not get test folders for %v, %v", dir, err)
}
var data []*Data
for _, folder := range folders {
datum := &Data{
CodeLens: make(CodeLens),
Diagnostics: make(Diagnostics),
CompletionItems: make(CompletionItems),
Completions: make(Completions),
CompletionSnippets: make(CompletionSnippets),
UnimportedCompletions: make(UnimportedCompletions),
DeepCompletions: make(DeepCompletions),
FuzzyCompletions: make(FuzzyCompletions),
RankCompletions: make(RankCompletions),
CaseSensitiveCompletions: make(CaseSensitiveCompletions),
Definitions: make(Definitions),
Implementations: make(Implementations),
Highlights: make(Highlights),
References: make(References),
Renames: make(Renames),
PrepareRenames: make(PrepareRenames),
Symbols: make(Symbols),
symbolsChildren: make(SymbolsChildren),
symbolInformation: make(SymbolInformation),
WorkspaceSymbols: make(WorkspaceSymbols),
FuzzyWorkspaceSymbols: make(WorkspaceSymbols),
CaseSensitiveWorkspaceSymbols: make(WorkspaceSymbols),
Signatures: make(Signatures),
Links: make(Links),
t: t,
dir: folder,
Folder: folder,
fragments: map[string]string{},
golden: map[string]*Golden{},
mappers: map[span.URI]*protocol.ColumnMapper{},
}
if !*UpdateGolden {
summary := filepath.Join(filepath.FromSlash(folder), summaryFile+goldenFileSuffix)
if _, err := os.Stat(summary); os.IsNotExist(err) {
t.Fatalf("could not find golden file summary.txt in %#v", folder)
}
archive, err := txtar.ParseFile(summary)
if err != nil {
t.Fatalf("could not read golden file %v/%v: %v", folder, summary, err)
}
datum.golden[summaryFile] = &Golden{
Filename: summary,
Archive: archive,
}
}
modules, _ := packagestest.GroupFilesByModules(folder)
for i, m := range modules {
for fragment, operation := range m.Files {
if trimmed := strings.TrimSuffix(fragment, goldenFileSuffix); trimmed != fragment {
delete(m.Files, fragment)
goldFile := filepath.Join(m.Name, fragment)
if i == 0 {
goldFile = filepath.Join(m.Name, "primarymod", fragment)
}
archive, err := txtar.ParseFile(goldFile)
if err != nil {
t.Fatalf("could not read golden file %v: %v", fragment, err)
}
datum.golden[trimmed] = &Golden{
Filename: goldFile,
Archive: archive,
}
} else if trimmed := strings.TrimSuffix(fragment, inFileSuffix); trimmed != fragment {
delete(m.Files, fragment)
m.Files[trimmed] = operation
} else if index := strings.Index(fragment, overlayFileSuffix); index >= 0 {
delete(m.Files, fragment)
partial := fragment[:index] + fragment[index+len(overlayFileSuffix):]
overlayFile := filepath.Join(m.Name, fragment)
if i == 0 {
overlayFile = filepath.Join(m.Name, "primarymod", fragment)
}
contents, err := ioutil.ReadFile(overlayFile)
if err != nil {
t.Fatal(err)
}
m.Overlay[partial] = contents
}
}
}
if len(modules) > 0 {
// For certain LSP related tests to run, make sure that the primary
// module for the passed in directory is testModule.
modules[0].Name = testModule
}
// Add exampleModule to provide tests with another pkg.
datum.Exported = packagestest.Export(t, exporter, modules)
for _, m := range modules {
for fragment := range m.Files {
filename := datum.Exported.File(m.Name, fragment)
datum.fragments[filename] = fragment
}
}
// Turn off go/packages debug logging.
datum.Exported.Config.Logf = nil
datum.Config.Logf = nil
// Merge the exported.Config with the view.Config.
datum.Config = *datum.Exported.Config
datum.Config.Fset = token.NewFileSet()
datum.Config.Context = Context(nil)
datum.Config.ParseFile = func(fset *token.FileSet, filename string, src []byte) (*ast.File, error) {
panic("ParseFile should not be called")
}
// Do a first pass to collect special markers for completion and workspace symbols.
if err := datum.Exported.Expect(map[string]interface{}{
"item": func(name string, r packagestest.Range, _ []string) {
datum.Exported.Mark(name, r)
},
"symbol": func(name string, r packagestest.Range, _ []string) {
datum.Exported.Mark(name, r)
},
}); err != nil {
t.Fatal(err)
}
// Collect any data that needs to be used by subsequent tests.
if err := datum.Exported.Expect(map[string]interface{}{
"codelens": datum.collectCodeLens,
"diag": datum.collectDiagnostics,
"item": datum.collectCompletionItems,
"complete": datum.collectCompletions(CompletionDefault),
"unimported": datum.collectCompletions(CompletionUnimported),
"deep": datum.collectCompletions(CompletionDeep),
"fuzzy": datum.collectCompletions(CompletionFuzzy),
"casesensitive": datum.collectCompletions(CompletionCaseSensitive),
"rank": datum.collectCompletions(CompletionRank),
"snippet": datum.collectCompletionSnippets,
"fold": datum.collectFoldingRanges,
"format": datum.collectFormats,
"import": datum.collectImports,
"godef": datum.collectDefinitions,
"implementations": datum.collectImplementations,
"typdef": datum.collectTypeDefinitions,
"hover": datum.collectHoverDefinitions,
"highlight": datum.collectHighlights,
"refs": datum.collectReferences,
"rename": datum.collectRenames,
"prepare": datum.collectPrepareRenames,
"symbol": datum.collectSymbols,
"signature": datum.collectSignatures,
"link": datum.collectLinks,
"suggestedfix": datum.collectSuggestedFixes,
}); err != nil {
t.Fatal(err)
}
for _, symbols := range datum.Symbols {
for i := range symbols {
children := datum.symbolsChildren[symbols[i].Name]
symbols[i].Children = children
}
}
// Collect names for the entries that require golden files.
if err := datum.Exported.Expect(map[string]interface{}{
"godef": datum.collectDefinitionNames,
"hover": datum.collectDefinitionNames,
"workspacesymbol": datum.collectWorkspaceSymbols(WorkspaceSymbolsDefault),
"workspacesymbolfuzzy": datum.collectWorkspaceSymbols(WorkspaceSymbolsFuzzy),
"workspacesymbolcasesensitive": datum.collectWorkspaceSymbols(WorkspaceSymbolsCaseSensitive),
}); err != nil {
t.Fatal(err)
}
data = append(data, datum)
}
return data
}
func Run(t *testing.T, tests Tests, data *Data) {
t.Helper()
checkData(t, data)
eachCompletion := func(t *testing.T, cases map[span.Span][]Completion, test func(*testing.T, span.Span, Completion, CompletionItems)) {
t.Helper()
for src, exp := range cases {
for i, e := range exp {
t.Run(SpanName(src)+"_"+strconv.Itoa(i), func(t *testing.T) {
t.Helper()
if (!haveCgo || runtime.GOOS == "android") && strings.Contains(t.Name(), "cgo") {
t.Skip("test requires cgo, not supported")
}
test(t, src, e, data.CompletionItems)
})
}
}
}
eachWorkspaceSymbols := func(t *testing.T, cases map[string][]protocol.SymbolInformation, test func(*testing.T, string, []protocol.SymbolInformation, map[string]struct{})) {
t.Helper()
for query, expectedSymbols := range cases {
name := query
if name == "" {
name = "EmptyQuery"
}
t.Run(name, func(t *testing.T) {
t.Helper()
dirs := make(map[string]struct{})
for _, si := range expectedSymbols {
d := filepath.Dir(si.Location.URI.SpanURI().Filename())
if _, ok := dirs[d]; !ok {
dirs[d] = struct{}{}
}
}
test(t, query, expectedSymbols, dirs)
})
}
}
t.Run("Completion", func(t *testing.T) {
t.Helper()
eachCompletion(t, data.Completions, tests.Completion)
})
t.Run("CompletionSnippets", func(t *testing.T) {
t.Helper()
for _, placeholders := range []bool{true, false} {
for src, expecteds := range data.CompletionSnippets {
for i, expected := range expecteds {
name := SpanName(src) + "_" + strconv.Itoa(i+1)
if placeholders {
name += "_placeholders"
}
t.Run(name, func(t *testing.T) {
t.Helper()
tests.CompletionSnippet(t, src, expected, placeholders, data.CompletionItems)
})
}
}
}
})
t.Run("UnimportedCompletion", func(t *testing.T) {
t.Helper()
eachCompletion(t, data.UnimportedCompletions, tests.UnimportedCompletion)
})
t.Run("DeepCompletion", func(t *testing.T) {
t.Helper()
eachCompletion(t, data.DeepCompletions, tests.DeepCompletion)
})
t.Run("FuzzyCompletion", func(t *testing.T) {
t.Helper()
eachCompletion(t, data.FuzzyCompletions, tests.FuzzyCompletion)
})
t.Run("CaseSensitiveCompletion", func(t *testing.T) {
t.Helper()
eachCompletion(t, data.CaseSensitiveCompletions, tests.CaseSensitiveCompletion)
})
t.Run("RankCompletions", func(t *testing.T) {
t.Helper()
eachCompletion(t, data.RankCompletions, tests.RankCompletion)
})
t.Run("CodeLens", func(t *testing.T) {
t.Helper()
for spn, want := range data.CodeLens {
// Check if we should skip this URI if the -modfile flag is not available.
if shouldSkip(data, spn.URI()) {
continue
}
t.Run(SpanName(spn), func(t *testing.T) {
t.Helper()
tests.CodeLens(t, spn, want)
})
}
})
t.Run("Diagnostics", func(t *testing.T) {
t.Helper()
for uri, want := range data.Diagnostics {
// Check if we should skip this URI if the -modfile flag is not available.
if shouldSkip(data, uri) {
continue
}
t.Run(uriName(uri), func(t *testing.T) {
t.Helper()
tests.Diagnostics(t, uri, want)
})
}
})
t.Run("FoldingRange", func(t *testing.T) {
t.Helper()
for _, spn := range data.FoldingRanges {
t.Run(uriName(spn.URI()), func(t *testing.T) {
t.Helper()
tests.FoldingRanges(t, spn)
})
}
})
t.Run("Format", func(t *testing.T) {
t.Helper()
for _, spn := range data.Formats {
t.Run(uriName(spn.URI()), func(t *testing.T) {
t.Helper()
tests.Format(t, spn)
})
}
})
t.Run("Import", func(t *testing.T) {
t.Helper()
for _, spn := range data.Imports {
t.Run(uriName(spn.URI()), func(t *testing.T) {
t.Helper()
tests.Import(t, spn)
})
}
})
t.Run("SuggestedFix", func(t *testing.T) {
t.Helper()
for _, spn := range data.SuggestedFixes {
// Check if we should skip this spn if the -modfile flag is not available.
if shouldSkip(data, spn.URI()) {
continue
}
t.Run(SpanName(spn), func(t *testing.T) {
t.Helper()
tests.SuggestedFix(t, spn)
})
}
})
t.Run("Definition", func(t *testing.T) {
t.Helper()
for spn, d := range data.Definitions {
t.Run(SpanName(spn), func(t *testing.T) {
t.Helper()
if (!haveCgo || runtime.GOOS == "android") && strings.Contains(t.Name(), "cgo") {
t.Skip("test requires cgo, not supported")
}
tests.Definition(t, spn, d)
})
}
})
t.Run("Implementation", func(t *testing.T) {
t.Helper()
for spn, m := range data.Implementations {
t.Run(SpanName(spn), func(t *testing.T) {
t.Helper()
tests.Implementation(t, spn, m)
})
}
})
t.Run("Highlight", func(t *testing.T) {
t.Helper()
for pos, locations := range data.Highlights {
t.Run(SpanName(pos), func(t *testing.T) {
t.Helper()
tests.Highlight(t, pos, locations)
})
}
})
t.Run("References", func(t *testing.T) {
t.Helper()
for src, itemList := range data.References {
t.Run(SpanName(src), func(t *testing.T) {
t.Helper()
tests.References(t, src, itemList)
})
}
})
t.Run("Renames", func(t *testing.T) {
t.Helper()
for spn, newText := range data.Renames {
t.Run(uriName(spn.URI())+"_"+newText, func(t *testing.T) {
t.Helper()
tests.Rename(t, spn, newText)
})
}
})
t.Run("PrepareRenames", func(t *testing.T) {
t.Helper()
for src, want := range data.PrepareRenames {
t.Run(SpanName(src), func(t *testing.T) {
t.Helper()
tests.PrepareRename(t, src, want)
})
}
})
t.Run("Symbols", func(t *testing.T) {
t.Helper()
for uri, expectedSymbols := range data.Symbols {
t.Run(uriName(uri), func(t *testing.T) {
t.Helper()
tests.Symbols(t, uri, expectedSymbols)
})
}
})
t.Run("WorkspaceSymbols", func(t *testing.T) {
t.Helper()
eachWorkspaceSymbols(t, data.WorkspaceSymbols, tests.WorkspaceSymbols)
})
t.Run("FuzzyWorkspaceSymbols", func(t *testing.T) {
t.Helper()
eachWorkspaceSymbols(t, data.FuzzyWorkspaceSymbols, tests.FuzzyWorkspaceSymbols)
})
t.Run("CaseSensitiveWorkspaceSymbols", func(t *testing.T) {
t.Helper()
eachWorkspaceSymbols(t, data.CaseSensitiveWorkspaceSymbols, tests.CaseSensitiveWorkspaceSymbols)
})
t.Run("SignatureHelp", func(t *testing.T) {
t.Helper()
for spn, expectedSignature := range data.Signatures {
t.Run(SpanName(spn), func(t *testing.T) {
t.Helper()
tests.SignatureHelp(t, spn, expectedSignature)
})
}
})
t.Run("Link", func(t *testing.T) {
t.Helper()
for uri, wantLinks := range data.Links {
// If we are testing GOPATH, then we do not want links with
// the versions attached (pkg.go.dev/repoa/moda@v1.1.0/pkg).
if data.Exported.Exporter == packagestest.GOPATH {
re := regexp.MustCompile(`@v\d+\.\d+\.[\w-]+`)
for i, link := range wantLinks {
wantLinks[i].Target = re.ReplaceAllString(link.Target, "")
}
}
t.Run(uriName(uri), func(t *testing.T) {
t.Helper()
tests.Link(t, uri, wantLinks)
})
}
})
if *UpdateGolden {
for _, golden := range data.golden {
if !golden.Modified {
continue
}
sort.Slice(golden.Archive.Files, func(i, j int) bool {
return golden.Archive.Files[i].Name < golden.Archive.Files[j].Name
})
if err := ioutil.WriteFile(golden.Filename, txtar.Format(golden.Archive), 0666); err != nil {
t.Fatal(err)
}
}
}
}
func checkData(t *testing.T, data *Data) {
buf := &bytes.Buffer{}
diagnosticsCount := 0
for _, want := range data.Diagnostics {
diagnosticsCount += len(want)
}
linksCount := 0
for _, want := range data.Links {
linksCount += len(want)
}
definitionCount := 0
typeDefinitionCount := 0
for _, d := range data.Definitions {
if d.IsType {
typeDefinitionCount++
} else {
definitionCount++
}
}
snippetCount := 0
for _, want := range data.CompletionSnippets {
snippetCount += len(want)
}
countCompletions := func(c map[span.Span][]Completion) (count int) {
for _, want := range c {
count += len(want)
}
return count
}
countCodeLens := func(c map[span.Span][]protocol.CodeLens) (count int) {
for _, want := range c {
count += len(want)
}
return count
}
fmt.Fprintf(buf, "CodeLensCount = %v\n", countCodeLens(data.CodeLens))
fmt.Fprintf(buf, "CompletionsCount = %v\n", countCompletions(data.Completions))
fmt.Fprintf(buf, "CompletionSnippetCount = %v\n", snippetCount)
fmt.Fprintf(buf, "UnimportedCompletionsCount = %v\n", countCompletions(data.UnimportedCompletions))
fmt.Fprintf(buf, "DeepCompletionsCount = %v\n", countCompletions(data.DeepCompletions))
fmt.Fprintf(buf, "FuzzyCompletionsCount = %v\n", countCompletions(data.FuzzyCompletions))
fmt.Fprintf(buf, "RankedCompletionsCount = %v\n", countCompletions(data.RankCompletions))
fmt.Fprintf(buf, "CaseSensitiveCompletionsCount = %v\n", countCompletions(data.CaseSensitiveCompletions))
fmt.Fprintf(buf, "DiagnosticsCount = %v\n", diagnosticsCount)
fmt.Fprintf(buf, "FoldingRangesCount = %v\n", len(data.FoldingRanges))
fmt.Fprintf(buf, "FormatCount = %v\n", len(data.Formats))
fmt.Fprintf(buf, "ImportCount = %v\n", len(data.Imports))
fmt.Fprintf(buf, "SuggestedFixCount = %v\n", len(data.SuggestedFixes))
fmt.Fprintf(buf, "DefinitionsCount = %v\n", definitionCount)
fmt.Fprintf(buf, "TypeDefinitionsCount = %v\n", typeDefinitionCount)
fmt.Fprintf(buf, "HighlightsCount = %v\n", len(data.Highlights))
fmt.Fprintf(buf, "ReferencesCount = %v\n", len(data.References))
fmt.Fprintf(buf, "RenamesCount = %v\n", len(data.Renames))
fmt.Fprintf(buf, "PrepareRenamesCount = %v\n", len(data.PrepareRenames))
fmt.Fprintf(buf, "SymbolsCount = %v\n", len(data.Symbols))
fmt.Fprintf(buf, "WorkspaceSymbolsCount = %v\n", len(data.WorkspaceSymbols))
fmt.Fprintf(buf, "FuzzyWorkspaceSymbolsCount = %v\n", len(data.FuzzyWorkspaceSymbols))
fmt.Fprintf(buf, "CaseSensitiveWorkspaceSymbolsCount = %v\n", len(data.CaseSensitiveWorkspaceSymbols))
fmt.Fprintf(buf, "SignaturesCount = %v\n", len(data.Signatures))
fmt.Fprintf(buf, "LinksCount = %v\n", linksCount)
fmt.Fprintf(buf, "ImplementationsCount = %v\n", len(data.Implementations))
want := string(data.Golden("summary", summaryFile, func() ([]byte, error) {
return buf.Bytes(), nil
}))
got := buf.String()
if want != got {
t.Errorf("test summary does not match, want\n%s\ngot:\n%s", want, got)
}
}
func (data *Data) Mapper(uri span.URI) (*protocol.ColumnMapper, error) {
data.mappersMu.Lock()
defer data.mappersMu.Unlock()
if _, ok := data.mappers[uri]; !ok {
content, err := data.Exported.FileContents(uri.Filename())
if err != nil {
return nil, err
}
converter := span.NewContentConverter(uri.Filename(), content)
data.mappers[uri] = &protocol.ColumnMapper{
URI: uri,
Converter: converter,
Content: content,
}
}
return data.mappers[uri], nil
}
func (data *Data) Golden(tag string, target string, update func() ([]byte, error)) []byte {
data.t.Helper()
fragment, found := data.fragments[target]
if !found {
if filepath.IsAbs(target) {
data.t.Fatalf("invalid golden file fragment %v", target)
}
fragment = target
}
golden := data.golden[fragment]
if golden == nil {
if !*UpdateGolden {
data.t.Fatalf("could not find golden file %v: %v", fragment, tag)
}
var subdir string
if fragment != summaryFile {
subdir = "primarymod"
}
golden = &Golden{
Filename: filepath.Join(data.dir, subdir, fragment+goldenFileSuffix),
Archive: &txtar.Archive{},
Modified: true,
}
data.golden[fragment] = golden
}
var file *txtar.File
for i := range golden.Archive.Files {
f := &golden.Archive.Files[i]
if f.Name == tag {
file = f
break
}
}
if *UpdateGolden {
if file == nil {
golden.Archive.Files = append(golden.Archive.Files, txtar.File{
Name: tag,
})
file = &golden.Archive.Files[len(golden.Archive.Files)-1]
}
contents, err := update()
if err != nil {
data.t.Fatalf("could not update golden file %v: %v", fragment, err)
}
file.Data = append(contents, '\n') // add trailing \n for txtar
golden.Modified = true
}
if file == nil {
data.t.Fatalf("could not find golden contents %v: %v", fragment, tag)
}
return file.Data[:len(file.Data)-1] // drop the trailing \n
}
func (data *Data) collectCodeLens(spn span.Span, title, cmd string) {
if _, ok := data.CodeLens[spn]; !ok {
data.CodeLens[spn] = []protocol.CodeLens{}
}
m, err := data.Mapper(spn.URI())
if err != nil {
return
}
rng, err := m.Range(spn)
if err != nil {
return
}
data.CodeLens[spn] = append(data.CodeLens[spn], protocol.CodeLens{
Range: rng,
Command: protocol.Command{
Title: title,
Command: cmd,
},
})
}
func (data *Data) collectDiagnostics(spn span.Span, msgSource, msg, msgSeverity string) {
if _, ok := data.Diagnostics[spn.URI()]; !ok {
data.Diagnostics[spn.URI()] = []source.Diagnostic{}
}
m, err := data.Mapper(spn.URI())
if err != nil {
return
}
rng, err := m.Range(spn)
if err != nil {
return
}
severity := protocol.SeverityError
switch msgSeverity {
case "error":
severity = protocol.SeverityError
case "warning":
severity = protocol.SeverityWarning
case "hint":
severity = protocol.SeverityHint
case "information":
severity = protocol.SeverityInformation
}
// This is not the correct way to do this, but it seems excessive to do the full conversion here.
want := source.Diagnostic{
Range: rng,
Severity: severity,
Source: msgSource,
Message: msg,
}
data.Diagnostics[spn.URI()] = append(data.Diagnostics[spn.URI()], want)
}
func (data *Data) collectCompletions(typ CompletionTestType) func(span.Span, []token.Pos) {
result := func(m map[span.Span][]Completion, src span.Span, expected []token.Pos) {
m[src] = append(m[src], Completion{
CompletionItems: expected,
})
}
switch typ {
case CompletionDeep:
return func(src span.Span, expected []token.Pos) {
result(data.DeepCompletions, src, expected)
}
case CompletionUnimported:
return func(src span.Span, expected []token.Pos) {
result(data.UnimportedCompletions, src, expected)
}
case CompletionFuzzy:
return func(src span.Span, expected []token.Pos) {
result(data.FuzzyCompletions, src, expected)
}
case CompletionRank:
return func(src span.Span, expected []token.Pos) {
result(data.RankCompletions, src, expected)
}
case CompletionCaseSensitive:
return func(src span.Span, expected []token.Pos) {
result(data.CaseSensitiveCompletions, src, expected)
}
default:
return func(src span.Span, expected []token.Pos) {
result(data.Completions, src, expected)
}
}
}
func (data *Data) collectCompletionItems(pos token.Pos, args []string) {
if len(args) < 3 {
loc := data.Exported.ExpectFileSet.Position(pos)
data.t.Fatalf("%s:%d: @item expects at least 3 args, got %d",
loc.Filename, loc.Line, len(args))
}
label, detail, kind := args[0], args[1], args[2]
var documentation string
if len(args) == 4 {
documentation = args[3]
}
data.CompletionItems[pos] = &source.CompletionItem{
Label: label,
Detail: detail,
Kind: protocol.ParseCompletionItemKind(kind),
Documentation: documentation,
}
}
func (data *Data) collectFoldingRanges(spn span.Span) {
data.FoldingRanges = append(data.FoldingRanges, spn)
}
func (data *Data) collectFormats(spn span.Span) {
data.Formats = append(data.Formats, spn)
}
func (data *Data) collectImports(spn span.Span) {
data.Imports = append(data.Imports, spn)
}
func (data *Data) collectSuggestedFixes(spn span.Span) {
data.SuggestedFixes = append(data.SuggestedFixes, spn)
}
func (data *Data) collectDefinitions(src, target span.Span) {
data.Definitions[src] = Definition{
Src: src,
Def: target,
}
}
func (data *Data) collectImplementations(src span.Span, targets []span.Span) {
data.Implementations[src] = targets
}
func (data *Data) collectHoverDefinitions(src, target span.Span) {
data.Definitions[src] = Definition{
Src: src,
Def: target,
OnlyHover: true,
}
}
func (data *Data) collectTypeDefinitions(src, target span.Span) {
data.Definitions[src] = Definition{
Src: src,
Def: target,
IsType: true,
}
}
func (data *Data) collectDefinitionNames(src span.Span, name string) {
d := data.Definitions[src]
d.Name = name
data.Definitions[src] = d
}
func (data *Data) collectHighlights(src span.Span, expected []span.Span) {
// Declaring a highlight in a test file: @highlight(src, expected1, expected2)
data.Highlights[src] = append(data.Highlights[src], expected...)
}
func (data *Data) collectReferences(src span.Span, expected []span.Span) {
data.References[src] = expected
}
func (data *Data) collectRenames(src span.Span, newText string) {
data.Renames[src] = newText
}
func (data *Data) collectPrepareRenames(src span.Span, rng span.Range, placeholder string) {
m, err := data.Mapper(src.URI())
if err != nil {
data.t.Fatal(err)
}
// Convert range to span and then to protocol.Range.
spn, err := rng.Span()
if err != nil {
data.t.Fatal(err)
}
prng, err := m.Range(spn)
if err != nil {
data.t.Fatal(err)
}
data.PrepareRenames[src] = &source.PrepareItem{
Range: prng,
Text: placeholder,
}
}
func (data *Data) collectSymbols(name string, spn span.Span, kind string, parentName string) {
m, err := data.Mapper(spn.URI())
if err != nil {
data.t.Fatal(err)
}
rng, err := m.Range(spn)
if err != nil {
data.t.Fatal(err)
}
sym := protocol.DocumentSymbol{
Name: name,
Kind: protocol.ParseSymbolKind(kind),
SelectionRange: rng,
}
if parentName == "" {
data.Symbols[spn.URI()] = append(data.Symbols[spn.URI()], sym)
} else {
data.symbolsChildren[parentName] = append(data.symbolsChildren[parentName], sym)
}
// Reuse @symbol in the workspace symbols tests.
si := protocol.SymbolInformation{
Name: sym.Name,
Kind: sym.Kind,
Location: protocol.Location{
URI: protocol.URIFromSpanURI(spn.URI()),
Range: sym.SelectionRange,
},
}
data.symbolInformation[spn] = si
}
func (data *Data) collectWorkspaceSymbols(typ WorkspaceSymbolsTestType) func(string, []span.Span) {
switch typ {
case WorkspaceSymbolsFuzzy:
return func(query string, targets []span.Span) {
for _, target := range targets {
data.FuzzyWorkspaceSymbols[query] = append(data.FuzzyWorkspaceSymbols[query], data.symbolInformation[target])
}
}
case WorkspaceSymbolsCaseSensitive:
return func(query string, targets []span.Span) {
for _, target := range targets {
data.CaseSensitiveWorkspaceSymbols[query] = append(data.CaseSensitiveWorkspaceSymbols[query], data.symbolInformation[target])
}
}
default:
return func(query string, targets []span.Span) {
for _, target := range targets {
data.WorkspaceSymbols[query] = append(data.WorkspaceSymbols[query], data.symbolInformation[target])
}
}
}
}
func (data *Data) collectSignatures(spn span.Span, signature string, activeParam int64) {
data.Signatures[spn] = &protocol.SignatureHelp{
Signatures: []protocol.SignatureInformation{
{
Label: signature,
},
},
ActiveParameter: float64(activeParam),
}
// Hardcode special case to test the lack of a signature.
if signature == "" && activeParam == 0 {
data.Signatures[spn] = nil
}
}
func (data *Data) collectCompletionSnippets(spn span.Span, item token.Pos, plain, placeholder string) {
data.CompletionSnippets[spn] = append(data.CompletionSnippets[spn], CompletionSnippet{
CompletionItem: item,
PlainSnippet: plain,
PlaceholderSnippet: placeholder,
})
}
func (data *Data) collectLinks(spn span.Span, link string, note *expect.Note, fset *token.FileSet) {
position := fset.Position(note.Pos)
uri := spn.URI()
data.Links[uri] = append(data.Links[uri], Link{
Src: spn,
Target: link,
NotePosition: position,
})
}
func uriName(uri span.URI) string {
return filepath.Base(strings.TrimSuffix(uri.Filename(), ".go"))
}
func SpanName(spn span.Span) string {
return fmt.Sprintf("%v_%v_%v", uriName(spn.URI()), spn.Start().Line(), spn.Start().Column())
}
func CopyFolderToTempDir(folder string) (string, error) {
if _, err := os.Stat(folder); err != nil {
return "", err
}
dst, err := ioutil.TempDir("", "modfile_test")
if err != nil {
return "", err
}
fds, err := ioutil.ReadDir(folder)
if err != nil {
return "", err
}
for _, fd := range fds {
srcfp := filepath.Join(folder, fd.Name())
stat, err := os.Stat(srcfp)
if err != nil {
return "", err
}
if !stat.Mode().IsRegular() {
return "", fmt.Errorf("cannot copy non regular file %s", srcfp)
}
contents, err := ioutil.ReadFile(srcfp)
if err != nil {
return "", err
}
if err := ioutil.WriteFile(filepath.Join(dst, fd.Name()), contents, stat.Mode()); err != nil {
return "", err
}
}
return dst, nil
}
func testFolders(root string) ([]string, error) {
// Check if this only has one test directory.
if _, err := os.Stat(filepath.Join(filepath.FromSlash(root), "primarymod")); !os.IsNotExist(err) {
return []string{root}, nil
}
folders := []string{}
root = filepath.FromSlash(root)
// Get all test directories that are one level deeper than root.
if err := filepath.Walk(root, func(path string, info os.FileInfo, err error) error {
if !info.IsDir() {
return nil
}
if filepath.Dir(path) == root {
folders = append(folders, filepath.ToSlash(path))
}
return nil
}); err != nil {
return nil, err
}
return folders, nil
}
func shouldSkip(data *Data, uri span.URI) bool {
if data.ModfileFlagAvailable {
return false
}
// If the -modfile flag is not available, then we do not want to run
// any tests on the go.mod file.
if strings.Contains(uri.Filename(), ".mod") {
return true
}
// If the -modfile flag is not available, then we do not want to test any
// uri that contains "go mod tidy".
m, err := data.Mapper(uri)
return err == nil && strings.Contains(string(m.Content), ", \"go mod tidy\",")
}