Tar extraction should set tar-level timestamps

[lissyx]

What would you like to have changed?

When extracting a tar (or tar composition, i.e., tar.xz, tar.gz, tar.bz2, etc), I would like the extracted files to have the proper timestamps (atime, mtime) that were recorded in the file

Why is this feature a useful, necessary, and/or important addition to this project?

This is the behavior of GNU tar, and lacking this feature breaks some usecases

What alternatives are there, or what are you doing in the meantime to work around the lack of this feature?

I can't use a tar of bazel's cache via generic-worker project (https://github.com/taskcluster/taskcluster/blob/222e8057550e7560412920446e76076d3db8e496/workers/generic-worker/mounts.go#L670-L679): newer timestamps breaks sanity checks from Bazel.

[lissyx]

I tried to make something that works pretty well in my case, but I'm really not at ease with Go to do more:

diff --git a/tar.go b/tar.go
index 6dda50a..d8ef9b0 100644
--- a/tar.go
+++ b/tar.go
@@ -230,16 +230,27 @@ func (t *Tar) untarFile(f File, destination string, hdr *tar.Header) error {
        case tar.TypeDir:
                return mkdir(to, f.Mode())
        case tar.TypeReg, tar.TypeRegA, tar.TypeChar, tar.TypeBlock, tar.TypeFifo, tar.TypeGNUSparse:
-               return writeNewFile(to, f, f.Mode())
+               err := writeNewFile(to, f, f.Mode())
+               if err == nil {
+                       return os.Chtimes(to, hdr.ModTime, hdr.ModTime)
+               }
        case tar.TypeSymlink:
-               return writeNewSymbolicLink(to, hdr.Linkname)
+               err := writeNewSymbolicLink(to, hdr.Linkname)
+               if err == nil {
+                       return os.Chtimes(to, hdr.ModTime, hdr.ModTime)
+               }
        case tar.TypeLink:
-               return writeNewHardLink(to, filepath.Join(destination, hdr.Linkname))
+               err := writeNewHardLink(to, filepath.Join(destination, hdr.Linkname))
+               if err == nil {
+                       return os.Chtimes(destination, hdr.ModTime, hdr.ModTime)
+               }
        case tar.TypeXGlobalHeader:
                return nil // ignore the pax global header from git-generated tarballs
        default:
                return fmt.Errorf("%s: unknown type flag: %c", hdr.Name, hdr.Typeflag)
        }
+
+       return nil
 }
 
 func (t *Tar) writeWalk(source, topLevelFolder, destination string) error {

[mholt]

Thanks for contributing a patch!

I've completely rewritten this library with some much needed improvements in #302, which will become v4 of this package. The new core APIs no longer write to disk, so this issue becomes irrelevant, as it is currently up to the user to write streams to disk in the way they want to. Maybe in the future there will be some helper functions or CLI commands to do this, at which point we can reassess; but for now I suppose this can be closed. Feel free to continue discussion and we can reopen if needed.

[lissyx]

Thanks, unfortunately I've moved topics and i dont depend on that anymore so i cant really continue to discuss that 😊