A written height is a ceiling, and the judge can see 592 more boxes

corpus_test.go

@@ -1585,8 +1585,16 @@ func readPdfium(name string) (map[string][]pdfiumBox, int, error) {
 // template did not name, which contributes nothing to [FormNode.Path] either.
 // The two are the same walk written differently, so this is a rewrite and not
 // a guess.
+//
+// A dot INSIDE a name is escaped, which is why the steps are cut by
+// [somSteps] and not by strings.Split. 35 of the 559 dumped forms name an
+// element that way — "ArtifactedHeader[0].A\.Original[0]" — and splitting on
+// every dot put a backslash in the rewritten path, so the key matched nothing
+// this package emits and 592 leaves were paired with nothing at all. The
+// agreement rates were never wrong, both sides being unpaired, but they could
+// not see those boxes.
 func somPath(som string) string {
-	steps := strings.Split(som, ".")
+	steps := somSteps(som)
 	var out []string
 	for _, s := range steps {
 		name, idx := s, 0
@@ -1605,6 +1613,27 @@ func somPath(som string) string {
 	return strings.Join(out, ".")
 }
 
+// somSteps cuts a SOM expression at its UNESCAPED dots. pdfium writes a name
+// holding a dot with a backslash before it (CXFA_Object::GetSOMExpression),
+// and the backslash is not part of the name.
+func somSteps(som string) []string {
+	var out []string
+	var cur strings.Builder
+	for i := 0; i < len(som); i++ {
+		switch {
+		case som[i] == '\\' && i+1 < len(som):
+			i++
+			cur.WriteByte(som[i])
+		case som[i] == '.':
+			out = append(out, cur.String())
+			cur.Reset()
+		default:
+			cur.WriteByte(som[i])
+		}
+	}
+	return append(out, cur.String())
+}
+
 // TestIntraLineAgainstPdfium checks where a box went ACROSS a line against a
 // reference that computes the answer.
 //

flow.go

@@ -203,15 +203,41 @@ func (p *placer) measure(n *FormNode, wide, colW Measure) (Measure, string) {
 	if err != nil {
 		return 0, fmt.Sprintf("its height is written as h=%q, which is not a length", n.Template.Get("h"))
 	}
-	if !ok {
-		own = 0
+	// pdfium disables the auto-size only for a height ABOVE
+	// kXFALayoutPrecision (0.0005pt, cxfa_contentlayoutprocessor.h:28), so a
+	// height written as nought is no height and the content decides. No
+	// container of the 560-form corpus writes one, which is why this is a
+	// guard and not a measured claim.
+	if own <= 0 {
+		ok = false
+	}
+	// A WRITTEN height is the height. It is a ceiling and not a floor: what
+	// the container holds does not grow it.
+	//
+	// pdf.js takes the other reading — Math.max(this[$extra].height + marginV,
+	// this.h || 0) (template.js:5222) — and this package followed it until the
+	// oracle was asked. pdfium decides it in two steps.
+	// CalculateContainerSpecifiedSize (cxfa_contentlayoutprocessor.cpp:97-137)
+	// turns bContainerHeightAutoSize off as soon as a subform writes an h above
+	// kXFALayoutPrecision, and CalculateContainerComponentSizeFromContentSize
+	// (:140-184) then leaves componentSize.height at that written height,
+	// never reading fContentCalculatedHeight at all.
+	//
+	// Measured over the corpus against pdfium, following pdfium: leaves
+	// agreeing on sheet AND y 161453/177676 -> 161937/177676, forms agreeing on
+	// every one of them 395/559 -> 403/559, no form worse and not one x moved.
+	// ca-cra__rc243-fill-26e's Detail2 row is the case in the small: it writes
+	// h="4.233mm" and holds children reaching 8.466mm, and pdfium puts the row
+	// that follows at 607.0961 — the written 11.9991 below the row above, to
+	// the point.
+	//
+	// The content height is still computed first, and a reason it could not be
+	// still wins, so a container holding something unmeasurable stays
+	// unmeasurable rather than falling back on what is written.
+	if ok {
+		return own, ""
 	}
-	// pdf.js: Math.max(this[$extra].height + marginV, this.h || 0)
-	// (template.js:5222). A container is as tall as what it holds even where
-	// the template writes a height, which is why holding a thing of unknown
-	// height leaves the container's own height unknown too rather than falling
-	// back on what is written.
-	return max(content+in.vertical(), own), ""
+	return content + in.vertical(), ""
 }
 
 // innerWide is how much horizontal room a container gives what it holds.

flow_test.go

@@ -481,3 +481,44 @@ func TestInsideAContainerThatMovesWholeAStackStillStops(t *testing.T) {
 		"f.G.B: a tb layout stacks its children, and the height of the one above it is not computed: " +
 			"its margin is not written in lengths"})
 }
+
+// TestAWrittenHeightIsACeilingNotAFloor pins the rule the pdfium oracle
+// settled: a container that writes an h is that tall, and what it holds does
+// not stretch it. pdf.js takes the max of the two (template.js:5222) and this
+// package followed that until the oracle was asked; pdfium's
+// CalculateContainerSpecifiedSize turns the auto-size off outright
+// (cxfa_contentlayoutprocessor.cpp:97-137).
+func TestAWrittenHeightIsACeilingNotAFloor(t *testing.T) {
+	for _, tc := range []struct {
+		what string
+		body string
+		want []string
+	}{
+		{"content taller than the written height does not stretch it",
+			// The box holds a child reaching 24 points and writes 12, so the
+			// draw that follows it begins at 12 and not at 24.
+			`<subform name="Box" h="12pt"><draw name="A" y="12pt" w="1pt" h="12pt"/></subform>
+			 <draw name="After" w="1pt" h="3pt"/>`,
+			[]string{"draw f.Box.A 0,12 1x12", "draw f.After 0,12 1x3"}},
+		{"content shorter than the written height does not shrink it",
+			`<subform name="Box" h="30pt"><draw name="A" w="1pt" h="4pt"/></subform>
+			 <draw name="After" w="1pt" h="3pt"/>`,
+			[]string{"draw f.Box.A 0,0 1x4", "draw f.After 0,30 1x3"}},
+		{"a height written as nought is no height, and the content decides",
+			// pdfium reads a height only above kXFALayoutPrecision
+			// (cxfa_contentlayoutprocessor.h:28), so this stacks as if the
+			// attribute were absent.
+			`<subform name="Box" h="0pt"><draw name="A" w="1pt" h="7pt"/></subform>
+			 <draw name="After" w="1pt" h="3pt"/>`,
+			[]string{"draw f.Box.A 0,0 1x7", "draw f.After 0,7 1x3"}},
+		{"a written height with a margin is still the whole of it",
+			`<subform name="Box" h="12pt"><margin topInset="5pt" bottomInset="5pt"/>
+			   <draw name="A" w="1pt" h="20pt"/></subform>
+			 <draw name="After" w="1pt" h="3pt"/>`,
+			[]string{"draw f.Box.A 0,5 1x20", "draw f.After 0,12 1x3"}},
+	} {
+		if got := laid(laidOut(t, page(`w="500pt" h="500pt"`, tc.body))); strings.Join(got, "\n") != strings.Join(tc.want, "\n") {
+			t.Errorf("%s:\n  got  %v\n  want %v", tc.what, got, tc.want)
+		}
+	}
+}