about summary refs log tree commit diff
path: root/tvix/nar-bridge/pkg/exporter/export.go
diff options
context:
space:
mode:
authorFlorian Klink <flokli@flokli.de>2023-10-03T11·29+0300
committerflokli <flokli@flokli.de>2023-10-05T06·17+0000
commitd3c44ab03d14bc143e790426bede465feb871cd5 (patch)
tree0c8abf1595ec6f5aa127639f60c8f4a7a808ca01 /tvix/nar-bridge/pkg/exporter/export.go
parent49b427d7736b11ba05892126f63557146a642a67 (diff)
refactor(tvix/nar-bridge): rename writer package to exporter r/6696
Change-Id: I7cf6a3ed0dca4c450fadd9228dc0154ad6254f51
Reviewed-on: https://cl.tvl.fyi/c/depot/+/9530
Tested-by: BuildkiteCI
Reviewed-by: Connor Brewster <cbrewster@hey.com>
Autosubmit: flokli <flokli@flokli.de>
Diffstat (limited to 'tvix/nar-bridge/pkg/exporter/export.go')
-rw-r--r--tvix/nar-bridge/pkg/exporter/export.go279
1 files changed, 279 insertions, 0 deletions
diff --git a/tvix/nar-bridge/pkg/exporter/export.go b/tvix/nar-bridge/pkg/exporter/export.go
new file mode 100644
index 000000000000..40ccc4b6380c
--- /dev/null
+++ b/tvix/nar-bridge/pkg/exporter/export.go
@@ -0,0 +1,279 @@
+package exporter
+
+import (
+	"fmt"
+	"io"
+	"path"
+
+	castorev1pb "code.tvl.fyi/tvix/castore/protos"
+	storev1pb "code.tvl.fyi/tvix/store/protos"
+	"github.com/nix-community/go-nix/pkg/nar"
+)
+
+type DirectoryLookupFn func([]byte) (*castorev1pb.Directory, error)
+type BlobLookupFn func([]byte) (io.ReadCloser, error)
+
+// Export will traverse a given PathInfo structure, and write the contents
+// in NAR format to the passed Writer.
+// It uses directoryLookupFn and blobLookupFn to resolve references.
+func Export(
+	w io.Writer,
+	pathInfo *storev1pb.PathInfo,
+	directoryLookupFn DirectoryLookupFn,
+	blobLookupFn BlobLookupFn,
+) error {
+	// initialize a NAR writer
+	narWriter, err := nar.NewWriter(w)
+	if err != nil {
+		return fmt.Errorf("unable to initialize nar writer: %w", err)
+	}
+	defer narWriter.Close()
+
+	// populate rootHeader
+	rootHeader := &nar.Header{
+		Path: "/",
+	}
+
+	// populate a stack
+	// we will push paths and directories to it when entering a directory,
+	// and emit individual elements to the NAR writer, draining the Directory object.
+	// once it's empty, we can pop it off the stack.
+	var stackPaths = []string{}
+	var stackDirectories = []*castorev1pb.Directory{}
+
+	// peek at the pathInfo root and assemble the root node and write to writer
+	// in the case of a regular file, we retrieve and write the contents, close and exit
+	// in the case of a symlink, we write the symlink, close and exit
+	switch v := (pathInfo.GetNode().GetNode()).(type) {
+	case *castorev1pb.Node_File:
+		rootHeader.Type = nar.TypeRegular
+		rootHeader.Size = int64(v.File.GetSize())
+		rootHeader.Executable = v.File.GetExecutable()
+		err := narWriter.WriteHeader(rootHeader)
+		if err != nil {
+			return fmt.Errorf("unable to write root header: %w", err)
+		}
+
+		// if it's a regular file, retrieve and write the contents
+		blobReader, err := blobLookupFn(v.File.GetDigest())
+		if err != nil {
+			return fmt.Errorf("unable to lookup blob: %w", err)
+		}
+		defer blobReader.Close()
+
+		_, err = io.Copy(narWriter, blobReader)
+		if err != nil {
+			return fmt.Errorf("unable to read from blobReader: %w", err)
+		}
+
+		err = blobReader.Close()
+		if err != nil {
+			return fmt.Errorf("unable to close content reader: %w", err)
+		}
+
+		err = narWriter.Close()
+		if err != nil {
+			return fmt.Errorf("unable to close nar reader: %w", err)
+		}
+
+		return nil
+
+	case *castorev1pb.Node_Symlink:
+		rootHeader.Type = nar.TypeSymlink
+		rootHeader.LinkTarget = string(v.Symlink.GetTarget())
+		err := narWriter.WriteHeader(rootHeader)
+		if err != nil {
+			return fmt.Errorf("unable to write root header: %w", err)
+		}
+
+		err = narWriter.Close()
+		if err != nil {
+			return fmt.Errorf("unable to close nar reader: %w", err)
+		}
+
+		return nil
+	case *castorev1pb.Node_Directory:
+		// We have a directory at the root, look it up and put in on the stack.
+		directory, err := directoryLookupFn(v.Directory.Digest)
+		if err != nil {
+			return fmt.Errorf("unable to lookup directory: %w", err)
+		}
+		stackDirectories = append(stackDirectories, directory)
+		stackPaths = append(stackPaths, "/")
+
+		err = narWriter.WriteHeader(&nar.Header{
+			Path: "/",
+			Type: nar.TypeDirectory,
+		})
+
+		if err != nil {
+			return fmt.Errorf("error writing header: %w", err)
+		}
+	}
+
+	// as long as the stack is not empty, we keep running.
+	for {
+		if len(stackDirectories) == 0 {
+			return nil
+		}
+
+		// Peek at the current top of the stack.
+		topOfStack := stackDirectories[len(stackDirectories)-1]
+		topOfStackPath := stackPaths[len(stackPaths)-1]
+
+		// get the next element that's lexicographically smallest, and drain it from
+		// the current directory on top of the stack.
+		nextNode := drainNextNode(topOfStack)
+
+		// If nextNode returns nil, there's nothing left in the directory node, so we
+		// can emit it from the stack.
+		// Contrary to the import case, we don't emit the node popping from the stack, but when pushing.
+		if nextNode == nil {
+			// pop off stack
+			stackDirectories = stackDirectories[:len(stackDirectories)-1]
+			stackPaths = stackPaths[:len(stackPaths)-1]
+
+			continue
+		}
+
+		switch n := (nextNode).(type) {
+		case *castorev1pb.DirectoryNode:
+			err := narWriter.WriteHeader(&nar.Header{
+				Path: path.Join(topOfStackPath, string(n.GetName())),
+				Type: nar.TypeDirectory,
+			})
+			if err != nil {
+				return fmt.Errorf("unable to write nar header: %w", err)
+			}
+
+			d, err := directoryLookupFn(n.GetDigest())
+			if err != nil {
+				return fmt.Errorf("unable to lookup directory: %w", err)
+			}
+
+			// add to stack
+			stackDirectories = append(stackDirectories, d)
+			stackPaths = append(stackPaths, path.Join(topOfStackPath, string(n.GetName())))
+		case *castorev1pb.FileNode:
+			err := narWriter.WriteHeader(&nar.Header{
+				Path:       path.Join(topOfStackPath, string(n.GetName())),
+				Type:       nar.TypeRegular,
+				Size:       int64(n.GetSize()),
+				Executable: n.GetExecutable(),
+			})
+			if err != nil {
+				return fmt.Errorf("unable to write nar header: %w", err)
+			}
+
+			// copy file contents
+			contentReader, err := blobLookupFn(n.GetDigest())
+			if err != nil {
+				return fmt.Errorf("unable to get blob: %w", err)
+			}
+			defer contentReader.Close()
+
+			_, err = io.Copy(narWriter, contentReader)
+			if err != nil {
+				return fmt.Errorf("unable to copy contents from contentReader: %w", err)
+			}
+
+			err = contentReader.Close()
+			if err != nil {
+				return fmt.Errorf("unable to close content reader: %w", err)
+			}
+		case *castorev1pb.SymlinkNode:
+			err := narWriter.WriteHeader(&nar.Header{
+				Path:       path.Join(topOfStackPath, string(n.GetName())),
+				Type:       nar.TypeSymlink,
+				LinkTarget: string(n.GetTarget()),
+			})
+			if err != nil {
+				return fmt.Errorf("unable to write nar header: %w", err)
+			}
+		}
+	}
+}
+
+// TODO: add validation functions to Directory in both rust and golang, to
+// validate the keys in directories, files and symlinks are sorted.
+
+// drainNextNode will drain a directory message with one of its child nodes,
+// whichever comes first alphabetically.
+func drainNextNode(d *castorev1pb.Directory) interface{} {
+	switch v := (smallestNode(d)).(type) {
+	case *castorev1pb.DirectoryNode:
+		d.Directories = d.Directories[1:]
+		return v
+	case *castorev1pb.FileNode:
+		d.Files = d.Files[1:]
+		return v
+	case *castorev1pb.SymlinkNode:
+		d.Symlinks = d.Symlinks[1:]
+		return v
+	case nil:
+		return nil
+	default:
+		panic("invalid type encountered")
+	}
+}
+
+// smallestNode will return the node from a directory message,
+// whichever comes first alphabetically.
+func smallestNode(d *castorev1pb.Directory) interface{} {
+	childDirectories := d.GetDirectories()
+	childFiles := d.GetFiles()
+	childSymlinks := d.GetSymlinks()
+
+	if len(childDirectories) > 0 {
+		if len(childFiles) > 0 {
+			if len(childSymlinks) > 0 {
+				// directories,files,symlinks
+				return smallerNode(smallerNode(childDirectories[0], childFiles[0]), childSymlinks[0])
+			} else {
+				// directories,files,!symlinks
+				return smallerNode(childDirectories[0], childFiles[0])
+			}
+		} else {
+			// directories,!files
+			if len(childSymlinks) > 0 {
+				// directories,!files,symlinks
+				return smallerNode(childDirectories[0], childSymlinks[0])
+			} else {
+				// directories,!files,!symlinks
+				return childDirectories[0]
+			}
+		}
+	} else {
+		// !directories
+		if len(childFiles) > 0 {
+			// !directories,files
+			if len(childSymlinks) > 0 {
+				// !directories,files,symlinks
+				return smallerNode(childFiles[0], childSymlinks[0])
+			} else {
+				// !directories,files,!symlinks
+				return childFiles[0]
+			}
+		} else {
+			//!directories,!files
+			if len(childSymlinks) > 0 {
+				//!directories,!files,symlinks
+				return childSymlinks[0]
+			} else {
+				//!directories,!files,!symlinks
+				return nil
+			}
+		}
+	}
+}
+
+// smallerNode compares two nodes by their name,
+// and returns the one with the smaller name.
+// both nodes may not be nil, we do check for these cases in smallestNode.
+func smallerNode(a interface{ GetName() []byte }, b interface{ GetName() []byte }) interface{ GetName() []byte } {
+	if string(a.GetName()) < string(b.GetName()) {
+		return a
+	} else {
+		return b
+	}
+}