image-codecs: read a JPEG's frame from its markers (jpeg_info)

For storytold/photocraft#12473: PhotoCraft lifts its 2^30 px JPEG cap
only after admitting the decode against free memory, and for that it
needs the frame before decoding. jpeg_info reads it from the markers
alone: the stored size, the layout decode returns, and
whole_coefficients, set when decode hands the file to craft-jpeg
(arithmetic-coded, lossless or 12-bit frames), whose working set is
several times the decoded image. None without a frame of 1, 3 or 4
components.

This was part of storytold/photocraft#12604 until
storytold/photocraft#14236 moved the codecs here.

Co-Authored-By: Νύξ (Nyx, AI) 🌑 <nyx@despotak.is>

crates/image-codecs/src/codecs/jpeg.rs

@@ -42,6 +42,18 @@ pub(crate) struct JpegMeta {
     pub frame: Option<(u32, u32, u8)>,
 }
 
+/// The first frame of `b` as [`crate::JpegInfo`], from its markers alone (see [`crate::jpeg_info`]).
+pub(crate) fn info(b: &[u8]) -> Option<crate::JpegInfo> {
+    let (width, height, components) = scan_metadata(b).frame?;
+    let layout = match components {
+        1 => ChannelLayout::Gray,
+        3 => ChannelLayout::Rgb,
+        4 => ChannelLayout::Cmyk,
+        _ => return None,
+    };
+    Some(crate::JpegInfo { width, height, layout, whole_coefficients: uses_craft_jpeg(b) })
+}
+
 /// Walk the marker segments up to SOS and collect metadata.
 pub(crate) fn scan_metadata(b: &[u8]) -> JpegMeta {
     let mut m = JpegMeta::default();
@@ -573,6 +585,21 @@ mod tests {
         bytes.get(2..11) == Some(b"\xff\xe0\0\x10JFIF\0")
     }
 
+    /// `info` reads the frame from the markers, and which decoder the file goes to.
+    #[test]
+    fn info_reads_the_frame_and_the_decoder() {
+        let img = Image::from_raw(37, 21, ChannelLayout::Rgb, SampleType::U8, vec![128; 37 * 21 * 3]).unwrap();
+        let rgb = crate::encode(&img, Format::Jpeg, &EncodeOptions::default()).unwrap();
+        assert_eq!(info(&rgb), Some(crate::JpegInfo { width: 37, height: 21, layout: ChannelLayout::Rgb, whole_coefficients: false }));
+        // A 12-bit extended frame (SOF1) declaring 40,000 × 30,000 gray, markers only.
+        let mut twelve = vec![0xFF, 0xD8, 0xFF, 0xC1, 0, 11, 12];
+        twelve.extend_from_slice(&30_000u16.to_be_bytes());
+        twelve.extend_from_slice(&40_000u16.to_be_bytes());
+        twelve.extend_from_slice(&[1, 1, 0x11, 0, 0xFF, 0xD9]);
+        assert_eq!(info(&twelve), Some(crate::JpegInfo { width: 40_000, height: 30_000, layout: ChannelLayout::Gray, whole_coefficients: true }));
+        assert_eq!(info(b"not a jpeg"), None);
+    }
+
     /// An APP1 Extended XMP fragment (GUID, total size, offset, data).
     fn extended_xmp(offset: u32) -> Vec<u8> {
         let mut seg = EXTENDED_XMP_HEADER.to_vec();

crates/image-codecs/src/lib.rs

@@ -212,6 +212,23 @@ pub fn probe_size(bytes: &[u8]) -> Result<(u32, u32), CodecError> {
     Ok(if (5..=8).contains(&o) { (h, w) } else { (w, h) })
 }
 
+/// A JPEG's frame as its header declares it, read without decoding pixels: the stored size, the
+/// layout [`decode`] returns, and whether [`decode`] hands the file to its whole-coefficient
+/// decoder (arithmetic-coded, lossless or 12-bit frames), whose working set is several times the
+/// decoded image.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct JpegInfo {
+    pub width: u32,
+    pub height: u32,
+    pub layout: ChannelLayout,
+    pub whole_coefficients: bool,
+}
+
+/// [`JpegInfo`] of `bytes`; `None` without a readable frame of 1, 3 or 4 components.
+pub fn jpeg_info(bytes: &[u8]) -> Option<JpegInfo> {
+    jpeg::info(bytes)
+}
+
 /// The colour table of a GIF's first frame (its Local Color Table, else the Global one) and the
 /// entry its Graphic Control Extension makes transparent, read without decoding pixels.
 pub fn gif_palette(bytes: &[u8]) -> Option<(Vec<[u8; 3]>, Option<u8>)> {