Files
sourcekit-lsp/Sources/SemanticIndex/UpdateIndexStoreTaskDescription.swift
Alex Hoppen eb784ebdd7 Log when the build graph starts being generated and when it finishes
The signposts aren’t easily visible in the log (you need to add `--signpost` to `log show`) and don’t get logged on non-Darwin platforms at all. Add logging for it.
2024-05-24 11:24:32 -07:00

520 lines
20 KiB
Swift

//===----------------------------------------------------------------------===//
//
// This source file is part of the Swift.org open source project
//
// Copyright (c) 2014 - 2024 Apple Inc. and the Swift project authors
// Licensed under Apache License v2.0 with Runtime Library Exception
//
// See https://swift.org/LICENSE.txt for license information
// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
//
//===----------------------------------------------------------------------===//
import CAtomics
import Foundation
import LSPLogging
import LanguageServerProtocol
import SKCore
import SKSupport
import struct TSCBasic.AbsolutePath
import class TSCBasic.Process
private nonisolated(unsafe) var updateIndexStoreIDForLogging = AtomicUInt32(initialValue: 1)
public enum FileToIndex: CustomLogStringConvertible {
/// A non-header file
case indexableFile(DocumentURI)
/// A header file where `mainFile` should be indexed to update the index of `header`.
case headerFile(header: DocumentURI, mainFile: DocumentURI)
/// The file whose index store should be updated.
///
/// This file might be a header file that doesn't have build settings associated with it. For the actual compiler
/// invocation that updates the index store, the `mainFile` should be used.
public var sourceFile: DocumentURI {
switch self {
case .indexableFile(let uri): return uri
case .headerFile(header: let header, mainFile: _): return header
}
}
/// The file that should be used for compiler invocations that update the index.
///
/// If the `sourceFile` is a header file, this will be a main file that includes the header. Otherwise, it will be the
/// same as `sourceFile`.
var mainFile: DocumentURI {
switch self {
case .indexableFile(let uri): return uri
case .headerFile(header: _, mainFile: let mainFile): return mainFile
}
}
public var description: String {
switch self {
case .indexableFile(let uri):
return uri.description
case .headerFile(header: let header, mainFile: let mainFile):
return "\(header.description) using main file \(mainFile.description)"
}
}
public var redactedDescription: String {
switch self {
case .indexableFile(let uri):
return uri.redactedDescription
case .headerFile(header: let header, mainFile: let mainFile):
return "\(header.redactedDescription) using main file \(mainFile.redactedDescription)"
}
}
}
/// A file to index and the target in which the file should be indexed.
public struct FileAndTarget: Sendable {
public let file: FileToIndex
public let target: ConfiguredTarget
}
/// Describes a task to index a set of source files.
///
/// This task description can be scheduled in a `TaskScheduler`.
public struct UpdateIndexStoreTaskDescription: IndexTaskDescription {
public static let idPrefix = "update-indexstore"
public let id = updateIndexStoreIDForLogging.fetchAndIncrement()
/// The files that should be indexed.
public let filesToIndex: [FileAndTarget]
/// The build system manager that is used to get the toolchain and build settings for the files to index.
private let buildSystemManager: BuildSystemManager
private let indexStoreUpToDateStatus: IndexUpToDateStatusManager<DocumentURI>
/// A reference to the underlying index store. Used to check if the index is already up-to-date for a file, in which
/// case we don't need to index it again.
private let index: UncheckedIndex
/// See `SemanticIndexManager.indexProcessDidProduceResult`
private let indexProcessDidProduceResult: @Sendable (IndexProcessResult) -> Void
/// Test hooks that should be called when the index task finishes.
private let testHooks: IndexTestHooks
/// The task is idempotent because indexing the same file twice produces the same result as indexing it once.
public var isIdempotent: Bool { true }
public var estimatedCPUCoreCount: Int { 1 }
public var description: String {
return self.redactedDescription
}
public var redactedDescription: String {
return "update-indexstore-\(id)"
}
init(
filesToIndex: [FileAndTarget],
buildSystemManager: BuildSystemManager,
index: UncheckedIndex,
indexStoreUpToDateStatus: IndexUpToDateStatusManager<DocumentURI>,
indexProcessDidProduceResult: @escaping @Sendable (IndexProcessResult) -> Void,
testHooks: IndexTestHooks
) {
self.filesToIndex = filesToIndex
self.buildSystemManager = buildSystemManager
self.index = index
self.indexStoreUpToDateStatus = indexStoreUpToDateStatus
self.indexProcessDidProduceResult = indexProcessDidProduceResult
self.testHooks = testHooks
}
public func execute() async {
// Only use the last two digits of the indexing ID for the logging scope to avoid creating too many scopes.
// See comment in `withLoggingScope`.
// The last 2 digits should be sufficient to differentiate between multiple concurrently running indexing operation.
await withLoggingSubsystemAndScope(subsystem: indexLoggingSubsystem, scope: "update-indexstore-\(id % 100)") {
let startDate = Date()
await testHooks.updateIndexStoreTaskDidStart?(self)
let filesToIndexDescription = filesToIndex.map {
$0.file.sourceFile.fileURL?.lastPathComponent ?? $0.file.sourceFile.stringValue
}
.joined(separator: ", ")
logger.log(
"Starting updating index store with priority \(Task.currentPriority.rawValue, privacy: .public): \(filesToIndexDescription)"
)
let filesToIndex = filesToIndex.sorted(by: { $0.file.sourceFile.stringValue < $1.file.sourceFile.stringValue })
// TODO (indexing): Once swiftc supports it, we should group files by target and index files within the same
// target together in one swiftc invocation.
// https://github.com/apple/sourcekit-lsp/issues/1268
for file in filesToIndex {
await updateIndexStore(forSingleFile: file.file, in: file.target)
}
await testHooks.updateIndexStoreTaskDidFinish?(self)
logger.log(
"Finished updating index store in \(Date().timeIntervalSince(startDate) * 1000, privacy: .public)ms: \(filesToIndexDescription)"
)
}
}
public func dependencies(
to currentlyExecutingTasks: [UpdateIndexStoreTaskDescription]
) -> [TaskDependencyAction<UpdateIndexStoreTaskDescription>] {
let selfMainFiles = Set(filesToIndex.map(\.file.mainFile))
return currentlyExecutingTasks.compactMap { (other) -> TaskDependencyAction<UpdateIndexStoreTaskDescription>? in
if !other.filesToIndex.lazy.map(\.file.mainFile).contains(where: { selfMainFiles.contains($0) }) {
// Disjoint sets of files can be indexed concurrently.
return nil
}
if self.filesToIndex.count < other.filesToIndex.count {
// If there is an index operation with more files already running, suspend it.
// The most common use case for this is if we schedule an entire target to be indexed in the background and then
// need a single file indexed for use interaction. We should suspend the target-wide indexing and just index
// the current file to get index data for it ASAP.
return .cancelAndRescheduleDependency(other)
} else {
return .waitAndElevatePriorityOfDependency(other)
}
}
}
private func updateIndexStore(forSingleFile file: FileToIndex, in target: ConfiguredTarget) async {
guard await !indexStoreUpToDateStatus.isUpToDate(file.sourceFile) else {
// If we know that the file is up-to-date without having ot hit the index, do that because it's fastest.
return
}
guard let sourceFileUrl = file.sourceFile.fileURL else {
// The URI is not a file, so there's nothing we can index.
return
}
guard !index.checked(for: .modifiedFiles).hasUpToDateUnit(for: sourceFileUrl, mainFile: file.mainFile.fileURL)
else {
logger.debug("Not indexing \(file.forLogging) because index has an up-to-date unit")
// We consider a file's index up-to-date if we have any up-to-date unit. Changing build settings does not
// invalidate the up-to-date status of the index.
return
}
if file.mainFile != file.sourceFile {
logger.log("Updating index store of \(file.forLogging) using main file \(file.mainFile.forLogging)")
}
guard let language = await buildSystemManager.defaultLanguage(for: file.mainFile) else {
logger.error("Not indexing \(file.forLogging) because its language could not be determined")
return
}
let buildSettings = await buildSystemManager.buildSettings(for: file.mainFile, in: target, language: language)
guard let buildSettings else {
logger.error("Not indexing \(file.forLogging) because it has no compiler arguments")
return
}
guard let toolchain = await buildSystemManager.toolchain(for: file.mainFile, language) else {
logger.error(
"Not updating index store for \(file.forLogging) because no toolchain could be determined for the document"
)
return
}
let startDate = Date()
switch language {
case .swift:
do {
try await updateIndexStore(
forSwiftFile: file.mainFile,
buildSettings: buildSettings,
toolchain: toolchain
)
} catch {
logger.error("Updating index store for \(file.forLogging) failed: \(error.forLogging)")
BuildSettingsLogger.log(settings: buildSettings, for: file.mainFile)
}
case .c, .cpp, .objective_c, .objective_cpp:
do {
try await updateIndexStore(
forClangFile: file.mainFile,
buildSettings: buildSettings,
toolchain: toolchain
)
} catch {
logger.error("Updating index store for \(file) failed: \(error.forLogging)")
BuildSettingsLogger.log(settings: buildSettings, for: file.mainFile)
}
default:
logger.error(
"Not updating index store for \(file) because it is a language that is not supported by background indexing"
)
}
await indexStoreUpToDateStatus.markUpToDate([file.sourceFile], updateOperationStartDate: startDate)
}
private func updateIndexStore(
forSwiftFile uri: DocumentURI,
buildSettings: FileBuildSettings,
toolchain: Toolchain
) async throws {
guard let swiftc = toolchain.swiftc else {
logger.error(
"Not updating index store for \(uri.forLogging) because toolchain \(toolchain.identifier) does not contain a Swift compiler"
)
return
}
let indexingArguments = adjustSwiftCompilerArgumentsForIndexStoreUpdate(
buildSettings.compilerArguments,
fileToIndex: uri
)
try await runIndexingProcess(
indexFile: uri,
buildSettings: buildSettings,
processArguments: [swiftc.pathString] + indexingArguments,
workingDirectory: buildSettings.workingDirectory.map(AbsolutePath.init(validating:))
)
}
private func updateIndexStore(
forClangFile uri: DocumentURI,
buildSettings: FileBuildSettings,
toolchain: Toolchain
) async throws {
guard let clang = toolchain.clang else {
logger.error(
"Not updating index store for \(uri.forLogging) because toolchain \(toolchain.identifier) does not contain clang"
)
return
}
let indexingArguments = adjustClangCompilerArgumentsForIndexStoreUpdate(
buildSettings.compilerArguments,
fileToIndex: uri
)
try await runIndexingProcess(
indexFile: uri,
buildSettings: buildSettings,
processArguments: [clang.pathString] + indexingArguments,
workingDirectory: buildSettings.workingDirectory.map(AbsolutePath.init(validating:))
)
}
private func runIndexingProcess(
indexFile: DocumentURI,
buildSettings: FileBuildSettings,
processArguments: [String],
workingDirectory: AbsolutePath?
) async throws {
if Task.isCancelled {
return
}
let start = ContinuousClock.now
let signposter = Logger(subsystem: LoggingScope.subsystem, category: "indexing").makeSignposter()
let signpostID = signposter.makeSignpostID()
let state = signposter.beginInterval(
"Indexing",
id: signpostID,
"Indexing \(indexFile.fileURL?.lastPathComponent ?? indexFile.pseudoPath)"
)
defer {
signposter.endInterval("Indexing", state)
}
let process = try Process.launch(
arguments: processArguments,
workingDirectory: workingDirectory
)
let result = try await process.waitUntilExitSendingSigIntOnTaskCancellation()
indexProcessDidProduceResult(
IndexProcessResult(
taskDescription: "Indexing \(indexFile.pseudoPath)",
processResult: result,
start: start
)
)
switch result.exitStatus.exhaustivelySwitchable {
case .terminated(code: 0):
break
case .terminated(code: let code):
// This most likely happens if there are compilation errors in the source file. This is nothing to worry about.
let stdout = (try? String(bytes: result.output.get(), encoding: .utf8)) ?? "<failed to decode stdout>"
let stderr = (try? String(bytes: result.stderrOutput.get(), encoding: .utf8)) ?? "<failed to decode stderr>"
// Indexing will frequently fail if the source code is in an invalid state. Thus, log the failure at a low level.
logger.debug(
"""
Updating index store for \(indexFile.forLogging) terminated with non-zero exit code \(code)
Stderr:
\(stderr)
Stdout:
\(stdout)
"""
)
BuildSettingsLogger.log(level: .debug, settings: buildSettings, for: indexFile)
case .signalled(signal: let signal):
if !Task.isCancelled {
// The indexing job finished with a signal. Could be because the compiler crashed.
// Ignore signal exit codes if this task has been cancelled because the compiler exits with SIGINT if it gets
// interrupted.
logger.error("Updating index store for \(indexFile.forLogging) signaled \(signal)")
BuildSettingsLogger.log(level: .error, settings: buildSettings, for: indexFile)
}
case .abnormal(exception: let exception):
if !Task.isCancelled {
logger.error("Updating index store for \(indexFile.forLogging) exited abnormally \(exception)")
BuildSettingsLogger.log(level: .error, settings: buildSettings, for: indexFile)
}
}
}
}
/// Adjust compiler arguments that were created for building to compiler arguments that should be used for indexing.
///
/// This removes compiler arguments that produce output files and adds arguments to index the file.
private func adjustSwiftCompilerArgumentsForIndexStoreUpdate(
_ compilerArguments: [String],
fileToIndex: DocumentURI
) -> [String] {
let optionsToRemove: [CompilerCommandLineOption] = [
.flag("c", [.singleDash]),
.flag("disable-cmo", [.singleDash]),
.flag("emit-dependencies", [.singleDash]),
.flag("emit-module-interface", [.singleDash]),
.flag("emit-module", [.singleDash]),
.flag("emit-objc-header", [.singleDash]),
.flag("incremental", [.singleDash]),
.flag("no-color-diagnostics", [.singleDash]),
.flag("parseable-output", [.singleDash]),
.flag("save-temps", [.singleDash]),
.flag("serialize-diagnostics", [.singleDash]),
.flag("use-frontend-parseable-output", [.singleDash]),
.flag("validate-clang-modules-once", [.singleDash]),
.flag("whole-module-optimization", [.singleDash]),
.option("clang-build-session-file", [.singleDash], [.separatedBySpace]),
.option("emit-module-interface-path", [.singleDash], [.separatedBySpace]),
.option("emit-module-path", [.singleDash], [.separatedBySpace]),
.option("emit-objc-header-path", [.singleDash], [.separatedBySpace]),
.option("emit-package-module-interface-path", [.singleDash], [.separatedBySpace]),
.option("emit-private-module-interface-path", [.singleDash], [.separatedBySpace]),
.option("num-threads", [.singleDash], [.separatedBySpace]),
// Technically, `-o` and the output file don't need to be separated by a space. Eg. `swiftc -oa file.swift` is
// valid and will write to an output file named `a`.
// We can't support that because the only way to know that `-output-file-map` is a different flag and not an option
// to write to an output file named `utput-file-map` is to know all compiler arguments of `swiftc`, which we don't.
.option("o", [.singleDash], [.separatedBySpace]),
.option("output-file-map", [.singleDash], [.separatedBySpace, .separatedByEqualSign]),
]
let removeFrontendFlags = [
"-experimental-skip-non-inlinable-function-bodies",
"-experimental-skip-all-function-bodies",
]
var result: [String] = []
result.reserveCapacity(compilerArguments.count)
var iterator = compilerArguments.makeIterator()
while let argument = iterator.next() {
switch optionsToRemove.firstMatch(for: argument) {
case .removeOption:
continue
case .removeOptionAndNextArgument:
_ = iterator.next()
continue
case nil:
break
}
if argument == "-Xfrontend" {
if let nextArgument = iterator.next() {
if removeFrontendFlags.contains(nextArgument) {
continue
}
result += [argument, nextArgument]
continue
}
}
result.append(argument)
}
result += [
"-index-file",
"-index-file-path", fileToIndex.pseudoPath,
// batch mode is not compatible with -index-file
"-disable-batch-mode",
// Fake an output path so that we get a different unit file for every Swift file we background index
"-index-unit-output-path", fileToIndex.pseudoPath + ".o",
]
return result
}
/// Adjust compiler arguments that were created for building to compiler arguments that should be used for indexing.
///
/// This removes compiler arguments that produce output files and adds arguments to index the file.
private func adjustClangCompilerArgumentsForIndexStoreUpdate(
_ compilerArguments: [String],
fileToIndex: DocumentURI
) -> [String] {
let optionsToRemove: [CompilerCommandLineOption] = [
// Disable writing of a depfile
.flag("M", [.singleDash]),
.flag("MD", [.singleDash]),
.flag("MMD", [.singleDash]),
.flag("MG", [.singleDash]),
.flag("MM", [.singleDash]),
.flag("MV", [.singleDash]),
// Don't create phony targets
.flag("MP", [.singleDash]),
// Don't write out compilation databases
.flag("MJ", [.singleDash]),
// Don't compile
.flag("c", [.singleDash]),
.flag("fmodules-validate-once-per-build-session", [.singleDash]),
// Disable writing of a depfile
.option("MT", [.singleDash], [.noSpace, .separatedBySpace]),
.option("MF", [.singleDash], [.noSpace, .separatedBySpace]),
.option("MQ", [.singleDash], [.noSpace, .separatedBySpace]),
// Don't write serialized diagnostic files
.option("serialize-diagnostics", [.singleDash, .doubleDash], [.separatedBySpace]),
.option("fbuild-session-file", [.singleDash], [.separatedByEqualSign]),
]
var result: [String] = []
result.reserveCapacity(compilerArguments.count)
var iterator = compilerArguments.makeIterator()
while let argument = iterator.next() {
switch optionsToRemove.firstMatch(for: argument) {
case .removeOption:
continue
case .removeOptionAndNextArgument:
_ = iterator.next()
continue
case nil:
break
}
result.append(argument)
}
result.append(
"-fsyntax-only"
)
return result
}
fileprivate extension Sequence {
/// Returns `true` if this sequence contains an element that is equal to an element in `otherSequence` when
/// considering two elements as equal if they satisfy `predicate`.
func hasIntersection(
with otherSequence: some Sequence<Element>,
where predicate: (Element, Element) -> Bool
) -> Bool {
for outer in self {
for inner in otherSequence {
if predicate(outer, inner) {
return true
}
}
}
return false
}
}