pdfdoc: page markers, the map for long documents, page ranges, grep

Pure render function on a *pdftext.Doc with no I/O: the whole text with
page markers when it fits, otherwise a map (title, outline with page
numbers, leading pages and how to ask for more), a page range, or grep
results with page numbers. Every length is in runes.

Co-Authored-By: ornith-1.5-35b-a3b <ornith-1.5-35b-a3b@llama.cpp.invalid>

docs/implementer-log.md

@@ -8,3 +8,4 @@ One row per task session that ends in a commit. Newest last.
 | p5a/02-pdftext | 2026-09-25 | done | 1 | gate: ok | none | Built the child-extraction package: `Extract` (reader.Open with ErrWrongPassword→ErrEncrypted, per-page recover, trailer /Info /Title, rule-3 UTF-16BE/Latin-1 decode), `outline.go` (/Dest and /GoTo /A dests, 10k-item and no-double-visit bounds, never-nil outline), `Runner` (re-exec'd child, env-isolated via empty env, 60 s deadline, 64 MiB output cap, 200 MiB stdin cap, stderr capped at 4 KiB, exit-code→error mapping, Doc parse plus ErrNoText), and the `sift pdftext` case in main.go. Removed the tools.go module pin; go.mod/go.sum/vendor/modules.txt confirmed unchanged. All 10 given tests pass under -race -count=3 including the deadline-kill test; make gate ok; given files byte-identical; `printf nope | bin/sift pdftext` exits 4. |
 | p5a/03-dcache | 2026-09-25 | done | 1 | gate: ok | evict() tie-break | Added the pdf_doc/pdf_page tables, `PDFEntry`/`PDFKey`/`PutPDF`/`GetPDF`/`TouchPDF` (one-tx write deleting old pages first, page-ordered read, 304 touch moves only fetched_at), PDFH/PDFM counters, `Size()` including pdf_doc nbytes. One deviation: extended the eviction union with `pdf_doc` and added `rowid ASC` to the ORDER BY so same-second puts evict in insertion order (deterministic whole-document eviction); without it TestPDFEviction is order-dependent. Left Stats() signature unchanged for task 07. Given test byte-identical. |
 | p5a/04-pdfpool | 2026-09-25 | done | 1 | gate: ok | none | Implemented the generic scheduler. `Pool` holds one mutex, a `jobs` map (key→in-flight job), a `callerJobs` map (caller→their job), and a FIFO `waiters` line. `Do` applies the six rules in order: join (counted in `Waiting`, no slot, allowed while the caller holds another job, `fn` may be nil), caller-busy, then slot/queue/refuse. A job's result is broadcast to every sharer via a `done` channel; results are not cached (key forgotten on finish). `fn` runs under `context.WithoutCancel` so the flight outlives a caller that leaves. Panic in `fn` is recovered into an error, freed the slot and caller. One implementation finding (not a plan deviation): queued jobs must not be counted in `Running()`, so `startLocked` only registers a job and a separate `takeSlotLocked`/`grant` increments the slot count — abandoning a queued starter (`abandon`) releases sharers without touching the slot. Given test byte-identical. |
+| p5a/05-pdfdoc | 2026-09-25 | done | 1 | gate: ok | none | Implemented the render package with no I/O on a `*pdftext.Doc`. `Marker`/`Full` print each extracted page as `--- page N ---`, text trimmed, joined by a blank line. `Render` dispatches grep → page range → whole. Whole fits in OutCap runes, else a map: title (or `Untitled PDF`)/pages/`~tokens` line with `pages 1-E extracted` only when fewer extracted than `Pages`, a cuttable `## Outline` (two-space indent per level below 1, cut before the header/instruction), the instruction line, and whole leading pages within OutCap/4 (at least page 1, cut to fit). Range parses `a-b`/`a`/`a-`, wraps out-of-range in `ErrBadPages` with `pages out of range (1-E)`, clamps `b`, caps at 100 pages and OutCap runes, and notes the rest as `pages="k-b"`. Grep is RE2 with smart-case `(?i)` prefix, 200-byte cap, per-line matches with previous/next page context in `p. P-`/`p. P:` form, first 100 shown, total counted; RE2 keeps the `(a+)+$` test linear. Every length is in runes. Given test byte-identical; make gate ok. |

internal/pdfdoc/grep.go

@@ -0,0 +1,153 @@
+package pdfdoc
+
+import (
+	"fmt"
+	"regexp"
+	"strconv"
+	"strings"
+	"unicode/utf8"
+
+	"git.wntrmute.dev/kyle/sift/internal/pdftext"
+)
+
+// maxGrepMatches bounds how many match blocks are shown; the count reported is
+// always the true total.
+const maxGrepMatches = 100
+
+// maxGrepPatternBytes bounds a grep pattern; a longer one is rejected before it
+// can compile.
+const maxGrepPatternBytes = 200
+
+// hit is one match: the 1-based page and the 0-based line index within it.
+type hit struct{ page, line int }
+
+// renderGrep searches every line of every page in the range (all pages when no
+// range is given) and returns the matches with their page numbers.
+func renderGrep(doc *pdftext.Doc, opt Options) (Rendered, error) {
+	pattern := opt.Grep
+	if len(pattern) > maxGrepPatternBytes {
+		return Rendered{}, fmt.Errorf("%w: pattern over %d bytes", ErrBadGrep, maxGrepPatternBytes)
+	}
+	re, err := regexp.Compile(grepPattern(pattern))
+	if err != nil {
+		return Rendered{}, fmt.Errorf("%w: %v", ErrBadGrep, err)
+	}
+
+	start, end := 1, len(doc.Text)
+	if opt.Pages != "" {
+		s, e, err := parsePages(opt.Pages, len(doc.Text))
+		if err != nil {
+			return Rendered{}, err
+		}
+		start, end = s, e
+	}
+
+	var hits []hit
+	linesByPage := make(map[int][]string, end-start+1)
+	for p := start; p <= end; p++ {
+		lines := splitLines(doc.Text[p-1])
+		linesByPage[p] = lines
+		for j := range lines {
+			if re.MatchString(lines[j]) {
+				hits = append(hits, hit{p, j})
+			}
+		}
+	}
+
+	n := len(hits)
+	var header string
+	switch {
+	case n == 0:
+		header = `grep "` + pattern + `": no matches`
+	case n == 1:
+		header = `grep "` + pattern + `": 1 match`
+	default:
+		header = fmt.Sprintf(`grep "%s": %d matches`, pattern, n)
+	}
+	if n > maxGrepMatches {
+		header += ", showing the first " + strconv.Itoa(maxGrepMatches)
+	}
+
+	maxBlocks := n
+	if maxBlocks > maxGrepMatches {
+		maxBlocks = maxGrepMatches
+	}
+
+	outLines := make([]string, 0, 16+len(hits))
+	outLines = append(outLines, header)
+	running := utf8.RuneCountInString(header)
+	shown := 0
+	for i := 0; i < maxBlocks; i++ {
+		block := matchBlock(hits[i], linesByPage)
+		add := utf8.RuneCountInString(strings.Join(block, "\n"))
+		if i == 0 {
+			if running+add > opt.OutCap {
+				break
+			}
+		} else if running+1+add > opt.OutCap {
+			break
+		}
+		outLines = append(outLines, block...)
+		running += add
+		if i > 0 {
+			running++
+		}
+		shown++
+	}
+	content := strings.Join(outLines, "\n")
+	if utf8.RuneCountInString(content) > opt.OutCap {
+		content = truncateRunes(content, opt.OutCap)
+	}
+
+	note := ""
+	if shown < maxBlocks {
+		note = fmt.Sprintf("showing %d of %d matches", shown, n)
+	}
+	return Rendered{Content: content, GrepHits: n, Note: note}, nil
+}
+
+// grepPattern applies smart case: it lowercases-sensitively prefixes "(?i)"
+// unless the pattern already contains an upper-case letter, which asks for a
+// case-sensitive search.
+func grepPattern(pattern string) string {
+	if hasUpper(pattern) {
+		return pattern
+	}
+	return "(?i)" + pattern
+}
+
+// hasUpper reports whether s contains an upper-case ASCII letter.
+func hasUpper(s string) bool {
+	for _, r := range s {
+		if r >= 'A' && r <= 'Z' {
+			return true
+		}
+	}
+	return false
+}
+
+// matchBlock renders one hit as its previous line, the matching line, the next
+// line and a separator. Context lines that do not exist (the match is on the
+// first or last line) are omitted.
+func matchBlock(h hit, linesByPage map[int][]string) []string {
+	lines := linesByPage[h.page]
+	var b []string
+	if h.line > 0 {
+		b = append(b, fmt.Sprintf("p. %d- %s", h.page, strings.TrimSpace(lines[h.line-1])))
+	}
+	b = append(b, fmt.Sprintf("p. %d: %s", h.page, strings.TrimSpace(lines[h.line])))
+	if h.line+1 < len(lines) {
+		b = append(b, fmt.Sprintf("p. %d- %s", h.page, strings.TrimSpace(lines[h.line+1])))
+	}
+	return append(b, "--")
+}
+
+// splitLines splits a page into its lines, dropping the single empty line a
+// trailing newline would otherwise introduce.
+func splitLines(s string) []string {
+	lines := strings.Split(s, "\n")
+	if len(lines) > 1 && lines[len(lines)-1] == "" && strings.HasSuffix(s, "\n") {
+		lines = lines[:len(lines)-1]
+	}
+	return lines
+}

internal/pdfdoc/pdfdoc.go

@@ -0,0 +1,285 @@
+// Package pdfdoc renders an extracted document for an agent: the whole text with
+// page markers when it fits, otherwise a map (title, outline with page numbers,
+// leading pages and how to ask for more), a page range, or grep results with page
+// numbers. It works on a *pdftext.Doc with no I/O; every length is in runes, which
+// is what bounds an agent's context regardless of the byte width of a code point.
+package pdfdoc
+
+import (
+	"errors"
+	"fmt"
+	"strconv"
+	"strings"
+	"unicode/utf8"
+
+	"git.wntrmute.dev/kyle/sift/internal/pdftext"
+)
+
+// The errors Render returns for a malformed Pages or Grep option.
+var (
+	ErrBadPages = errors.New("invalid pages")
+	ErrBadGrep  = errors.New("invalid grep")
+)
+
+// maxPages bounds a range request so a caller cannot ask for the whole of a
+// million-page document at once; the cap below still applies on top of it.
+const maxPages = 100
+
+// Options selects how Render draws from a document.
+type Options struct {
+	Pages  string // "" = no range; "a-b", "a" or "a-" (1-based, inclusive)
+	Grep   string // "" = no grep; an RE2 pattern
+	OutCap int    // characters (runes) the content may use
+}
+
+// Rendered is what an agent sees. When Map is true Content is the map of a
+// document longer than OutCap; a grep answer sets GrepHits and leaves First and
+// Last at zero.
+type Rendered struct {
+	Content     string
+	Map         bool   // Content is the map of a document longer than OutCap
+	First, Last int    // pages whose text is in Content; 0, 0 for a grep answer
+	GrepHits    int    // matches found (more may exist than are shown)
+	Note        string // "" or how the answer was cut and how to continue
+}
+
+// Marker is the header printed before a page's text.
+func Marker(page int) string {
+	return fmt.Sprintf("--- page %d ---", page)
+}
+
+// Full renders every extracted page with its marker, trimmed of surrounding
+// whitespace. It is the unit the map and page ranges are built from.
+func Full(doc *pdftext.Doc) string {
+	parts := make([]string, 0, len(doc.Text))
+	for i := range doc.Text {
+		parts = append(parts, Marker(i+1)+"\n\n"+strings.TrimSpace(doc.Text[i]))
+	}
+	return strings.Join(parts, "\n\n")
+}
+
+// Render dispatches on the option that is set: grep, a page range, or the whole
+// document.
+func Render(doc *pdftext.Doc, opt Options) (Rendered, error) {
+	switch {
+	case opt.Grep != "":
+		return renderGrep(doc, opt)
+	case opt.Pages != "":
+		return renderPages(doc, opt)
+	default:
+		return renderWhole(doc, opt)
+	}
+}
+
+// renderWhole shows the whole document when it fits OutCap, otherwise a map that
+// fits.
+func renderWhole(doc *pdftext.Doc, opt Options) (Rendered, error) {
+	full := Full(doc)
+	if utf8.RuneCountInString(full) <= opt.OutCap {
+		return Rendered{Content: full, First: 1, Last: len(doc.Text)}, nil
+	}
+	return renderMap(doc, opt), nil
+}
+
+// header is the fixed top of a map: the title line and the pages/token line,
+// which notes how many pages were extracted when fewer than the document has.
+func header(doc *pdftext.Doc) string {
+	title := doc.Title
+	if title == "" {
+		title = "Untitled PDF"
+	}
+	pages := fmt.Sprintf("%d pages · ~%d tokens", doc.Pages, tokenEstimate(doc))
+	if len(doc.Text) < doc.Pages {
+		pages += fmt.Sprintf(" · pages 1-%d extracted", len(doc.Text))
+	}
+	return "# " + title + "\n\n" + pages + "\n\n"
+}
+
+// tokenEstimate roughs a document's size in tokens as a quarter of its runes.
+func tokenEstimate(doc *pdftext.Doc) int {
+	return utf8.RuneCountInString(Full(doc)) / 4
+}
+
+// instruction is the fixed line that tells the agent how to ask for more.
+const instruction = "This document is longer than one answer: ask for pages=\"a-b\" or grep=\"pattern\".\n\n"
+
+// renderMap assembles the header, a cuttable outline, the instruction, and whole
+// leading pages that fit in a quarter of the cap.
+func renderMap(doc *pdftext.Doc, opt Options) Rendered {
+	h := header(doc)
+	outlineLines := outlineItems(doc.Outline)
+
+	// The outline is the only flexible part: cut it rather than the header or the
+	// instruction when the document's outline would not fit.
+	pagesBudget := opt.OutCap / 4
+	outlineBudget := opt.OutCap - utf8.RuneCountInString(h) - utf8.RuneCountInString(instruction) - pagesBudget
+	if outlineBudget < 0 {
+		outlineBudget = 0
+	}
+	var kept []string
+	running := 0
+	for _, line := range outlineLines {
+		line += "\n"
+		if running+utf8.RuneCountInString(line) > outlineBudget {
+			break
+		}
+		running += utf8.RuneCountInString(line)
+		kept = append(kept, strings.TrimRight(line, "\n"))
+	}
+	outlineSection := ""
+	if len(kept) > 0 {
+		outlineSection = "## Outline\n" + strings.Join(kept, "\n") + "\n\n"
+	}
+
+	pages := leadingPages(doc, opt.OutCap-utf8.RuneCountInString(h)-utf8.RuneCountInString(outlineSection)-utf8.RuneCountInString(instruction), pagesBudget)
+	content := h + outlineSection + instruction + pages.text
+	if utf8.RuneCountInString(content) > opt.OutCap {
+		content = truncateRunes(content, opt.OutCap)
+	}
+	return Rendered{Content: content, Map: true, First: 1, Last: pages.last}
+}
+
+// outlineItems renders each outline item as "- <indent><title> (p. <page>)", two
+// spaces of indent per level below the top.
+func outlineItems(items []pdftext.OutlineItem) []string {
+	lines := make([]string, 0, len(items))
+	for _, it := range items {
+		indent := strings.Repeat(" ", max(0, (it.Level-1)*2))
+		lines = append(lines, indent+"- "+it.Title+" (p. "+strconv.Itoa(it.Page)+")")
+	}
+	return lines
+}
+
+// pagesResult is a run of whole pages plus the last one shown.
+type pagesResult struct {
+	text string
+	last int
+}
+
+// leadingPages renders whole pages from page 1 into budget runes, at least one,
+// cutting the first to fit if it is alone larger than the budget.
+func leadingPages(doc *pdftext.Doc, budget, pagesBudget int) pagesResult {
+	if budget < 0 {
+		budget = 0
+	}
+	if pagesBudget > budget {
+		pagesBudget = budget
+	}
+	extracted := len(doc.Text)
+	var parts []string
+	running := 0
+	last := 0
+	for p := 1; p <= extracted; p++ {
+		part := Marker(p) + "\n\n" + strings.TrimSpace(doc.Text[p-1])
+		if p == 1 && utf8.RuneCountInString(part) > pagesBudget {
+			parts = append(parts, truncateRunes(part, pagesBudget))
+			return pagesResult{text: strings.Join(parts, "\n\n"), last: 1}
+		}
+		if running+utf8.RuneCountInString(part) > pagesBudget {
+			break
+		}
+		running += utf8.RuneCountInString(part)
+		parts = append(parts, part)
+		last = p
+	}
+	return pagesResult{text: strings.Join(parts, "\n\n"), last: last}
+}
+
+// renderPages renders a page range as whole pages, capped at maxPages pages and
+// at OutCap runes, noting how to ask for the rest.
+func renderPages(doc *pdftext.Doc, opt Options) (Rendered, error) {
+	start, end, err := parsePages(opt.Pages, len(doc.Text))
+	if err != nil {
+		return Rendered{}, err
+	}
+
+	limit := end
+	if end-start+1 > maxPages {
+		limit = start + maxPages - 1
+	}
+
+	var parts []string
+	running := 0
+	last := start - 1
+	note := ""
+	for p := start; p <= limit; p++ {
+		part := Marker(p) + "\n\n" + strings.TrimSpace(doc.Text[p-1])
+		if p == start && utf8.RuneCountInString(part) > opt.OutCap {
+			parts = append(parts, truncateRunes(part, opt.OutCap))
+			note = fmt.Sprintf("page %d cut to fit the answer", p)
+			last = p
+			break
+		}
+		if running+utf8.RuneCountInString(part) > opt.OutCap {
+			break
+		}
+		running += utf8.RuneCountInString(part)
+		parts = append(parts, part)
+		last = p
+	}
+	if last < end {
+		note = fmt.Sprintf("cut at page %d; ask for pages=\"%d-%d\"", last, last+1, end)
+	}
+	return Rendered{Content: strings.Join(parts, "\n\n"), First: start, Last: last, Note: note}, nil
+}
+
+// parsePages parses a "a-b", "a" or "a-" spec against the number of extracted
+// pages. It returns a range error wrapping ErrBadPages when the numbers are out
+// of range, and a format error wrapping ErrBadPages otherwise.
+func parsePages(spec string, extracted int) (start, end int, err error) {
+	s := strings.TrimSpace(spec)
+	start, end, endToExtracted, ok := splitSpec(s)
+	if !ok {
+		return 0, 0, fmt.Errorf("%w: invalid pages", ErrBadPages)
+	}
+	if endToExtracted {
+		end = extracted
+	}
+	if start < 1 || end < start || start > extracted {
+		return 0, 0, fmt.Errorf("%w: pages out of range (1-%d)", ErrBadPages, extracted)
+	}
+	if end > extracted {
+		end = extracted
+	}
+	return start, end, nil
+}
+
+// splitSpec turns a spec into its numbers. It reports ok=false for a malformed
+// spec so the caller returns a format error. A bare "a" is a single page; "a-"
+// spans to the last extracted page.
+func splitSpec(s string) (start, end int, endToExtracted, ok bool) {
+	if !strings.Contains(s, "-") {
+		n, err := strconv.Atoi(s)
+		if err != nil {
+			return 0, 0, false, false
+		}
+		return n, n, false, true
+	}
+	parts := strings.Split(s, "-")
+	if len(parts) != 2 || parts[0] == "" {
+		return 0, 0, false, false
+	}
+	a, err := strconv.Atoi(strings.TrimSpace(parts[0]))
+	if err != nil {
+		return 0, 0, false, false
+	}
+	if strings.TrimSpace(parts[1]) == "" {
+		return a, 0, true, true
+	}
+	b, err := strconv.Atoi(strings.TrimSpace(parts[1]))
+	if err != nil {
+		return 0, 0, false, false
+	}
+	return a, b, false, true
+}
+
+// truncateRunes returns the first n runes of s, never splitting a multi-byte one.
+func truncateRunes(s string, n int) string {
+	if n <= 0 {
+		return ""
+	}
+	if utf8.RuneCountInString(s) <= n {
+		return s
+	}
+	return string([]rune(s)[:n])
+}

internal/pdfdoc/pdfdoc_test.go

@@ -0,0 +1,249 @@
+package pdfdoc_test
+
+// P5a task 05: rendering an extracted document for an agent — the whole text with page markers,
+// the map for a long document, a page range, and grep.
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+	"testing"
+	"time"
+	"unicode/utf8"
+
+	"git.wntrmute.dev/kyle/sift/internal/pdfdoc"
+	"git.wntrmute.dev/kyle/sift/internal/pdftext"
+)
+
+func small() *pdftext.Doc {
+	return &pdftext.Doc{
+		Meta: pdftext.Meta{Pages: 3, Title: "Three Pages", Outline: []pdftext.OutlineItem{{Title: "1 Intro", Page: 1, Level: 1}, {Title: "2 Fox", Page: 2, Level: 1}, {Title: "2.1 Den", Page: 2, Level: 2}}},
+		Text: []string{"Chapter one begins here.\nSecond line.\n", "  The quick brown fox is on page two.\nIt jumps.", "Last page."},
+	}
+}
+
+// big returns n pages of about 1,000 characters each, every line naming its page.
+func big(n int) *pdftext.Doc {
+	d := &pdftext.Doc{Meta: pdftext.Meta{Pages: n, Title: "Long Thesis", Outline: []pdftext.OutlineItem{{Title: "1 Introduction", Page: 1, Level: 1}, {Title: "2 Method", Page: 12, Level: 1}, {Title: "2.3 Setup", Page: 15, Level: 2}}}}
+	for p := 1; p <= n; p++ {
+		var lines []string
+		for l := 1; l <= 12; l++ {
+			lines = append(lines, fmt.Sprintf("page %d line %d lorem ipsum dolor sit amet consectetur adipiscing elit sed do eiusmod", p, l))
+		}
+		if p == 37 {
+			lines[4] = "the Needle is here on page 37"
+		}
+		d.Text = append(d.Text, strings.Join(lines, "\n"))
+	}
+	return d
+}
+
+func TestFull(t *testing.T) {
+	want := "--- page 1 ---\n\nChapter one begins here.\nSecond line.\n\n--- page 2 ---\n\nThe quick brown fox is on page two.\nIt jumps.\n\n--- page 3 ---\n\nLast page."
+	if got := pdfdoc.Full(small()); got != want {
+		t.Errorf("Full =\n%q\nwant\n%q", got, want)
+	}
+	if pdfdoc.Marker(12) != "--- page 12 ---" {
+		t.Errorf("Marker(12) = %q", pdfdoc.Marker(12))
+	}
+}
+
+func TestWholeDocumentWhenItFits(t *testing.T) {
+	r, err := pdfdoc.Render(small(), pdfdoc.Options{OutCap: 10000})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if r.Map || r.First != 1 || r.Last != 3 || r.Content != pdfdoc.Full(small()) {
+		t.Errorf("got %+v", r)
+	}
+}
+
+func TestMapForALongDocument(t *testing.T) {
+	d := big(60) // ~60k characters
+	r, err := pdfdoc.Render(d, pdfdoc.Options{OutCap: 20000})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !r.Map || r.First != 1 || r.Last < 1 {
+		t.Fatalf("got Map=%v First=%d Last=%d, want a map from page 1", r.Map, r.First, r.Last)
+	}
+	c := r.Content
+	for _, want := range []string{
+		"# Long Thesis\n",
+		"60 pages · ~",
+		"## Outline\n",
+		"- 1 Introduction (p. 1)\n",
+		"- 2 Method (p. 12)\n",
+		"  - 2.3 Setup (p. 15)\n", // level 2 is indented two spaces
+		`pages="a-b"`,
+		`grep="pattern"`,
+		"--- page 1 ---",
+	} {
+		if !strings.Contains(c, want) {
+			t.Errorf("map lacks %q", want)
+		}
+	}
+	if utf8.RuneCountInString(c) > 20000 {
+		t.Errorf("map is %d characters, over OutCap 20000", utf8.RuneCountInString(c))
+	}
+	// Leading pages take at most a quarter of the cap, and only whole pages.
+	if strings.Contains(c, fmt.Sprintf("--- page %d ---", r.Last+1)) {
+		t.Errorf("content has page %d past Last %d", r.Last+1, r.Last)
+	}
+	if !strings.Contains(c, fmt.Sprintf("page %d line 12", r.Last)) {
+		t.Errorf("the last page shown (%d) is cut; pages in a map are whole", r.Last)
+	}
+	idx := strings.Index(c, "--- page 1 ---")
+	if pagesPart := c[idx:]; utf8.RuneCountInString(pagesPart) > 20000/4 {
+		t.Errorf("leading pages take %d characters, over OutCap/4", utf8.RuneCountInString(pagesPart))
+	}
+}
+
+func TestMapNotesPartialExtraction(t *testing.T) {
+	d := big(30)
+	d.Pages = 2500 // the child stopped at its page cap
+	r, _ := pdfdoc.Render(d, pdfdoc.Options{OutCap: 5000})
+	if !strings.Contains(r.Content, "2500 pages") || !strings.Contains(r.Content, "pages 1-30 extracted") {
+		t.Errorf("map must say how much was extracted:\n%s", r.Content[:min(600, len(r.Content))])
+	}
+}
+
+func TestUntitledAndNoOutline(t *testing.T) {
+	d := big(40)
+	d.Title, d.Outline = "", nil
+	r, _ := pdfdoc.Render(d, pdfdoc.Options{OutCap: 5000})
+	if !strings.HasPrefix(r.Content, "# Untitled PDF\n") || strings.Contains(r.Content, "## Outline") {
+		t.Errorf("untitled, outline-less map:\n%s", r.Content[:min(400, len(r.Content))])
+	}
+}
+
+func TestPages(t *testing.T) {
+	d := big(60)
+	for _, tc := range []struct {
+		spec        string
+		first, last int
+	}{
+		{"2-3", 2, 3},
+		{"7", 7, 7},
+		{" 58- ", 58, 60},
+		{"60-60", 60, 60},
+	} {
+		r, err := pdfdoc.Render(d, pdfdoc.Options{Pages: tc.spec, OutCap: 200000})
+		if err != nil {
+			t.Errorf("%q: %v", tc.spec, err)
+			continue
+		}
+		if r.Map || r.First != tc.first || r.Last != tc.last {
+			t.Errorf("%q: First %d Last %d Map %v, want %d-%d", tc.spec, r.First, r.Last, r.Map, tc.first, tc.last)
+		}
+		if !strings.HasPrefix(r.Content, pdfdoc.Marker(tc.first)+"\n\n") || strings.Contains(r.Content, pdfdoc.Marker(tc.last+1)) || (tc.first > 1 && strings.Contains(r.Content, pdfdoc.Marker(tc.first-1)+"\n")) {
+			t.Errorf("%q: wrong pages in content: %.120q…", tc.spec, r.Content)
+		}
+	}
+	for _, bad := range []string{"0-1", "5-4", "61", "999", "a-b", "1-2-3", "", "-5"} {
+		if bad == "" {
+			continue // empty means "no range", tested above
+		}
+		_, err := pdfdoc.Render(d, pdfdoc.Options{Pages: bad, OutCap: 200000})
+		if !errors.Is(err, pdfdoc.ErrBadPages) {
+			t.Errorf("%q: %v, want ErrBadPages", bad, err)
+		}
+	}
+	_, err := pdfdoc.Render(d, pdfdoc.Options{Pages: "61", OutCap: 200000})
+	if err == nil || !strings.Contains(err.Error(), "pages out of range (1-60)") {
+		t.Errorf("out-of-range message: %v", err)
+	}
+}
+
+// A range is at most 100 pages and at most OutCap characters, cut at a whole page with a note
+// that says how to continue.
+func TestPagesAreCapped(t *testing.T) {
+	d := big(300)
+	r, err := pdfdoc.Render(d, pdfdoc.Options{Pages: "1-300", OutCap: 1 << 30})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if r.Last != 100 || !strings.Contains(r.Note, `pages="101-300"`) {
+		t.Errorf("1-300: Last %d note %q; want 100 and a pointer to 101-300", r.Last, r.Note)
+	}
+	r, _ = pdfdoc.Render(d, pdfdoc.Options{Pages: "10-40", OutCap: 5000})
+	if utf8.RuneCountInString(r.Content) > 5000 || r.First != 10 || r.Last < 10 || r.Last >= 40 {
+		t.Errorf("10-40 under a 5000 cap: First %d Last %d, %d chars", r.First, r.Last, utf8.RuneCountInString(r.Content))
+	}
+	if !strings.Contains(r.Note, fmt.Sprintf(`pages="%d-40"`, r.Last+1)) {
+		t.Errorf("note %q must point at the rest", r.Note)
+	}
+	// One page bigger than the cap is truncated rather than dropped.
+	huge := &pdftext.Doc{Meta: pdftext.Meta{Pages: 1}, Text: []string{strings.Repeat("w ", 10000)}}
+	r, _ = pdfdoc.Render(huge, pdfdoc.Options{Pages: "1", OutCap: 1000})
+	if r.First != 1 || r.Last != 1 || utf8.RuneCountInString(r.Content) > 1000 || r.Note == "" {
+		t.Errorf("oversized single page: %+v (%d chars)", r, utf8.RuneCountInString(r.Content))
+	}
+}
+
+func TestGrep(t *testing.T) {
+	d := big(60)
+	r, err := pdfdoc.Render(d, pdfdoc.Options{Grep: "needle", OutCap: 200000})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if r.GrepHits != 1 || r.First != 0 || r.Map {
+		t.Errorf("got %+v", r)
+	}
+	for _, want := range []string{
+		`grep "needle": 1 match`,
+		"p. 37- page 37 line 4",
+		"p. 37: the Needle is here on page 37",
+		"p. 37- page 37 line 6",
+	} {
+		if !strings.Contains(r.Content, want) {
+			t.Errorf("grep output lacks %q:\n%s", want, r.Content)
+		}
+	}
+	// Smart case: an upper-case letter makes the pattern case-sensitive.
+	if r, _ := pdfdoc.Render(d, pdfdoc.Options{Grep: "NEEDLE", OutCap: 200000}); r.GrepHits != 0 {
+		t.Errorf("NEEDLE matched %d times, want 0 (case-sensitive)", r.GrepHits)
+	}
+	if r, _ := pdfdoc.Render(d, pdfdoc.Options{Grep: "Needle", OutCap: 200000}); r.GrepHits != 1 {
+		t.Errorf("Needle matched %d times, want 1", r.GrepHits)
+	}
+	// No match is an answer, not an error.
+	r, err = pdfdoc.Render(d, pdfdoc.Options{Grep: "zebra", OutCap: 200000})
+	if err != nil || r.GrepHits != 0 || !strings.Contains(r.Content, `grep "zebra": no matches`) {
+		t.Errorf("no match: %+v %v", r, err)
+	}
+	// Within a page range.
+	r, _ = pdfdoc.Render(d, pdfdoc.Options{Grep: "line 3 ", Pages: "5-6", OutCap: 200000})
+	if r.GrepHits != 2 || strings.Contains(r.Content, "p. 7") {
+		t.Errorf("grep in 5-6: %d hits\n%s", r.GrepHits, r.Content)
+	}
+}
+
+func TestGrepLimits(t *testing.T) {
+	d := big(200)
+	r, err := pdfdoc.Render(d, pdfdoc.Options{Grep: "lorem", OutCap: 1 << 30}) // 2,399 matches: every line but the needle
+	if err != nil {
+		t.Fatal(err)
+	}
+	if r.GrepHits != 2399 || strings.Count(r.Content, ": page ") != 100 || !strings.Contains(r.Content, "showing the first 100") {
+		t.Errorf("hits %d, shown %d", r.GrepHits, strings.Count(r.Content, ": page "))
+	}
+	if _, err := pdfdoc.Render(d, pdfdoc.Options{Grep: strings.Repeat("a", 201), OutCap: 1000}); !errors.Is(err, pdfdoc.ErrBadGrep) {
+		t.Errorf("201-byte pattern: %v, want ErrBadGrep", err)
+	}
+	if _, err := pdfdoc.Render(d, pdfdoc.Options{Grep: "(unclosed", OutCap: 1000}); !errors.Is(err, pdfdoc.ErrBadGrep) {
+		t.Errorf("bad regexp: %v, want ErrBadGrep", err)
+	}
+	// RE2 is linear: the classic catastrophic pattern over a 1 MB page finishes quickly.
+	evil := &pdftext.Doc{Meta: pdftext.Meta{Pages: 1}, Text: []string{strings.Repeat("a", 1<<20) + "!"}}
+	start := time.Now()
+	_, _ = pdfdoc.Render(evil, pdfdoc.Options{Grep: "(a+)+$", OutCap: 1000})
+	if d := time.Since(start); d > time.Second {
+		t.Errorf("(a+)+$ took %v", d)
+	}
+	// Output stays under the cap.
+	r, _ = pdfdoc.Render(d, pdfdoc.Options{Grep: "lorem", OutCap: 3000})
+	if utf8.RuneCountInString(r.Content) > 3000 {
+		t.Errorf("grep output %d characters over a 3000 cap", utf8.RuneCountInString(r.Content))
+	}
+}