Track non-image Directory sequence in Reader

A bit nasty, since we are storing the current Directory within the Image
instead of a separate attribute so we can not have one without the other
directly. Instead we prioritize with another attribute.

src/decoder/image.rs

@@ -86,7 +86,19 @@ impl Image {
         decoder: &mut ValueReader<R>,
         ifd: Directory,
     ) -> TiffResult<Image> {
-        let mut tag_reader = TagReader { decoder, ifd: &ifd };
+        let img = Self::from_ref(decoder, &ifd)?;
+
+        Ok(Image {
+            ifd: Some(ifd),
+            ..img
+        })
+    }
+
+    pub(crate) fn from_ref<R: Read + Seek>(
+        decoder: &mut ValueReader<R>,
+        ifd: &Directory,
+    ) -> TiffResult<Image> {
+        let mut tag_reader = TagReader { decoder, ifd };
 
         let width = tag_reader.require_tag(Tag::ImageWidth)?.into_u32()?;
         let height = tag_reader.require_tag(Tag::ImageLength)?.into_u32()?;
@@ -288,7 +300,7 @@ impl Image {
         };
 
         Ok(Image {
-            ifd: Some(ifd),
+            ifd: None,
             width,
             height,
             bits_per_sample: bits_per_sample[0],

src/decoder/mod.rs

@@ -23,7 +23,7 @@ mod stream;
 mod tag_reader;
 
 /// Result of a decoding process
-#[derive(Debug)]
+#[derive(Debug, PartialEq)]
 pub enum DecodingResult {
     /// A vector of unsigned bytes
     U8(Vec<u8>),
@@ -349,6 +349,10 @@ where
     ifd_offsets: Vec<IfdPointer>,
     /// Map from the ifd into the `ifd_offsets` ordered list.
     seen_ifds: cycles::IfdCycles,
+    /// The directory, if we have not yet read it as an image.
+    /// This is prioritized _over_ the image. Hence it must _not_ be set if we are currently
+    /// reading a true image IFD that is instead stored in the `image` attribute.
+    non_image_ifd: Option<Directory>,
     image: Image,
 }
 
@@ -624,6 +628,7 @@ impl<R: Read + Seek> Decoder<R> {
                 chunk_offsets: Vec::new(),
                 chunk_bytes: Vec::new(),
             },
+            non_image_ifd: None,
         };
         decoder.next_image()?;
         Ok(decoder)
@@ -653,48 +658,85 @@ impl<R: Read + Seek> Decoder<R> {
 
     /// Loads the IFD at the specified index in the list, if one exists
     pub fn seek_to_image(&mut self, ifd_index: usize) -> TiffResult<()> {
-        // Check whether we have seen this IFD before, if so then the index will be less than the length of the list of ifd offsets
-        if ifd_index >= self.ifd_offsets.len() {
-            // We possibly need to load in the next IFD
-            if self.next_ifd.is_none() {
-                self.current_ifd = None;
+        let ifd = self.seek_to_directory(ifd_index)?;
+        self.non_image_ifd = None;
+        self.image = Image::from_reader(&mut self.value_reader, ifd)?;
+        Ok(())
+    }
 
-                return Err(TiffError::FormatError(
-                    TiffFormatError::ImageFileDirectoryNotFound,
-                ));
-            }
+    fn seek_to_directory(&mut self, ifd_index: usize) -> TiffResult<Directory> {
+        if ifd_index < self.ifd_offsets.len() {
+            // If the index is within the list of ifds then we can load the selected image/IFD
+            let ifd_offset = self.ifd_offsets[ifd_index];
+            let ifd = self.value_reader.read_directory(ifd_offset)?;
 
-            loop {
-                // Follow the list until we find the one we want, or we reach the end, whichever happens first
-                let ifd = self.next_ifd()?;
-
-                if ifd.next().is_none() {
-                    break;
-                }
-
-                if ifd_index < self.ifd_offsets.len() {
-                    break;
-                }
-            }
-        }
-
-        // If the index is within the list of ifds then we can load the selected image/IFD
-        if let Some(ifd_offset) = self.ifd_offsets.get(ifd_index) {
-            let ifd = self.value_reader.read_directory(*ifd_offset)?;
             self.next_ifd = ifd.next();
-            self.current_ifd = Some(*ifd_offset);
-            self.image = Image::from_reader(&mut self.value_reader, ifd)?;
+            self.current_ifd = Some(ifd_offset);
 
-            Ok(())
-        } else {
-            Err(TiffError::FormatError(
-                TiffFormatError::ImageFileDirectoryNotFound,
-            ))
+            return Ok(ifd);
         }
+
+        // Follow the list until we find the one we want, or we reach the end, whichever happens
+        // first. How many IFDs to read only for their `next` field?
+        let step_over = self.ifd_offsets.len() - ifd_index;
+
+        for _ in 0..step_over {
+            // FIXME: for optimization we only need to read the offset of this one, not its whole
+            // data. However on buffered files this should be a rather small difference unless
+            // you're traversing a lot of directories.
+
+            // Detecting an end-of-file is done by `next_ifd`. We ignore the IFD itself (but not
+            // via an ignore pattern to avoid silencing must_use accidentally).
+            self.next_ifd()?;
+        }
+
+        // self.next_ifd, self.current_ifd will be setup by `next_ifd`.
+        self.next_ifd()
+    }
+
+    /// Start the chain of image directories from a new location.
+    ///
+    /// This enters a new chain of image file directories with the indicated offset as the new root
+    /// (similar to the file having its initial offset at the given IFD position). After this call,
+    /// [`Self::seek_to_image`] has the new root at index `0` and works relative to the new root.
+    ///
+    /// This is not atomic with regards to errors. If the function returns an error, the decoder is
+    /// left in an intermediate state where it does not point to any image. A valid state can be
+    /// recovered by calling [`Self::restart_at_image`] with a valid offset and a successful seek.
+    pub fn restart_at_image(&mut self, offset: IfdPointer) -> TiffResult<()> {
+        let ifd = self.restart_at_offset(offset)?;
+        self.non_image_ifd = None;
+        self.image = Image::from_reader(&mut self.value_reader, ifd)?;
+        Ok(())
+    }
+
+    /// Start the chain of non-image directories from a new location.
+    ///
+    /// See [`Self::restart_at_image`] for details, except this method does not attempt to
+    /// interpret the directory in the sequence as an image. Instead, it may be used to read a
+    /// sequence of auxiliary IFDs that are not necessarily images. For instance, a directory
+    /// referred to in the SubIfd tag may be a thumbnail
+    pub fn restart_at_directory(&mut self, offset: IfdPointer) -> TiffResult<()> {
+        let ifd = self.restart_at_offset(offset)?;
+        self.non_image_ifd = Some(ifd);
+        Ok(())
+    }
+
+    fn restart_at_offset(&mut self, offset: IfdPointer) -> TiffResult<Directory> {
+        self.ifd_offsets.clear();
+        self.ifd_offsets.push(offset);
+
+        self.next_ifd = Some(offset);
+        self.current_ifd = None;
+
+        self.next_ifd()
     }
 
     fn next_ifd(&mut self) -> TiffResult<Directory> {
         let Some(next_ifd) = self.next_ifd.take() else {
+            self.current_ifd = None;
+            self.non_image_ifd = None;
+
             return Err(TiffError::FormatError(
                 TiffFormatError::ImageFileDirectoryNotFound,
             ));
@@ -721,10 +763,44 @@ impl<R: Read + Seek> Decoder<R> {
     /// To determine whether there are more images call `TIFFDecoder::more_images` instead.
     pub fn next_image(&mut self) -> TiffResult<()> {
         let ifd = self.next_ifd()?;
+        self.non_image_ifd = None;
         self.image = Image::from_reader(&mut self.value_reader, ifd)?;
         Ok(())
     }
 
+    /// Read the next directory without interpreting it as an image.
+    ///
+    /// If there is no further image in the TIFF file a format error is returned. To determine
+    /// whether there are more images call `TIFFDecoder::more_directories` instead.
+    pub fn next_directory(&mut self) -> TiffResult<()> {
+        let ifd = self.next_ifd()?;
+        self.non_image_ifd = Some(ifd);
+        Ok(())
+    }
+
+    /// Interpret the current directory as an image.
+    ///
+    /// This method is used after having called [`Self::restart_at`] or [`Self::next_directory`] to
+    /// iterate the sequence of image file directories, having read a directory without having read
+    /// its tags as image data.
+    pub fn current_directory_as_image(&mut self) -> TiffResult<()> {
+        let current_ifd = self.current_ifd.ok_or(TiffError::FormatError(
+            TiffFormatError::ImageFileDirectoryNotFound,
+        ))?;
+
+        if let Some(ifd) = &self.non_image_ifd {
+            self.image = Image::from_ref(&mut self.value_reader, ifd)?;
+            // Definitely sets an IFD but this works without unwraps.
+            self.image.ifd = self.non_image_ifd.take();
+        } else {
+            // Probably re-reading an image but that's fine.
+            let ifd = self.read_directory(current_ifd)?;
+            self.image = Image::from_reader(&mut self.value_reader, ifd)?;
+        }
+
+        Ok(())
+    }
+
     /// Returns `true` if there is at least one more image available.
     pub fn more_images(&self) -> bool {
         self.next_ifd.is_some()
@@ -1227,11 +1303,22 @@ impl<R: Read + Seek> Decoder<R> {
     }
 
     /// Get the IFD decoder for our current image IFD.
-    fn image_ifd(&mut self) -> IfdDecoder<'_> {
+    fn current_directory_ifd(&mut self) -> IfdDecoder<'_> {
+        // Special fallback. We do not want to error handle not having read a current directory, in
+        // particular as the behavior having a directory without tags will produce these errors
+        // anyways (most likely). Note that an empty directory is invalid in a TIFF.
+        static NO_IFD: Directory = Directory::empty();
+
+        let ifd = self
+            .non_image_ifd
+            .as_ref()
+            .or(self.image.ifd.as_ref())
+            .unwrap_or(&NO_IFD);
+
         IfdDecoder {
             inner: tag_reader::TagReader {
                 decoder: &mut self.value_reader,
-                ifd: self.image.ifd.as_ref().unwrap(),
+                ifd,
             },
         }
     }
@@ -1272,25 +1359,25 @@ impl<R: Read + Seek> Decoder<R> {
     /// Tries to retrieve a tag from the current image directory.
     /// Return `Ok(None)` if the tag is not present.
     pub fn find_tag(&mut self, tag: Tag) -> TiffResult<Option<ifd::Value>> {
-        self.image_ifd().find_tag(tag)
+        self.current_directory_ifd().find_tag(tag)
     }
 
     /// Tries to retrieve a tag in the current image directory and convert it to the desired
     /// unsigned type.
     pub fn find_tag_unsigned<T: TryFrom<u64>>(&mut self, tag: Tag) -> TiffResult<Option<T>> {
-        self.image_ifd().find_tag_unsigned(tag)
+        self.current_directory_ifd().find_tag_unsigned(tag)
     }
 
     /// Tries to retrieve a tag from the current image directory and convert it to the desired
     /// unsigned type. Returns an error if the tag is not present.
     pub fn get_tag_unsigned<T: TryFrom<u64>>(&mut self, tag: Tag) -> TiffResult<T> {
-        self.image_ifd().get_tag_unsigned(tag)
+        self.current_directory_ifd().get_tag_unsigned(tag)
     }
 
     /// Tries to retrieve a tag from the current image directory.
     /// Returns an error if the tag is not present
     pub fn get_tag(&mut self, tag: Tag) -> TiffResult<ifd::Value> {
-        self.image_ifd().get_tag(tag)
+        self.current_directory_ifd().get_tag(tag)
     }
 
     pub fn get_tag_u32(&mut self, tag: Tag) -> TiffResult<u32> {
@@ -1345,7 +1432,7 @@ impl<R: Read + Seek> Decoder<R> {
     }
 
     pub fn tag_iter(&mut self) -> impl Iterator<Item = TiffResult<(Tag, ifd::Value)>> + '_ {
-        self.image_ifd().tag_iter()
+        self.current_directory_ifd().tag_iter()
     }
 }
 

src/directory.rs

@@ -29,7 +29,7 @@ pub struct Directory {
 impl Directory {
     /// Create a directory in an initial state without entries. Note that an empty directory can
     /// not be encoded in a file, it must contain at least one entry.
-    pub fn empty() -> Self {
+    pub const fn empty() -> Self {
         Directory {
             entries: BTreeMap::new(),
             next_ifd: None,

tests/subsubifds.rs

@@ -1,4 +1,7 @@
-use tiff::decoder::Decoder;
+use tiff::{
+    decoder::Decoder,
+    tags::{IfdPointer, Tag},
+};
 
 // This file has the following IFD structure:
 //
@@ -39,3 +42,128 @@ fn decode_seek_chain() {
     assert_eq!(Some(offset2), decoder.ifd_pointer());
     assert!(!decoder.more_images());
 }
+
+#[test]
+fn decode_seek_recover() {
+    let file = File::open(TEST_IMAGE_SUBIFD).expect("Cannot open test image");
+    let mut decoder = Decoder::new(file).expect("Invalid format to create decoder");
+
+    // Remember the offsets in the first iteration.
+    let offset0 = decoder.ifd_pointer().expect("First IFD pointer not found");
+    decoder.next_image().unwrap();
+    let offset1 = decoder.ifd_pointer().expect("Second IFD pointer not found");
+    decoder.next_image().unwrap();
+    let offset2 = decoder.ifd_pointer().expect("Third IFD pointer not found");
+
+    // oops!
+    let fails = decoder.restart_at_image(IfdPointer(0xdead_beef));
+    assert!(fails.is_err());
+
+    // However we can recover by restarting our seek.
+    decoder
+        .restart_at_image(offset0)
+        .expect("Failed to restart IFDs from first image");
+    assert_eq!(Some(offset0), decoder.ifd_pointer());
+    decoder.next_image().unwrap();
+    assert_eq!(Some(offset1), decoder.ifd_pointer());
+    decoder.next_image().unwrap();
+    assert_eq!(Some(offset2), decoder.ifd_pointer());
+}
+
+#[test]
+fn decode_seek_directory() {
+    let file = File::open(TEST_IMAGE_SUBIFD).expect("Cannot open test image");
+    let mut decoder = Decoder::new(file).expect("Invalid format to create decoder");
+
+    let offset = decoder.ifd_pointer().expect("First IFD pointer not found");
+    let img0 = decoder.read_image().unwrap();
+
+    decoder.next_image().unwrap();
+    let img1 = decoder.read_image().unwrap();
+    assert_ne!(img0, img1);
+
+    {
+        // Due to the special test file structure.
+        let subifd = decoder
+            .get_tag(Tag::SubIfd)
+            .unwrap()
+            .into_ifd_vec()
+            .unwrap()[0];
+
+        // Let's try to manually interpret this as an image.
+        decoder
+            .restart_at_directory(subifd)
+            .expect("Failed to restart IFDs at SubIfd");
+        decoder
+            .current_directory_as_image()
+            .expect("Failed to read SubIfd image");
+
+        assert_eq!(decoder.dimensions().unwrap(), (32, 32));
+        let subimage = decoder.read_image().unwrap();
+        // Verify we did not accidentally read another image.
+        assert_ne!(img0, subimage);
+        assert_ne!(img1, subimage);
+    }
+
+    // And finally, seek back to the first image.
+    decoder
+        .restart_at_image(offset)
+        .expect("First IFD no longer readable");
+    assert_eq!(Some(offset), decoder.ifd_pointer());
+
+    let img0_try2 = decoder.read_image().unwrap();
+    assert_eq!(img0, img0_try2);
+}
+
+#[test]
+fn decode_subifds() {
+    let explore_chain = |decoder: &mut Decoder<_>, subifds: &mut Vec<_>| {
+        let mut len = 0usize;
+
+        loop {
+            len += 1;
+
+            if let Ok(subifd) = decoder.get_tag(Tag::SubIfd) {
+                let prelen = subifds.len();
+                subifds.extend(subifd.into_ifd_vec().expect("Failed to decode SubIfd tag"));
+                eprintln!("New roots: {:?}", &subifds[prelen..]);
+            } else {
+                eprintln!("No roots");
+            }
+
+            if !decoder.more_images() {
+                break;
+            }
+
+            decoder.next_image().unwrap();
+        }
+
+        len
+    };
+
+    let file = File::open(TEST_IMAGE_SUBIFD).expect("Cannot open test image");
+    let mut decoder = Decoder::new(file).expect("Invalid format to create decoder");
+
+    let mut subifds = vec![];
+    let mut chain_lengths = vec![];
+    chain_lengths.push(explore_chain(&mut decoder, &mut subifds));
+
+    while !subifds.is_empty() {
+        let root = subifds.remove(0);
+        let manual_directory = decoder.read_directory(root).unwrap();
+        eprintln!("Manual directory: {:?}", manual_directory);
+
+        decoder.restart_at_directory(root).unwrap();
+        let prelen = subifds.len();
+        chain_lengths.push(explore_chain(&mut decoder, &mut subifds));
+        eprintln!("Chains roots: {:?}", &subifds[prelen..]);
+    }
+
+    // Traversal Order in this graph:
+    // Image 0 -> Image 1 -> Image 2
+    //            |-> Image 3
+    //            |-> Image 4 -> Image 5
+    //            |              \-> Image 7 -> Image 8
+    //            \-> Image 6
+    assert_eq!(chain_lengths, [3, 1, 2, 1, 2]);
+}