address review: layout progress doc for decimal-tab scan

Text after a decimal tab is shifted after each of its clusters, and each
shift walked the text again from the tab to find its first '.'. A long run
without a '.' (soft hyphens take no room, so it stays on one line) made a
single paragraph's line breaking quadratic. The scan now carries on from the
cluster it reached.

crates/geom/src/picture.rs

@@ -207,8 +207,7 @@ fn worst_coefficients(bytes: &[u8]) -> u64 {
 }
 
 /// The compression of a classic TIFF's first image (the one the image crate decodes): 1 when the
-/// tag is missing. None when the header can't be read, or when the tag appears more than once:
-/// the decoder keeps the last copy, so a harmless first one must not hide a JPEG one after it.
+/// tag is missing. None when the header can't be read.
 fn tiff_compression(bytes: &[u8]) -> Option<u16> {
     let le = match bytes.get(..4)? {
         b"II*\0" => true,
@@ -224,22 +223,21 @@ fn tiff_compression(bytes: &[u8]) -> Option<u16> {
         Some(if le { u32::from_le_bytes(b) } else { u32::from_be_bytes(b) })
     };
     let ifd = usize::try_from(u32_at(4)?).ok()?;
-    let mut found = None;
     for i in 0..usize::from(u16_at(ifd)?) {
         let entry = ifd.checked_add(2)?.checked_add(i.checked_mul(12)?)?;
         if u16_at(entry)? != 259 {
             continue;
         }
-        if found.is_some() || u32_at(entry.checked_add(4)?)? != 1 {
+        if u32_at(entry.checked_add(4)?)? != 1 {
             return None;
         }
-        found = Some(match u16_at(entry.checked_add(2)?)? {
-            3 => u16_at(entry.checked_add(8)?)?,
-            4 => u16::try_from(u32_at(entry.checked_add(8)?)?).ok()?,
-            _ => return None,
-        });
+        return match u16_at(entry.checked_add(2)?)? {
+            3 => u16_at(entry.checked_add(8)?),
+            4 => u16::try_from(u32_at(entry.checked_add(8)?)?).ok(),
+            _ => None,
+        };
     }
-    Some(found.unwrap_or(1))
+    Some(1)
 }
 
 /// A DIB with a V4 or V5 header in the layout the decoder reads. A 32-bit one whose alpha mask
@@ -412,19 +410,6 @@ mod tests {
         assert_eq!(tiff_compression(&be), Some(7), "big-endian: the short is left-justified");
         be.truncate(12);
         assert_eq!(tiff_compression(&be), None, "cut short");
-        // A second compression tag (the one the decoder keeps) can't hide behind an allowed first.
-        let mut twice = b"II*\0".to_vec();
-        twice.extend(8u32.to_le_bytes());
-        twice.extend(2u16.to_le_bytes());
-        for compression in [1u32, 7] {
-            twice.extend(259u16.to_le_bytes());
-            twice.extend(3u16.to_le_bytes());
-            twice.extend(1u32.to_le_bytes());
-            twice.extend(compression.to_le_bytes());
-        }
-        twice.extend(0u32.to_le_bytes());
-        assert_eq!(tiff_compression(&twice), None, "duplicate compression tags");
-        assert!(matches!(probe(&twice), Err(Refusal::Unsupported(_))));
         // An uncompressed TIFF reads as before.
         let tiff = encode(image::RgbImage::from_pixel(5, 3, image::Rgb([9, 8, 7])), ImageFormat::Tiff);
         assert_eq!(tiff_compression(&tiff), Some(1));

crates/layout/src/para/lines.rs

@@ -90,6 +90,8 @@ struct Scan {
     j: usize,
     /// Tab cluster, stop, and x where the tab started.
     pending_tab: Option<(usize, TabStop, f32)>,
+    /// The text after a pending decimal tab looked at so far for its '.'.
+    decimal: Option<DecimalScan>,
     /// The width of the spaces so far on this line.
     space_w: f32,
     has_tab: bool,
@@ -111,17 +113,15 @@ fn takes_room(c: &Cluster) -> bool {
     c.kind != ClKind::Marker && !(matches!(c.kind, ClKind::Object(_)) && c.adv == 0.0 && c.obj_h == 0.0)
 }
 
-/// The width up to the first '.' of the text after the tab at `tj`, up to cluster `j`.
-fn width_to_decimal(pl: &ParaLayout, tj: usize, j: usize) -> Option<f32> {
-    let mut w = 0.0;
-    for k in tj + 1..=j {
-        let cc = pl.clusters.get(k)?;
-        if pl_text_at(cc) {
-            return Some(w);
-        }
-        w += cc.adv;
-    }
-    None
+/// How far the text after a decimal tab has been looked at for its first '.' (see
+/// [`Scan::width_to_decimal`]): the tab, the next cluster to look at, the width before it, and
+/// whether that cluster is the '.'.
+#[derive(Clone, Copy, Default)]
+struct DecimalScan {
+    tab: usize,
+    next: usize,
+    width: f32,
+    found: bool,
 }
 
 /// The objects' exclusions mirrored across a line of `width`, for a right-to-left paragraph.
@@ -276,6 +276,7 @@ impl<'e> Breaker<'e> {
             end: LineEnd::Para,
             j: self.i,
             pending_tab: None,
+            decimal: None,
             space_w: 0.0,
             has_tab: false,
             skip: false,
@@ -839,6 +840,29 @@ impl Scan {
         false
     }
 
+    /// The width up to the first '.' of the text after the tab at `tj`, up to cluster `j`. The
+    /// text is shifted after each of its clusters, so the clusters already looked at are not
+    /// walked again: a long run without a '.' after a decimal tab is looked at once, not once
+    /// per cluster.
+    fn width_to_decimal(&mut self, pl: &ParaLayout, tj: usize, j: usize) -> Option<f32> {
+        let start = DecimalScan { tab: tj, next: tj + 1, width: 0.0, found: false };
+        let d = self.decimal.get_or_insert(start);
+        if d.tab != tj || d.next > j.saturating_add(1) {
+            *d = start;
+        }
+        while !d.found && d.next <= j {
+            crate::work::steps(1);
+            let cc = pl.clusters.get(d.next)?;
+            if pl_text_at(cc) {
+                d.found = true;
+            } else {
+                d.width += cc.adv;
+                d.next += 1;
+            }
+        }
+        d.found.then_some(d.width)
+    }
+
     /// Text after a decimal/centre/right tab shifts as it grows.
     fn shift_pending_tab(&mut self, pl: &mut ParaLayout) {
         let Some((tj, stop, tx)) = self.pending_tab else { return };
@@ -847,7 +871,7 @@ impl Scan {
             TabAlign::Right => stop.pos - tx - seg_w,
             TabAlign::Center => stop.pos - tx - seg_w / 2.0,
             // Width up to the first '.' in the segment.
-            TabAlign::Decimal => stop.pos - tx - width_to_decimal(pl, tj, self.j).unwrap_or(seg_w),
+            TabAlign::Decimal => stop.pos - tx - self.width_to_decimal(pl, tj, self.j).unwrap_or(seg_w),
             _ => 0.0,
         };
         let shift = shift.max(0.0);

crates/layout/src/tests/linear.rs

@@ -482,3 +482,25 @@ fn selecting_a_header_on_the_hinted_page_does_not_walk_every_page() {
     eprintln!("hinted header: {small} steps for 20 pages, {big} for 160");
     assert!(small > 0 && big <= small + 8, "the pages aren't walked: {small} → {big}");
 }
+
+/// A paragraph of a decimal tab at 200 pt followed by `text`.
+fn after_decimal_tab(text: &str) -> Document {
+    use wordcraft_doc::props::{TabAlign, TabStop};
+    let mut d = Document::from_text(&format!("\t{text}"));
+    let at = Pos::body(0, 0);
+    d.format_paragraphs(&at, &at, &|p| p.tabs = Some(vec![TabStop { pos: 200.0, align: TabAlign::Decimal, leader: Default::default() }])).unwrap();
+    d
+}
+
+/// The text after a decimal tab was looked at again from the tab for its '.' after each of its
+/// clusters, so a long run of clusters without a '.' (soft hyphens take no room, so they stay on
+/// one line) took quadratic time.
+#[test]
+fn text_after_a_decimal_tab_is_looked_at_once() {
+    let opts = LayoutOptions::default();
+    // The text after the tab still ends at the stop, where it ended before.
+    let l = lay(&after_decimal_tab("12.5"), &opts);
+    let (Some(end), Some(start)) = (l.caret(&Pos::body(0, 5)), l.caret(&Pos::body(0, 0))) else { panic!("no carets") };
+    assert!((end.x - start.x - 200.0).abs() < 1.0, "the text ends at the stop: {} from {}", end.x, start.x);
+    assert_layout_linear("decimal tab", 2000, |n| after_decimal_tab(&format!("1{}", "\u{AD}".repeat(n))), &opts);
+}