Merge the fixes to the PR below into this one

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AmsdeUCdTaD72jkDw2pJKY

crates/optimize/README.md

@@ -65,7 +65,8 @@ let report = optimize_with_progress(&mut doc, &settings, &mut |stage| keep_going
     RGB packed to one bit among them), image masks, and indexed images over `DeviceGray` or
     `DeviceRGB`, whose palette is read (Rec. 601 luma for RGB), with no decode array or one of two
     different numbers, and not drawn or listed where `/DefaultGray` or `/DefaultRGB` is or may be
-    in force. Not a separation or a DeviceN, whose tint transform isn't read, calibrated or ICC
+    in force (an indexed image referred to from beyond the page and form resources read, as an
+    annotation's appearance, may be). Not a separation or a DeviceN, whose tint transform isn't read, calibrated or ICC
     grey, or an indexed image over another base or whose two entries look alike. Not one with a soft mask, whose own size would have to
     follow; not a halftone or a dither, most of whose new pixels would cover both black and white,
     so that taking the sample covering more would turn its tones solid (a 50% screen black, a 25%

crates/optimize/src/images.rs

@@ -300,6 +300,14 @@ fn device_rgb(doc: &Document, r: ObjRef) -> bool {
         && cs.as_deref().and_then(Object::as_name).is_some_and(|n| matches!(n, b"DeviceRGB" | b"RGB"))
 }
 
+/// Whether image `r` is indexed: its palette says which of its samples is dark ([`dark_is_one`]),
+/// which default colour spaces in force where it is drawn may change.
+fn indexed(doc: &Document, r: ObjRef) -> bool {
+    let Object::Stream(s) = &*doc.get(r) else { return false };
+    let cs = s.dict.get(b"ColorSpace").map(|cs| doc.resolve(cs));
+    matches!(cs.as_deref(), Some(Object::Array(a)) if matches!(a.first().and_then(Object::as_name), Some(b"Indexed" | b"I")))
+}
+
 /// Call `f` with every reference in `o`, a dictionary's or array's included, a stream's dictionary
 /// too, but not those in objects referred to.
 fn each_ref(o: &Object, mut f: impl FnMut(ObjRef)) {
@@ -1145,11 +1153,13 @@ pub(crate) fn run(
     progress: &mut dyn FnMut(Stage) -> bool,
 ) -> Result<(), OptimizeError> {
     let (ppi, mut defaults, read) = walk_pages(doc, pages);
-    // Black-and-white RGB becomes grey, which default colour spaces would draw otherwise: one used
-    // where the walk didn't look may be drawn so.
+    // Black-and-white RGB becomes grey, which default colour spaces would draw otherwise, and an
+    // indexed image's palette says which of its samples a resampled one keeps, which they may
+    // change: one used where the walk didn't look may be drawn so.
     let packs = settings.mono.compression != MonoCompression::Retain;
-    let rgb: HashSet<u32> = ppi.keys().filter(|r| packs && device_rgb(doc, **r)).map(|r| r.num).collect();
-    defaults.extend(used_elsewhere(doc, &rgb, &read));
+    let checked: HashSet<u32> =
+        ppi.keys().filter(|r| (packs && device_rgb(doc, **r)) || (settings.mono.downsample && indexed(doc, **r))).map(|r| r.num).collect();
+    defaults.extend(used_elsewhere(doc, &checked, &read));
     // Soft masks are processed with their image, not on their own.
     let masks: HashSet<ObjRef> = ppi
         .keys()

crates/optimize/tests/optimize.rs

@@ -3573,6 +3573,41 @@ fn indexed_thin_rules_survive_averaging_for_both_palette_orders_and_decodes() {
     }
 }
 
+/// An indexed image also drawn where the optimizer doesn't read (an annotation's appearance, where
+/// other default colour spaces may be in force and change which palette entry is darker) keeps its
+/// resolution, as black-and-white RGB keeps its colour space; drawn only on the page, it is
+/// averaged. A grey image is averaged either way: sample 0 stays dark in a grey space, and only an
+/// indexed image's palette depends on what's in force where it is drawn.
+#[test]
+fn an_indexed_image_used_beyond_the_page_content_read_is_not_averaged() {
+    let (w, h) = (1200u32, 480u32);
+    let stride = (w as usize).div_ceil(8);
+    let mut data = vec![0xFF; stride * h as usize];
+    for y in 0..h as usize {
+        data[y * stride + 101 / 8] &= !(0x80 >> (101 % 8));
+    }
+    let palette = Object::String(pdfcraft_cos::PdfString::literal(vec![0, 0, 0, 255, 255, 255]));
+    let indexed = Object::Array(vec![Object::name("Indexed"), Object::name("DeviceRGB"), 1.into(), palette]);
+    let cases = [(&indexed, false, 600), (&indexed, true, 1200), (&Object::name("DeviceGray"), false, 600), (&Object::name("DeviceGray"), true, 600)];
+    for (cs, in_annotation, width) in cases {
+        let mut doc = Document::new_empty();
+        let image = mono_image(&mut doc, w, h, &data, None, &[("BitsPerComponent", 1.into()), ("ColorSpace", cs.clone())]);
+        // 144 × 57.6 pt is 600 ppi.
+        let p = page(&mut doc, &[("Im", image)], "q 144 0 0 57.6 20 600 cm /Im Do Q");
+        if in_annotation {
+            let resources = Object::Dict({
+                let mut r = Dict::new();
+                r.set(b"XObject".to_vec(), Object::Dict(xobject_dict(&[("Im", image)])));
+                r
+            });
+            let appearance = drawing(&mut doc, Some(resources), &form_entries());
+            annotate(&mut doc, p, appearance);
+        }
+        optimize(&mut doc, &Settings::default()).unwrap();
+        assert_eq!(stream(&doc, image).dict.int(b"Width"), Some(width), "{cs:?}, in an annotation: {in_annotation}");
+    }
+}
+
 /// White margins must not hide a small screened photograph from the tone guard.
 #[test]
 fn mixed_text_and_small_halftone_regions_keep_their_tones_with_wide_white_margins() {