Accessibility check asks for a manual check where page content ran past the decode budget

crates/a11y/src/content.rs

@@ -20,6 +20,8 @@ pub(crate) struct Scan {
     pub untagged: BTreeMap<usize, usize>,
     /// Page → fonts used.
     pub fonts: BTreeMap<usize, BTreeSet<ObjRef>>,
+    /// Pages read only in part: content or forms past the page's decode budget are missing.
+    pub incomplete: BTreeSet<usize>,
 }
 
 const MAX_FORM_DEPTH: usize = 12;
@@ -32,6 +34,9 @@ impl Scan {
             let res = page.dict.get(b"Resources").map(|r| doc.resolve(r)).and_then(|r| r.as_dict().cloned()).unwrap_or_default();
             let mut budget = DecodeBudget::page();
             let data = crate::alt::page_content_within(doc, &page.dict, &mut budget);
+            if budget.remaining() == 0 {
+                s.incomplete.insert(p);
+            }
             let mut w = Walk { doc, scan: &mut s, page: p, forms: HashSet::new(), budget };
             w.content(&data, &res, false, 0);
         }
@@ -100,7 +105,20 @@ impl Walk<'_> {
                             self.scan.has_images = true;
                             untagged += usize::from(!inside);
                         }
-                        Some(b"Form") if depth < MAX_FORM_DEPTH && !self.forms.contains(&r) && self.budget.enter_form() => {
+                        Some(b"Form") if depth < MAX_FORM_DEPTH && !self.forms.contains(&r) => {
+                            if !self.budget.enter_form() {
+                                self.scan.incomplete.insert(self.page);
+                                continue;
+                            }
+                            let data = match self.budget.decode(x) {
+                                Ok(data) => data,
+                                Err(_) => {
+                                    if self.budget.exhausted() {
+                                        self.scan.incomplete.insert(self.page);
+                                    }
+                                    continue;
+                                }
+                            };
                             self.forms.insert(r);
                             let inner = x
                                 .dict
@@ -108,7 +126,6 @@ impl Walk<'_> {
                                 .map(|o| self.doc.resolve(o))
                                 .and_then(|o| o.as_dict().cloned())
                                 .unwrap_or_else(|| res.clone());
-                            let data = self.budget.decode(x).unwrap_or_default();
                             self.content(&data, &inner, inside, depth + 1);
                             self.forms.remove(&r);
                         }

crates/a11y/src/lib.rs

@@ -681,6 +681,18 @@ pub fn check(doc: &Document, options: &Options) -> Report {
                 }
             }
         }
+        // A page read only in part may hold what the scan didn't find: text that makes the
+        // document more than images, untagged content, or a font without Unicode mappings.
+        let unsure = !scan.incomplete.is_empty()
+            && match rule {
+                Rule::ImageOnly => !scan.has_text,
+                Rule::TaggedContent | Rule::CharacterEncoding => findings.is_empty(),
+                _ => false,
+            };
+        if unsure {
+            results.push(RuleResult { rule, status: Status::Manual, findings: Vec::new() });
+            continue;
+        }
         if total > MAX_FINDINGS {
             let more = total - MAX_FINDINGS;
             findings.truncate(MAX_FINDINGS);

crates/a11y/src/tests.rs

@@ -1278,3 +1278,31 @@ fn marking_decorative_refuses_content_past_the_decode_budget() {
     assert!(mark_decorative(&mut doc, ObjRef::new(7, 0)).is_err());
     assert_eq!(figures(&doc).len(), 2, "the figure stays");
 }
+
+/// Text past the page's decode budget is unread, so a page that shows images and no text so far
+/// can't be called image-only, nor its content all tagged.
+#[test]
+fn content_past_the_decode_budget_needs_a_manual_check() {
+    use pdfcraft_cos::{Dict, Object, ObjRef, Stream};
+    let mut doc = pdf(
+        &[
+            "<< /Type /Catalog /Pages 2 0 R >>".into(),
+            "<< /Type /Pages /Kids [3 0 R] /Count 1 >>".into(),
+            "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 200] /Contents 4 0 R /Resources << /XObject << /Im 5 0 R >> /Font << /F1 6 0 R >> >> >>".into(),
+            stream("/Artifact BMC q 200 0 0 200 0 0 cm /Im Do Q EMC"),
+            "<< /Type /XObject /Subtype /Image /Width 1 /Height 1 /ColorSpace /DeviceGray /BitsPerComponent 8 /Length 1 >>\nstream\n\u{0}\nendstream"
+                .into(),
+            FONT.into(),
+        ],
+        "",
+    );
+    let filler = doc.add(Stream::flate(Dict::new(), &vec![b' '; 1 << 20]));
+    let text = doc.add(Stream::from_raw(Dict::new(), b"BT /F1 12 Tf 10 10 Td (Hello) Tj ET".to_vec()));
+    let mut contents = vec![Object::Ref(ObjRef::new(4, 0))];
+    contents.extend(std::iter::repeat_n(Object::Ref(filler), 256));
+    contents.push(Object::Ref(text));
+    doc.update_dict(ObjRef::new(3, 0), |p| p.set(b"Contents".to_vec(), Object::Array(contents))).unwrap();
+    let report = check(&doc, &Options::default());
+    assert_eq!(status(&report, Rule::ImageOnly), Status::Manual);
+    assert_eq!(status(&report, Rule::TaggedContent), Status::Manual);
+}