textmachine/backend/internal/chunk/epubtoc_test.go

package chunk

import (
	"os"
	"path/filepath"
	"strings"
	"testing"

	"textmachine/backend/internal/chunk/chunktest"
	"textmachine/backend/internal/lang"
	"textmachine/backend/internal/text"
)

// ch is a content document with enough prose to be a chapter.
func ch(id, href, body string) chunktest.Chapter {
	return chunktest.Chapter{ID: id, Href: href, Body: "<p>" + body + "</p>"}
}

func refs(ids ...string) []chunktest.SpineRef {
	out := make([]chunktest.SpineRef, 0, len(ids))
	for _, id := range ids {
		out = append(out, chunktest.SpineRef{ID: id})
	}
	return out
}

// TestEPUBNavDrawsChaptersNotTheSpine is the pack's central claim: the spine is reading ORDER and the nav is
// the book's own statement about chapters. Six documents, three nav targets ⇒ three chapters, and the
// documents between targets are FOLDED IN rather than becoming chapters of their own.
func TestEPUBNavDrawsChaptersNotTheSpine(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("c1", "c1.xhtml", "one a"), ch("c1b", "c1b.xhtml", "one b"),
			ch("c2", "c2.xhtml", "two a"), ch("c2b", "c2b.xhtml", "two b"),
			ch("c3", "c3.xhtml", "three a"), ch("c3b", "c3b.xhtml", "three b"),
		},
		Spine: refs("c1", "c1b", "c2", "c2b", "c3", "c3b"),
		Nav:   &chunktest.Nav{TOC: []string{"c1.xhtml", "c2.xhtml", "c3.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 3 {
		t.Fatalf("chapters = %d, want 3 (the nav named three)", len(doc.Chapters))
	}
	if doc.Structure != StructureDeclared {
		t.Fatalf("structure = %q, want %q — a nav IS the format declaring its chapters", doc.Structure, StructureDeclared)
	}
	// ⛔ No text was dropped: every document's prose survives, in its own chapter.
	for i, want := range []string{"one a", "one b"} {
		if !strings.Contains(doc.Chapters[0], want) {
			t.Fatalf("chapter 1 lost %q (part %d): %q", want, i, doc.Chapters[0])
		}
	}
	if doc.DocumentsAttached != 3 {
		t.Fatalf("documents attached = %d, want 3 (one folded into each chapter)", doc.DocumentsAttached)
	}
}

// TestEPUB2NCXDrawsChaptersWhenThereIsNoNav is the EPUB 2 half. An EPUB 2 has no navigation document at all,
// so a reader that knows only about nav falls straight through to the spine and reports the wrong chapters.
func TestEPUB2NCXDrawsChaptersWhenThereIsNoNav(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("c1", "c1.xhtml", "one a"), ch("c1b", "c1b.xhtml", "one b"),
			ch("c2", "c2.xhtml", "two a"),
		},
		Spine: refs("c1", "c1b", "c2"),
		NCX:   []string{"c1.xhtml", "c2.xhtml"},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 || doc.Structure != StructureDeclared {
		t.Fatalf("chapters = %d structure = %q, want 2 / %q", len(doc.Chapters), doc.Structure, StructureDeclared)
	}
	if !strings.Contains(doc.Chapters[0], "one b") {
		t.Fatalf("the document between NCX targets was lost: %q", doc.Chapters[0])
	}
}

// TestEPUBSpineIsDelimitedNotDeclared pins the vocabulary fix: with no nav and no NCX the engine still cuts
// on the spine, but it no longer calls that the format declaring its chapters.
func TestEPUBSpineIsDelimitedNotDeclared(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two")},
		Spine:    refs("c1", "c2"),
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if doc.Structure != StructureDelimited {
		t.Fatalf("structure = %q, want %q — a spine is reading order, not a chapter list", doc.Structure, StructureDelimited)
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2", len(doc.Chapters))
	}
}

// TestEPUB3LandmarksExcludeTocAndCover: EPUB 3 declares service pages inside the navigation document.
func TestEPUB3LandmarksExcludeTocAndCover(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("cov", "cover.xhtml", "COVERPAGE"), ch("toc", "toc.xhtml", "TOCPAGE"),
			ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two"),
		},
		Spine: refs("cov", "toc", "c1", "c2"),
		Nav: &chunktest.Nav{
			TOC:       []string{"c1.xhtml", "c2.xhtml"},
			Landmarks: map[string]string{"cover.xhtml": "cover", "toc.xhtml": "toc"},
			InSpine:   true,
		},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2", len(doc.Chapters))
	}
	for _, bad := range []string{"COVERPAGE", "TOCPAGE"} {
		for i, c := range doc.Chapters {
			if strings.Contains(c, bad) {
				t.Fatalf("service page %q leaked into chapter %d", bad, i+1)
			}
		}
	}
	// The nav document itself is in the spine here and is excluded on its `properties`, so three pages went.
	if len(doc.Excluded) != 3 {
		t.Fatalf("documents excluded = %d, want 3 (cover + toc + the nav document)", len(doc.Excluded))
	}
}

// TestEPUB2GuideExcludesTocAndCover is the case a landmarks-only reader gets wrong. An EPUB 2 has no
// navigation document, so the ONLY declaration of "this page is the toc" is the OPF <guide>. Miss it and a
// text table of contents is glued to chapter one — the mass case, not a corner one.
func TestEPUB2GuideExcludesTocAndCover(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("cov", "cover.xhtml", "COVERPAGE"), ch("toc", "toc.xhtml", "TOCPAGE"),
			ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two"),
		},
		Spine: refs("cov", "toc", "c1", "c2"),
		NCX:   []string{"c1.xhtml", "c2.xhtml"},
		Guide: []chunktest.GuideRef{{Type: "cover", Href: "cover.xhtml"}, {Type: "toc", Href: "toc.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2", len(doc.Chapters))
	}
	if strings.Contains(doc.Chapters[0], "TOCPAGE") {
		t.Fatal("the EPUB 2 table-of-contents page was glued to chapter one — <guide> was not read")
	}
	if len(doc.Excluded) != 2 {
		t.Fatalf("documents excluded = %d, want 2", len(doc.Excluded))
	}
}

// TestEPUBCoverImagePropertyExcludesNothing: `properties="cover-image"` sits on an IMAGE, not on a document.
// A reader that greps the attribute for "cover" would drop a real chapter.
func TestEPUBCoverImagePropertyExcludesNothing(t *testing.T) {
	chapters := []chunktest.Chapter{ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two")}
	chapters[0].Properties = "cover-image"
	p := chunktest.EPUB{Chapters: chapters, Spine: refs("c1", "c2")}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 || len(doc.Excluded) != 0 {
		t.Fatalf("chapters = %d excluded = %d, want 2 / 0 — cover-image is not the cover DOCUMENT",
			len(doc.Chapters), len(doc.Excluded))
	}
}

// TestEPUBNonLinearNeverStartsAChapter: linear="no" marks appendices and footnote pages. They are text and
// are kept, but they do not open a chapter even when the nav points at them.
func TestEPUBNonLinearNeverStartsAChapter(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("c1", "c1.xhtml", "one"), ch("notes", "notes.xhtml", "FOOTNOTES"), ch("c2", "c2.xhtml", "two"),
		},
		Spine: []chunktest.SpineRef{{ID: "c1"}, {ID: "notes", Linear: "no"}, {ID: "c2"}},
		Nav:   &chunktest.Nav{TOC: []string{"c1.xhtml", "notes.xhtml", "c2.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2 — the non-linear page must not open one", len(doc.Chapters))
	}
	if !strings.Contains(doc.Chapters[0], "FOOTNOTES") {
		t.Fatalf("the non-linear page was DROPPED instead of attached: %q", doc.Chapters[0])
	}
}

// TestEPUBUncoveredBeforeFirstTargetJoinsChapterOne mirrors the txt preamble rule: a title page the nav does
// not mention joins chapter one rather than becoming a chapter or vanishing.
func TestEPUBUncoveredBeforeFirstTargetJoinsChapterOne(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("title", "title.xhtml", "TITLEPAGE"), ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two"),
		},
		Spine: refs("title", "c1", "c2"),
		Nav:   &chunktest.Nav{TOC: []string{"c1.xhtml", "c2.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2", len(doc.Chapters))
	}
	if !strings.Contains(doc.Chapters[0], "TITLEPAGE") {
		t.Fatalf("the uncovered title page was lost: %q", doc.Chapters[0])
	}
}

// TestEPUBNavWithNoResolvableTargetsIsAbsent: a nav pointing at documents that are not in the spine is a
// BROKEN nav. It must fall through to the next witness, not produce zero chapters.
func TestEPUBNavWithNoResolvableTargetsIsAbsent(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two")},
		Spine:    refs("c1", "c2"),
		Nav:      &chunktest.Nav{TOC: []string{"gone-a.xhtml", "gone-b.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 || doc.Structure != StructureDelimited {
		t.Fatalf("chapters = %d structure = %q, want 2 / %q", len(doc.Chapters), doc.Structure, StructureDelimited)
	}
	if doc.TOCUnresolved != 2 {
		t.Fatalf("unresolved targets = %d, want 2 — the count is the whole alarm", doc.TOCUnresolved)
	}
}

// TestEPUBDuplicateNavTargetsCollapse: several nav entries pointing into ONE document (chapter + its
// sections) are one boundary, not several empty ones.
func TestEPUBDuplicateNavTargetsCollapse(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two")},
		Spine:    refs("c1", "c2"),
		Nav:      &chunktest.Nav{TOC: []string{"c1.xhtml", "c1.xhtml#s2", "c2.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2 — a fragment into the same document is the same boundary", len(doc.Chapters))
	}
	if doc.TargetsCollapsed != 1 {
		t.Fatalf("targets collapsed = %d, want 1", doc.TargetsCollapsed)
	}
	// ⛔ AND THE PROVENANCE IS DOWNGRADED. The nav named three chapters; this package hands back two, because
	// it groups whole documents and does not slice one by its anchors. Reporting that as `declared` would put
	// the engine's own coarser answer out under the format's name — the very lie the spine used to tell.
	if doc.Structure != StructureDelimited {
		t.Fatalf("structure = %q, want %q — the format named more chapters than the cut returned",
			doc.Structure, StructureDelimited)
	}
}

// TestEPUBNavOracleReproducesTheTxtCut is the code oracle the format work is measured by: ground truth with
// no hand annotation and no model call. The nav DECLARES where chapters begin; unroll the same book into a
// plain txt whose chapters open with source headers, and the txt path must recover the same boundaries.
// Two independent readers, one answer.
func TestEPUBNavOracleReproducesTheTxtCut(t *testing.T) {
	bodies := []string{"第一章\n" + strings.Repeat("蛊", 200), "第二章\n" + strings.Repeat("蛊", 200), "第三章\n" + strings.Repeat("蛊", 200)}
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("c1", "c1.xhtml", bodies[0]), ch("c2", "c2.xhtml", bodies[1]), ch("c3", "c3.xhtml", bodies[2]),
		},
		Spine: refs("c1", "c2", "c3"),
		Nav:   &chunktest.Nav{TOC: []string{"c1.xhtml", "c2.xhtml", "c3.xhtml"}},
	}.Build(t)

	fromEPUB, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	txt := writeTempTXT(t, strings.Join(bodies, "\n"))
	fromTXT, err := ingest(txt)
	if err != nil {
		t.Fatal(err)
	}
	if len(fromTXT.Chapters) != len(fromEPUB.Chapters) {
		t.Fatalf("txt cut %d chapters, the nav declared %d", len(fromTXT.Chapters), len(fromEPUB.Chapters))
	}
	for i := range fromEPUB.Chapters {
		if fromTXT.Chapters[i] != fromEPUB.Chapters[i] {
			t.Fatalf("chapter %d differs:\n txt: %.60q\nepub: %.60q", i+1, fromTXT.Chapters[i], fromEPUB.Chapters[i])
		}
	}
	if fromEPUB.Structure != StructureDeclared || fromTXT.Structure != StructureDetected {
		t.Fatalf("provenance should differ even where the cut agrees: epub=%q txt=%q", fromEPUB.Structure, fromTXT.Structure)
	}
}

// writeTempTXT writes body to a temp .txt and returns its path.
func writeTempTXT(t *testing.T, body string) string {
	t.Helper()
	p := filepath.Join(t.TempDir(), "src.txt")
	if err := os.WriteFile(p, []byte(body), 0o644); err != nil {
		t.Fatal(err)
	}
	return p
}

// TestEPUBNavTargetsFollowSpineOrderNotTocOrder: a table of contents may list its entries in any order (an
// appendix first, a foreword last). Boundaries are spine POSITIONS, so a mis-ordered toc must not interleave
// the chapters or hand a later document to an earlier chapter.
func TestEPUBNavTargetsFollowSpineOrderNotTocOrder(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("c1", "c1.xhtml", "ONE"), ch("c1b", "c1b.xhtml", "ONEB"),
			ch("c2", "c2.xhtml", "TWO"), ch("c2b", "c2b.xhtml", "TWOB"),
		},
		Spine: refs("c1", "c1b", "c2", "c2b"),
		Nav:   &chunktest.Nav{TOC: []string{"c2.xhtml", "c1.xhtml"}}, // declared out of reading order
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2", len(doc.Chapters))
	}
	if !strings.Contains(doc.Chapters[0], "ONE") || !strings.Contains(doc.Chapters[0], "ONEB") {
		t.Fatalf("chapter 1 is not the FIRST span of the book: %q", doc.Chapters[0])
	}
	if !strings.Contains(doc.Chapters[1], "TWO") || !strings.Contains(doc.Chapters[1], "TWOB") {
		t.Fatalf("chapter 2 is not the SECOND span of the book: %q", doc.Chapters[1])
	}
}

// TestEPUBPartiallyUnresolvedNavStillCuts: a scrape whose toc points half at real documents and half at
// nothing must use what resolved and COUNT the rest, not fall back and not fail.
func TestEPUBPartiallyUnresolvedNavStillCuts(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two")},
		Spine:    refs("c1", "c2"),
		Nav:      &chunktest.Nav{TOC: []string{"c1.xhtml", "gone.xhtml", "c2.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if len(doc.Chapters) != 2 || doc.Structure != StructureDeclared {
		t.Fatalf("chapters = %d structure = %q, want 2 / %q", len(doc.Chapters), doc.Structure, StructureDeclared)
	}
	if doc.TOCUnresolved != 1 {
		t.Fatalf("unresolved = %d, want 1", doc.TOCUnresolved)
	}
}

// TestEPUBAllServiceSpineFailsLoud: a book whose every spine document is a declared service page has no text
// to translate. That must be a loud refusal, never an empty book that costs money to discover.
func TestEPUBAllServiceSpineFailsLoud(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("cov", "cover.xhtml", "COVER"), ch("toc", "toc.xhtml", "TOC")},
		Spine:    refs("cov", "toc"),
		NCX:      []string{"cover.xhtml"},
		Guide:    []chunktest.GuideRef{{Type: "cover", Href: "cover.xhtml"}, {Type: "toc", Href: "toc.xhtml"}},
	}.Build(t)

	if _, err := ingest(p); err == nil {
		t.Fatal("a spine of nothing but service pages must fail loud, not return an empty book")
	}
}

// TestEPUBSingleDocumentChapterIsByteIdenticalToTheOldPath guards the refactor itself: grouping now JOINS a
// chapter's documents, and the one-document case must go through that join unchanged — otherwise every
// existing EPUB book silently re-cuts and re-pays.
func TestEPUBSingleDocumentChapterIsByteIdenticalToTheOldPath(t *testing.T) {
	body := "<p>Первый абзац.</p><p>Второй абзац.</p>"
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{{ID: "c1", Href: "c1.xhtml", Body: body}, {ID: "c2", Href: "c2.xhtml", Body: body}},
		Spine:    refs("c1", "c2"),
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	// The expectation is COMPUTED from the primitives, not guessed: whatever extractXHTML produces for this
	// document, a one-document chapter must be exactly that, normalized — the join must add nothing.
	raw, _, err := extractXHTML([]byte(`<?xml version="1.0" encoding="utf-8"?>
<html xmlns="http://www.w3.org/1999/xhtml"><head><title>c</title><style>.x{color:red}</style></head>
<body>` + body + `</body></html>`))
	if err != nil {
		t.Fatal(err)
	}
	want := text.NormalizeSource(raw)
	for i, c := range doc.Chapters {
		if c != want {
			t.Fatalf("chapter %d = %q, want %q — the join altered a single-document chapter", i+1, c, want)
		}
	}
}

// TestANewLanguageCutsFromItsDataFileAlone answers the project's default review question end to end: a pair
// that is NOT in this repository ships one data file and gets a real chapter cut, with no Go change.
//
// Korean proves both halves. With no file it falls to the embedded CJK default, whose marker 第 never appears
// in a Korean book, so the text stays ONE chapter. With ko/structure.txt naming 제 and 장 — the same binary,
// the same text — it cuts into three and reports `detected`.
func TestANewLanguageCutsFromItsDataFileAlone(t *testing.T) {
	prose := strings.Repeat("가", 200)
	body := "제1장 시작\n" + prose + "\n제2장 전개\n" + prose + "\n제3장 결말\n" + prose
	src := writeTempTXT(t, body)

	bare, err := lang.LoadSourceStructure("", "ko")
	if err != nil {
		t.Fatal(err)
	}
	before, err := IngestEncoded(src, "", "ko", bare)
	if err != nil {
		t.Fatal(err)
	}
	if len(before.Chapters) != 1 {
		t.Fatalf("without its own grammar a Korean book must stay one chapter, got %d", len(before.Chapters))
	}

	root := t.TempDir()
	if err := os.MkdirAll(filepath.Join(root, "ko"), 0o755); err != nil {
		t.Fatal(err)
	}
	if err := os.WriteFile(filepath.Join(root, "ko", "structure.txt"), []byte("marker\t제\nunits\t장\n"), 0o644); err != nil {
		t.Fatal(err)
	}
	st, err := lang.LoadSourceStructure(root, "ko")
	if err != nil {
		t.Fatal(err)
	}
	after, err := IngestEncoded(src, "", "ko", st)
	if err != nil {
		t.Fatal(err)
	}
	if len(after.Chapters) != 3 {
		t.Fatalf("chapters = %d, want 3 — the language's own grammar did not cut", len(after.Chapters))
	}
	if after.Structure != StructureDetected {
		t.Fatalf("structure = %q, want %q", after.Structure, StructureDetected)
	}
	if after.Titles[1] != "제2장 전개" {
		t.Fatalf("title_raw of chapter 2 = %q, want %q", after.Titles[1], "제2장 전개")
	}
}

// TestEPUBServiceRoleWithAFragmentDoesNotDeleteTheDocument is the text-loss guard.
//
// ⛔ `<reference type="toc" href="c2.xhtml#pos"/>` says the table of contents BEGINS at that point — inside a
// document that may hold prose before and after it. Reading it as "this whole file is the toc" deletes book
// text while the chapter count stays plausible and the provenance stays a confident `declared`: nothing
// looks wrong. The form is not hypothetical — the owner's Kristoff EPUB carries exactly it
// (`type="toc" href="index_split_124.html#filepos2096584"`), and there the split file happened to be pure
// front matter. One calibre split the other way and it is a chapter.
func TestEPUBServiceRoleWithAFragmentDoesNotDeleteTheDocument(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("c1", "c1.xhtml", "PROSEONE"), ch("c2", "c2.xhtml", "PROSETWO"), ch("c3", "c3.xhtml", "PROSETHREE"),
		},
		Spine: refs("c1", "c2", "c3"),
		NCX:   []string{"c1.xhtml", "c2.xhtml", "c3.xhtml"},
		Guide: []chunktest.GuideRef{{Type: "toc", Href: "c2.xhtml#tocpos"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	joined := strings.Join(doc.Chapters, "\x00")
	for _, want := range []string{"PROSEONE", "PROSETWO", "PROSETHREE"} {
		if !strings.Contains(joined, want) {
			t.Fatalf("%s was deleted: a role declared at a POINT inside a document excluded the whole file", want)
		}
	}
	if len(doc.Excluded) != 0 {
		t.Fatalf("documents excluded = %d, want 0 — the declaration named a place, not a document", len(doc.Excluded))
	}
}

// TestEPUBWholeDocumentServiceRoleStillExcludes is the other half: without a fragment the declaration IS
// about the document, and the page must still go.
func TestEPUBWholeDocumentServiceRoleStillExcludes(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("toc", "toc.xhtml", "TOCPAGE"), ch("c1", "c1.xhtml", "PROSE")},
		Spine:    refs("toc", "c1"),
		NCX:      []string{"c1.xhtml"},
		Guide:    []chunktest.GuideRef{{Type: "toc", Href: "toc.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if strings.Contains(strings.Join(doc.Chapters, ""), "TOCPAGE") {
		t.Fatal("a whole-document toc declaration must still exclude the page")
	}
	if len(doc.Excluded) != 1 {
		t.Fatalf("documents excluded = %d, want 1", len(doc.Excluded))
	}
}

// TestEPUBOrdinaryBookDoesNotReportABrokenTOC: a table of contents that lists the book's own cover resolves
// perfectly — the cover is right there in the spine — it is simply not a chapter boundary. Counting that as
// "the toc named nothing" makes every ordinary EPUB 2 report a broken table, and an alarm that fires on the
// normal case is one nobody reads.
func TestEPUBOrdinaryBookDoesNotReportABrokenTOC(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("cov", "cover.xhtml", "COVER"), ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two"),
		},
		Spine: refs("cov", "c1", "c2"),
		NCX:   []string{"cover.xhtml", "c1.xhtml", "c2.xhtml"},
		Guide: []chunktest.GuideRef{{Type: "cover", Href: "cover.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if doc.TOCUnresolved != 0 {
		t.Fatalf("unresolved = %d, want 0 — every target named a document this book has", doc.TOCUnresolved)
	}
	if len(doc.Excluded) != 1 {
		t.Fatalf("documents excluded = %d, want 1 — the cover is counted ONCE, as an exclusion", len(doc.Excluded))
	}
	if len(doc.Chapters) != 2 {
		t.Fatalf("chapters = %d, want 2", len(doc.Chapters))
	}
}

// TestEPUBDanglingNavCountSurvivesTheFallToNCX: a nav whose targets all dangle is replaced by the NCX, and
// how much of it dangled is still a fact about the book. Overwriting the resolved table wholesale threw that
// count away with the table.
func TestEPUBDanglingNavCountSurvivesTheFallToNCX(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two")},
		Spine:    refs("c1", "c2"),
		Nav:      &chunktest.Nav{TOC: []string{"gone-a.xhtml", "gone-b.xhtml"}},
		NCX:      []string{"c1.xhtml", "c2.xhtml"},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if doc.Structure != StructureDeclared || len(doc.Chapters) != 2 {
		t.Fatalf("structure = %q chapters = %d, want %q / 2 — the NCX must take over", doc.Structure, len(doc.Chapters), StructureDeclared)
	}
	if doc.TOCUnresolved != 2 {
		t.Fatalf("unresolved = %d, want 2 — the broken nav's dangling targets are still a fact", doc.TOCUnresolved)
	}
}

// TestEPUBTitleRawComesFromTheNavLabel: the EPUB half of title_raw had no test at all — removing the label
// collection entirely left every chapter with an empty title and the battery green.
func TestEPUBTitleRawComesFromTheNavLabel(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("c1", "c1.xhtml", "one"), ch("c1b", "c1b.xhtml", "one more"), ch("c2", "c2.xhtml", "two"),
		},
		Spine: refs("c1", "c1b", "c2"),
		Nav:   &chunktest.Nav{TOC: []string{"c1.xhtml", "c2.xhtml"}},
		NCX:   []string{"c1.xhtml", "c2.xhtml"}, // present but LOSING: nav takes precedence
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	want := []string{"NAV c1.xhtml", "NAV c2.xhtml"}
	if len(doc.Titles) != len(want) {
		t.Fatalf("titles = %d, want %d", len(doc.Titles), len(want))
	}
	for i, w := range want {
		if doc.Titles[i] != w {
			// A wrong SOURCE is the interesting failure: an "NCX …" here means the losing table supplied it.
			t.Fatalf("title %d = %q, want %q", i+1, doc.Titles[i], w)
		}
	}
}

// TestEPUB2TitleRawComesFromTheNCXLabel is the EPUB 2 half, where navLabel lives in the NCX.
func TestEPUB2TitleRawComesFromTheNCXLabel(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two")},
		Spine:    refs("c1", "c2"),
		NCX:      []string{"c1.xhtml", "c2.xhtml"},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	want := []string{"NCX c1.xhtml", "NCX c2.xhtml"}
	for i, w := range want {
		if doc.Titles[i] != w {
			t.Fatalf("title %d = %q, want %q", i+1, doc.Titles[i], w)
		}
	}
}

// TestEPUBTitleRawFollowsTheChapterOpener: when documents are folded in AHEAD of a boundary, the chapter's
// title is the boundary's, not the first document's. The opener is stored rather than derived precisely
// because "first element of the group" is the wrong answer exactly here.
func TestEPUBTitleRawFollowsTheChapterOpener(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{
			ch("title", "title.xhtml", "TITLEPAGE"), ch("c1", "c1.xhtml", "one"), ch("c2", "c2.xhtml", "two"),
		},
		Spine: refs("title", "c1", "c2"),
		Nav:   &chunktest.Nav{TOC: []string{"c1.xhtml", "c2.xhtml"}},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if doc.Titles[0] != "NAV c1.xhtml" {
		t.Fatalf("chapter 1 title = %q — it must name the BOUNDARY, not the title page folded in ahead of it",
			doc.Titles[0])
	}
}

// TestEPUB3LandmarkWithAFragmentDoesNotDeleteTheDocument is the EPUB 3 half of the same rule the guide half
// keeps: a landmark naming a POINT inside a document marks a place, not the document.
func TestEPUB3LandmarkWithAFragmentDoesNotDeleteTheDocument(t *testing.T) {
	p := chunktest.EPUB{
		Chapters: []chunktest.Chapter{ch("c1", "c1.xhtml", "PROSEONE"), ch("c2", "c2.xhtml", "PROSETWO")},
		Spine:    refs("c1", "c2"),
		Nav: &chunktest.Nav{
			TOC:       []string{"c1.xhtml", "c2.xhtml"},
			Landmarks: map[string]string{"c1.xhtml#tocstart": "toc"},
		},
	}.Build(t)

	doc, err := ingest(p)
	if err != nil {
		t.Fatal(err)
	}
	if !strings.Contains(strings.Join(doc.Chapters, ""), "PROSEONE") {
		t.Fatal("a landmark at a POINT inside a document excluded the whole file and deleted its prose")
	}
	if len(doc.Excluded) != 0 {
		t.Fatalf("documents excluded = %d, want 0", len(doc.Excluded))
	}
	if doc.ServiceRolesInsideDocuments != 1 {
		t.Fatalf("service roles inside documents = %d, want 1 — a refusal must be counted, not silent",
			doc.ServiceRolesInsideDocuments)
	}
}

// TestNavLabelIsTrimmedLikeItsSiblings pins the answer to be the same whichever table the book carries.
//
// ⛔ Real books indent their nav. Untrimmed, one title arrives as "\n        Chapter One\n      " through a
// nav and as "Chapter One" through an NCX — the same chapter, two answers, decided by which table the
// publisher happened to ship. It goes on the wire as title_raw, so a reader renders the whitespace.
// The pack's own fixtures could not catch it: they write labels inline, with no indentation to trim.
func TestNavLabelIsTrimmedLikeItsSiblings(t *testing.T) {
	indented := []byte(`<html xmlns:epub="http://www.idpf.org/2007/ops"><body>
<nav epub:type="toc">
  <ol>
    <li>
      <a href="c1.xhtml">
        Chapter One
      </a>
    </li>
  </ol>
</nav>
</body></html>`)
	_, navLabels, _, _ := parseNavDoc(indented, "OEBPS")
	ncx := []byte(`<ncx><navMap><navPoint><navLabel><text>
        Chapter One
      </text></navLabel><content src="c1.xhtml"/></navPoint></navMap></ncx>`)
	_, ncxLabels := parseNCX(ncx, "OEBPS")

	const key, want = "OEBPS/c1.xhtml", "Chapter One"
	if navLabels[key] != want {
		t.Fatalf("nav label = %q, want %q", navLabels[key], want)
	}
	if navLabels[key] != ncxLabels[key] {
		t.Fatalf("one title, two answers: nav %q vs ncx %q", navLabels[key], ncxLabels[key])
	}
}