Add an append_link() method to handle long link targets (#273)

We should support appending long symlink targets, because this
occurs in real world filesystems.  In my case, RPM set up
`.build-id` symlinks which can get long.

Add an `append_link()` method following the precendent of
`append_path()` - we're just supporting *two* potentially
long filenames.

As a side benefit, we can just do the `std::io::empty()` dance
internally and not require the caller to specify it.

The addition of special case `append()` methods is unfortunate,
because the header API methods are then really an attractive nuisance.
We should potentially consider deprecating them.

Closes: #192

Author: cgwalters

src/builder.rs

@@ -162,6 +162,54 @@ impl<W: Write> Builder<W> {
         self.append(&header, data)
     }
 
+    /// Adds a new link (symbolic or hard) entry to this archive with the specified path and target.
+    ///
+    /// This function is similar to [`Self::append_data`] which supports long filenames,
+    /// but also supports long link targets using GNU extensions if necessary.
+    /// You must set the entry type to either [`EntryType::Link`] or [`EntryType::Symlink`].
+    /// The `set_cksum` method will be invoked after setting the path. No other metadata in the
+    /// header will be modified.
+    ///
+    /// If you are intending to use GNU extensions, you must use this method over calling
+    /// [`Header::set_link_name`] because that function will fail on long links.
+    ///
+    /// Similar constraints around the position of the archive and completion
+    /// apply as with [`Self::append_data`].
+    ///
+    /// # Errors
+    ///
+    /// This function will return an error for any intermittent I/O error which
+    /// occurs when either reading or writing.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use tar::{Builder, Header, EntryType};
+    ///
+    /// let mut ar = Builder::new(Vec::new());
+    /// let mut header = Header::new_gnu();
+    /// header.set_username("foo");
+    /// header.set_entry_type(EntryType::Symlink);
+    /// header.set_size(0);
+    /// ar.append_link(&mut header, "really/long/path/to/foo", "other/really/long/target").unwrap();
+    /// let data = ar.into_inner().unwrap();
+    /// ```
+    pub fn append_link<P: AsRef<Path>, T: AsRef<Path>>(
+        &mut self,
+        header: &mut Header,
+        path: P,
+        target: T,
+    ) -> io::Result<()> {
+        self._append_link(header, path.as_ref(), target.as_ref())
+    }
+
+    fn _append_link(&mut self, header: &mut Header, path: &Path, target: &Path) -> io::Result<()> {
+        prepare_header_path(self.get_mut(), header, path)?;
+        prepare_header_link(self.get_mut(), header, target)?;
+        header.set_cksum();
+        self.append(&header, std::io::empty())
+    }
+
     /// Adds a file on the local filesystem to this archive.
     ///
     /// This function will open the file specified by `path` and insert the file

src/header.rs

@@ -420,6 +420,8 @@ impl Header {
     /// in the appropriate format. May fail if the link name is too long or if
     /// the path specified is not Unicode and this is a Windows platform. Will
     /// strip out any "." path component, which signifies the current directory.
+    ///
+    /// To use GNU long link names, prefer instead [`crate::Builder::append_link`].
     pub fn set_link_name<P: AsRef<Path>>(&mut self, p: P) -> io::Result<()> {
         self._set_link_name(p.as_ref())
     }

tests/all.rs

@@ -914,6 +914,27 @@ fn long_linkname_trailing_nul() {
     assert_eq!(&*e.link_name_bytes().unwrap(), b"foo");
 }
 
+#[test]
+fn long_linkname_gnu() {
+    for t in [tar::EntryType::Symlink, tar::EntryType::Link] {
+        let mut b = Builder::new(Vec::<u8>::new());
+        let mut h = Header::new_gnu();
+        h.set_entry_type(t);
+        h.set_size(0);
+        let path = "usr/lib/.build-id/05/159ed904e45ff5100f7acd3d3b99fa7e27e34f";
+        let target = "../../../../usr/lib64/qt5/plugins/wayland-graphics-integration-server/libqt-wayland-compositor-xcomposite-egl.so";
+        t!(b.append_link(&mut h, path, target));
+
+        let contents = t!(b.into_inner());
+        let mut a = Archive::new(&contents[..]);
+
+        let e = &t!(t!(a.entries()).next().unwrap());
+        assert_eq!(e.header().entry_type(), t);
+        assert_eq!(e.path().unwrap().to_str().unwrap(), path);
+        assert_eq!(e.link_name().unwrap().unwrap().to_str().unwrap(), target);
+    }
+}
+
 #[test]
 fn encoded_long_name_has_trailing_nul() {
     let td = t!(TempBuilder::new().prefix("tar-rs").tempdir());