Move The Last Pieces for Cross-Module Incremental Builds

We're going to play a dirty, dirty trick - but it'll make our users'
lives better in the end so stick with me here.

In order to build up an incremental compilation, we need two sources of
dependency information:

1) "Priors" - Swiftdeps with dependency information from the past
   build(s)
2) "Posteriors" - Swiftdeps with dependencies from after we rebuild the
   file or module or whatever

With normal swift files built in incremental mode, the priors are given by the
swiftdeps files which are generated parallel to a swift file and usually
placed in the build directory alongside the object files. Because we
have entries in the output file map, we can always know where these
swiftdeps files are. The priors are integrated by the driver and then
the build is scheduled. As the build runs and jobs complete, their
swiftdeps are reloaded and re-integrated. The resulting changes are then
traversed and more jobs are scheduled if necessary. These give us the
posteriors we desire.

A module flips this on its head. The swiftdeps information serialized
in a module functions as the *posterior* since the driver consuming the
module has no way of knowing how to rebuild the module, and because its
dependencies are, for all intents and purposes, fixed in time. The
missing piece of the puzzle is the priors. That is, we need some way of
knowing what the "past" interface of the module looked like so we can
compare it to the "present" interface. Moreover, we need to always know
where to look for these priors.

We solve this problem by serializing a file alongside the build record:
the "external" build record. This is given by a... creative encoding
of multiple source file dependency graphs into a single source file
dependency graph. The rough structure of this is:

   SourceFile => interface <BUILD_RECORD>.external
   | - Incremental External Dependency => interface <MODULE_1>.swiftmodule
   | | - <dependency> ...
   | | - <dependency> ...
   | | - <dependency> ...
   | - Incremental External Dependency => interface <MODULE_2>.swiftmodule
   | | - <dependency> ...
   | | - <dependency> ...
   | - Incremental External Dependency => interface <MODULE_3>.swiftmodule
   | - ...

Sorta, `cat`'ing a bunch of source file dependency graphs together but
with incremental external dependency nodes acting as glue.

Now for the trick:

We have to unpack this structure and integrate it to get our priors.
This is easy. The tricky bit comes in integrate itself. Because the
top-level source file node points directly at the external build record,
not the original swift modules that defined these dependency nodes, we
swap the key it wants to use (the external build record) for the
incremental external dependency acting as the "parent" of the dependency
node. We do this by following the arc we carefully laid down in the
structure above.

For rdar://69595010
Goes a long way towards rdar://48955139, rdar://64238133

Author: CodaFi

include/swift/Basic/FileTypes.def

@@ -61,6 +61,7 @@ TYPE("llvm-bc",             LLVM_BC,                   "bc",              "")
 TYPE("diagnostics",         SerializedDiagnostics,     "dia",             "")
 TYPE("objc-header",         ObjCHeader,                "h",               "")
 TYPE("swift-dependencies",  SwiftDeps,                 "swiftdeps",       "")
+TYPE("external-swift-dependencies", ExternalSwiftDeps,  "swiftdeps.external",  "")
 TYPE("swift-ranges",        SwiftRanges,               "swiftranges",     "")
 TYPE("compiled-source",     CompiledSource,            "compiledsource",  "")
 TYPE("remap",               Remapping,                 "remap",           "")

include/swift/Driver/Compilation.h

@@ -508,6 +508,17 @@ class Compilation {
   /// \sa types::isPartOfSwiftCompilation
   const char *getAllSourcesPath() const;
 
+  /// Retrieve the path to the external swift deps file.
+  ///
+  /// For cross-module incremental builds, this file contains the dependencies
+  /// from all the modules integrated over the prior build.
+  ///
+  /// Currently this patch is relative to the build record, but we may want
+  /// to allow the output file map to customize this at some point.
+  std::string getExternalSwiftDepsFilePath() const {
+    return CompilationRecordPath + ".external";
+  }
+
   /// Asks the Compilation to perform the Jobs which it knows about.
   ///
   /// \param TQ The TaskQueue used to schedule jobs for execution.

lib/Basic/FileTypes.cpp

@@ -100,6 +100,7 @@ bool file_types::isTextual(ID Id) {
   case file_types::TY_SerializedDiagnostics:
   case file_types::TY_ClangModuleFile:
   case file_types::TY_SwiftDeps:
+  case file_types::TY_ExternalSwiftDeps:
   case file_types::TY_SwiftRanges:
   case file_types::TY_CompiledSource:
   case file_types::TY_Nothing:
@@ -145,6 +146,7 @@ bool file_types::isAfterLLVM(ID Id) {
   case file_types::TY_SerializedDiagnostics:
   case file_types::TY_ClangModuleFile:
   case file_types::TY_SwiftDeps:
+  case file_types::TY_ExternalSwiftDeps:
   case file_types::TY_SwiftRanges:
   case file_types::TY_CompiledSource:
   case file_types::TY_Nothing:
@@ -197,6 +199,7 @@ bool file_types::isPartOfSwiftCompilation(ID Id) {
   case file_types::TY_SerializedDiagnostics:
   case file_types::TY_ClangModuleFile:
   case file_types::TY_SwiftDeps:
+  case file_types::TY_ExternalSwiftDeps:
   case file_types::TY_SwiftRanges:
   case file_types::TY_CompiledSource:
   case file_types::TY_Nothing:

lib/Driver/Compilation.cpp

@@ -15,6 +15,7 @@
 #include "swift/AST/DiagnosticEngine.h"
 #include "swift/AST/DiagnosticsDriver.h"
 #include "swift/AST/FineGrainedDependencies.h"
+#include "swift/AST/FineGrainedDependencyFormat.h"
 #include "swift/Basic/OutputFileMap.h"
 #include "swift/Basic/Program.h"
 #include "swift/Basic/STLExtras.h"
@@ -1185,6 +1186,30 @@ namespace driver {
       return ExternallyDependentJobs;
     }
 
+    using ChangeSet = fine_grained_dependencies::ModuleDepGraph::Changes::value_type;
+    static void
+    pruneChangeSetFromExternalDependency(ChangeSet &changes) {
+      // The changeset includes detritus from the graph that gets consed up
+      // in \c writePriorDependencyGraph. We need to ignore the fake
+      // source file provides nodes and the fake incremental external
+      // dependencies linked to them.
+      swift::erase_if(
+          changes, [&](fine_grained_dependencies::ModuleDepGraphNode *node) {
+            if (node->getKey().getKind() ==
+                    fine_grained_dependencies::NodeKind::sourceFileProvide ||
+                node->getKey().getKind() ==
+                    fine_grained_dependencies::NodeKind::
+                        incrementalExternalDepend) {
+              return true;
+            }
+            if (node->getKey().getAspect() ==
+                fine_grained_dependencies::DeclAspect::implementation) {
+              return true;
+            }
+            return !node->getIsProvides();
+          });
+    }
+  
     SmallVector<const Job *, 16>
     collectIncrementalExternallyDependentJobsFromDependencyGraph(
         const bool forRanges) {
@@ -1196,6 +1221,17 @@ namespace driver {
         }
       };
 
+      // Load our priors, which are always adjacent to the build record. We
+      // don't care if this load succeeds or not. If it fails, and we succeed at
+      // integrating one of the external files below, the changeset will be the
+      // entire module!
+      const auto *externalPriorJob = Comp.addExternalJob(
+          std::make_unique<Job>(Comp.getDerivedOutputFileMap(),
+                                Comp.getExternalSwiftDepsFilePath()));
+      getFineGrainedDepGraph(forRanges).loadFromPath(
+          externalPriorJob, Comp.getExternalSwiftDepsFilePath(),
+          Comp.getDiags());
+
       for (auto external : getFineGrainedDepGraph(forRanges)
                                .getIncrementalExternalDependencies()) {
         llvm::sys::fs::file_status depStatus;
@@ -1231,20 +1267,23 @@ namespace driver {
         // code's internal invariants.
         const auto *externalJob = Comp.addExternalJob(
             std::make_unique<Job>(Comp.getDerivedOutputFileMap(), external));
-        auto subChanges =
+        auto maybeChanges =
             getFineGrainedDepGraph(forRanges).loadFromSwiftModuleBuffer(
                 externalJob, *buffer.get(), Comp.getDiags());
 
         // If the incremental dependency graph failed to load, fall back to
         // treating this as plain external job.
-        if (!subChanges.hasValue()) {
+        if (!maybeChanges.hasValue()) {
           fallbackToExternalBehavior(external);
           continue;
         }
 
-        for (auto *CMD :
-             getFineGrainedDepGraph(forRanges)
-                 .findJobsToRecompileWhenNodesChange(subChanges.getValue())) {
+        // Prune away the detritus from the build record.
+        auto &changes = maybeChanges.getValue();
+        pruneChangeSetFromExternalDependency(changes);
+
+        for (auto *CMD : getFineGrainedDepGraph(forRanges)
+                             .findJobsToRecompileWhenNodesChange(changes)) {
           if (CMD == externalJob) {
             continue;
           }
@@ -1861,6 +1900,76 @@ static void writeCompilationRecord(StringRef path, StringRef argsHash,
   }
 }
 
+using SourceFileDepGraph = swift::fine_grained_dependencies::SourceFileDepGraph;
+
+/// Render out the unified module dependency graph to the given \p path, which
+/// is expected to be a path relative to the build record.
+static void withPriorDependencyGraph(StringRef path,
+                                     const Compilation::Result &result,
+                                     llvm::function_ref<void(SourceFileDepGraph &&)> cont) {
+  // Building a source file dependency graph from the module dependency graph
+  // is a strange task on its face because a source file dependency graph is
+  // usually built for exactly one file. However, the driver is going to use
+  // some encoding tricks to get the dependencies for each incremental external
+  // dependency into one big file. Note that these tricks
+  // are undone in \c pruneChangeSetFromExternalDependency, so if you modify
+  // this you need to go fix that algorithm up as well. This is a diagrammatic
+  // view of the structure of the dependencies this function builds:
+  //
+  // SourceFile => interface <BUILD_RECORD>.external
+  // | - Incremetal External Dependency => interface <MODULE_1>.swiftmodule
+  // | | - <dependency> ...
+  // | | - <dependency> ...
+  // | | - <dependency> ...
+  // | - Incremetal External Dependency => interface <MODULE_2>.swiftmodule
+  // | | - <dependency> ...
+  // | | - <dependency> ...
+  // | - Incremetal External Dependency => interface <MODULE_3>.swiftmodule
+  // | - ...
+  //
+  // Where each <dependency> node has an arc back to its parent swiftmodule.
+  // That swiftmodule, in turn, takes the form of as an incremental external
+  // dependency. This formulation allows us to easily discern the original
+  // swiftmodule that a <dependency> came from just by examining that arc. This
+  // is used in integrate to "move" the <dependency> from the build record to
+  // the swiftmodule by swapping the key it uses.
+  using namespace swift::fine_grained_dependencies;
+  SourceFileDepGraph g;
+  const auto &resultModuleGraph = result.depGraph;
+  // Create the key for the entire external build record.
+  auto fileKey =
+      DependencyKey::createKeyForWholeSourceFile(DeclAspect::interface, path);
+  auto fileNodePair = g.findExistingNodePairOrCreateAndAddIfNew(fileKey, None);
+  for (StringRef incrExternalDep :
+       resultModuleGraph.getIncrementalExternalDependencies()) {
+    // Now make a node for each incremental external dependency.
+    auto interfaceKey =
+        DependencyKey(NodeKind::incrementalExternalDepend,
+                      DeclAspect::interface, "", incrExternalDep.str());
+    auto ifaceNode = g.findExistingNodeOrCreateIfNew(interfaceKey, None,
+                                                     false /* = !isProvides */);
+    resultModuleGraph.forEachNodeInJob(incrExternalDep, [&](const auto *node) {
+      // Reject
+      // 1) Implementation nodes: We don't care about the interface nodes
+      //    for cross-module dependencies because the client cannot observe it
+      //    by definition.
+      // 2) Source file nodes: we're about to define our own.
+      if (!node->getKey().isInterface() ||
+          node->getKey().getKind() == NodeKind::sourceFileProvide) {
+        return;
+      }
+      assert(node->getIsProvides() &&
+             "Found a node in module depdendencies that is not a provides!");
+      auto *newNode = new SourceFileDepGraphNode(
+          node->getKey(), node->getFingerprint(), /*isProvides*/ true);
+      g.addNode(newNode);
+      g.addArc(ifaceNode, newNode);
+    });
+    g.addArc(fileNodePair.getInterface(), ifaceNode);
+  }
+  return cont(std::move(g));
+}
+
 static void writeInputJobsToFilelist(llvm::raw_fd_ostream &out, const Job *job,
                                      const file_types::ID infoType) {
   // FIXME: Duplicated from ToolChains.cpp.
@@ -1960,6 +2069,15 @@ Compilation::performJobsImpl(std::unique_ptr<TaskQueue> &&TQ) {
     auto result = std::move(State).takeResult();
     writeCompilationRecord(CompilationRecordPath, ArgsHash, BuildStartTime,
                            InputInfo);
+    if (EnableCrossModuleIncrementalBuild) {
+      // Write out our priors adjacent to the build record so we can pick
+      // the up in a subsequent build.
+      withPriorDependencyGraph(getExternalSwiftDepsFilePath(), result,
+                               [&](SourceFileDepGraph &&g) {
+        writeFineGrainedDependencyGraphToPath(
+            Diags, getExternalSwiftDepsFilePath(), g);
+      });
+    }
     return result;
   } else {
     return std::move(State).takeResult();

lib/Driver/Driver.cpp

@@ -2039,6 +2039,7 @@ void Driver::buildActions(SmallVectorImpl<const Action *> &TopLevelActions,
       case file_types::TY_ObjCHeader:
       case file_types::TY_ClangModuleFile:
       case file_types::TY_SwiftDeps:
+      case file_types::TY_ExternalSwiftDeps:
       case file_types::TY_SwiftRanges:
       case file_types::TY_CompiledSource:
       case file_types::TY_Remapping:

lib/Driver/FineGrainedDependencyDriverGraph.cpp

@@ -287,15 +287,29 @@ ModuleDepGraph::Changes ModuleDepGraph::integrate(const SourceFileDepGraph &g,
 
   g.forEachNode([&](const SourceFileDepGraphNode *integrand) {
     const auto &key = integrand->getKey();
-
-    auto preexistingMatch = findPreexistingMatch(swiftDepsOfJob, integrand);
+    StringRef realSwiftDepsPath = swiftDepsOfJob;
+    // If we're doing a cross-module incremental build, we'll see these
+    // `.external` "swiftdeps" files. See \c writePriorDependencyGraph for
+    // the structure of the graph we're traversing here. Essentially, follow
+    // the arc laid down there to discover the file path for the swiftmodule
+    // where this dependency node originally came from.
+    if (swiftDepsOfJob.endswith(file_types::getExtension(file_types::TY_ExternalSwiftDeps)) &&
+        integrand->getKey().getKind() != NodeKind::sourceFileProvide) {
+      integrand->forEachDefIDependUpon([&](size_t seqNum) {
+        auto &external = g.getNode(seqNum)->getKey();
+        if (external.getKind() == NodeKind::incrementalExternalDepend) {
+          realSwiftDepsPath = external.getName();
+        }
+      });
+    }
+    auto preexistingMatch = findPreexistingMatch(realSwiftDepsPath, integrand);
     if (preexistingMatch.hasValue() &&
         preexistingMatch.getValue().first == LocationOfPreexistingNode::here)
       disappearedNodes.erase(key); // Node was and still is. Do not erase it.
 
     Optional<NullablePtr<ModuleDepGraphNode>> newNodeOrChangedNode =
         integrateSourceFileDepGraphNode(g, integrand, preexistingMatch,
-                                        swiftDepsOfJob);
+                                        realSwiftDepsPath);
 
     if (!newNodeOrChangedNode)
       changedNodes = None;

lib/Driver/ToolChains.cpp

@@ -619,6 +619,7 @@ const char *ToolChain::JobContext::computeFrontendModeForCompile() const {
   case file_types::TY_ObjCHeader:
   case file_types::TY_Image:
   case file_types::TY_SwiftDeps:
+  case file_types::TY_ExternalSwiftDeps:
   case file_types::TY_SwiftRanges:
   case file_types::TY_CompiledSource:
   case file_types::TY_ModuleTrace:
@@ -881,6 +882,7 @@ ToolChain::constructInvocation(const BackendJobAction &job,
     case file_types::TY_ObjCHeader:
     case file_types::TY_Image:
     case file_types::TY_SwiftDeps:
+    case file_types::TY_ExternalSwiftDeps:
     case file_types::TY_SwiftRanges:
     case file_types::TY_CompiledSource:
     case file_types::TY_Remapping: