mirror of
https://github.com/Stirling-Tools/Stirling-PDF.git
synced 2026-09-03 05:10:16 +03:00
Merge remote-tracking branch 'origin/main' into fix/sanitize-empty-aa-dictionary
This commit is contained in:
Anthony Stirling
-341
@@ -1,341 +0,0 @@
|
||||
package stirling.software.proprietary.service.ua;
|
||||
|
||||
import static org.junit.jupiter.api.Assertions.assertTrue;
|
||||
|
||||
import java.io.ByteArrayOutputStream;
|
||||
import java.io.IOException;
|
||||
import java.util.ArrayList;
|
||||
import java.util.List;
|
||||
import java.util.Locale;
|
||||
|
||||
import org.apache.pdfbox.Loader;
|
||||
import org.apache.pdfbox.pdmodel.PDDocument;
|
||||
import org.apache.pdfbox.pdmodel.PDPage;
|
||||
import org.apache.pdfbox.pdmodel.PDPageContentStream;
|
||||
import org.apache.pdfbox.pdmodel.common.PDRectangle;
|
||||
import org.apache.pdfbox.pdmodel.font.PDFont;
|
||||
import org.junit.jupiter.api.BeforeAll;
|
||||
import org.junit.jupiter.api.DisplayName;
|
||||
import org.junit.jupiter.api.Test;
|
||||
|
||||
import stirling.software.common.service.CustomPDFDocumentFactory;
|
||||
import stirling.software.common.service.PdfMetadataService;
|
||||
import stirling.software.proprietary.pdf.ua.DocumentStructure;
|
||||
import stirling.software.proprietary.pdf.ua.LayoutAnalyzer;
|
||||
import stirling.software.proprietary.pdf.ua.PageContent;
|
||||
import stirling.software.proprietary.pdf.ua.PdfUaProfile;
|
||||
import stirling.software.proprietary.pdf.ua.PdfUaTagger;
|
||||
import stirling.software.proprietary.pdf.ua.TaggedContentExtractor;
|
||||
import stirling.software.proprietary.pdf.ua.TaggingOptions;
|
||||
|
||||
/**
|
||||
* Measures where conversion time and memory go; meant to be read, not to gate CI. Assertions catch
|
||||
* only order-of-magnitude regressions - wall-clock numbers are no contract.
|
||||
*/
|
||||
class PdfUaBenchmarkTest {
|
||||
|
||||
private static PdfUaConversionService service;
|
||||
private static PdfUaValidationService validation;
|
||||
|
||||
@BeforeAll
|
||||
static void setUp() {
|
||||
validation = new PdfUaValidationService();
|
||||
validation.initialise();
|
||||
service =
|
||||
new PdfUaConversionService(
|
||||
validation,
|
||||
new FontEmbeddingService(),
|
||||
new CustomPDFDocumentFactory(
|
||||
org.mockito.Mockito.mock(PdfMetadataService.class)));
|
||||
}
|
||||
|
||||
/** A realistic page: heading, prose, a small table, a bullet list. */
|
||||
private static byte[] document(int pages) throws IOException {
|
||||
try (PDDocument document = new PDDocument()) {
|
||||
PDFont font = null;
|
||||
for (int p = 0; p < pages; p++) {
|
||||
PDPage page = new PDPage(PDRectangle.A4);
|
||||
document.addPage(page);
|
||||
if (font == null) {
|
||||
font = PdfUaTestDocuments.font(document);
|
||||
}
|
||||
try (PDPageContentStream cs = new PDPageContentStream(document, page)) {
|
||||
float y = 790;
|
||||
write(cs, font, 9, 60, 810, "Benchmark Corpus Running Head");
|
||||
write(cs, font, 18, 60, y, "Section " + (p + 1));
|
||||
y -= 30;
|
||||
for (int line = 0; line < 22; line++) {
|
||||
write(
|
||||
cs,
|
||||
font,
|
||||
11,
|
||||
60,
|
||||
y,
|
||||
"Body line " + line + " of section " + (p + 1) + " with prose.");
|
||||
y -= 15;
|
||||
}
|
||||
for (int row = 0; row < 4; row++) {
|
||||
cs.beginText();
|
||||
cs.setFont(font, 11);
|
||||
cs.newLineAtOffset(60, y);
|
||||
cs.showText("Row " + row);
|
||||
cs.newLineAtOffset(160, 0);
|
||||
cs.showText(String.valueOf(row * 120));
|
||||
cs.newLineAtOffset(140, 0);
|
||||
cs.showText(String.valueOf(row * 480));
|
||||
cs.endText();
|
||||
y -= 16;
|
||||
}
|
||||
write(cs, font, 11, 60, y - 10, "• First bullet point");
|
||||
write(cs, font, 11, 60, y - 25, "• Second bullet point");
|
||||
write(cs, font, 9, 300, 30, "Page " + (p + 1));
|
||||
}
|
||||
}
|
||||
ByteArrayOutputStream out = new ByteArrayOutputStream();
|
||||
document.save(out);
|
||||
return out.toByteArray();
|
||||
}
|
||||
}
|
||||
|
||||
private static void write(
|
||||
PDPageContentStream cs, PDFont font, float size, float x, float y, String text)
|
||||
throws IOException {
|
||||
cs.beginText();
|
||||
cs.setFont(font, size);
|
||||
cs.newLineAtOffset(x, y);
|
||||
cs.showText(text);
|
||||
cs.endText();
|
||||
}
|
||||
|
||||
private static TaggingOptions options() {
|
||||
return TaggingOptions.builder()
|
||||
.profile(PdfUaProfile.UA1)
|
||||
.language("en-GB")
|
||||
.title("Benchmark")
|
||||
.embedFonts(false)
|
||||
.existingTags(TaggingOptions.ExistingTags.REBUILD)
|
||||
.build();
|
||||
}
|
||||
|
||||
private static long usedHeap() {
|
||||
Runtime runtime = Runtime.getRuntime();
|
||||
System.gc();
|
||||
return runtime.totalMemory() - runtime.freeMemory();
|
||||
}
|
||||
|
||||
@Test
|
||||
@DisplayName("reports throughput and memory across document sizes")
|
||||
void throughputAcrossSizes() throws Exception {
|
||||
int[] sizes = {1, 10, 50, 150};
|
||||
StringBuilder report =
|
||||
new StringBuilder("\nPDF/UA conversion throughput\n")
|
||||
.append(
|
||||
String.format(
|
||||
" %-7s %-10s %-12s %-12s %-10s %s%n",
|
||||
"pages",
|
||||
"input",
|
||||
"convert ms",
|
||||
"ms/page",
|
||||
"pages/s",
|
||||
"heap MB"));
|
||||
|
||||
// Warm up so the first timed run is not measuring class loading and JIT.
|
||||
service.convert(document(5), options());
|
||||
|
||||
for (int pages : sizes) {
|
||||
byte[] input = document(pages);
|
||||
long heapBefore = usedHeap();
|
||||
long start = System.nanoTime();
|
||||
var outcome = service.convert(input, options());
|
||||
long elapsedMs = (System.nanoTime() - start) / 1_000_000;
|
||||
long heapDelta = (usedHeap() - heapBefore) / (1024 * 1024);
|
||||
|
||||
assertTrue(outcome.pdfBytes().length > 0);
|
||||
report.append(
|
||||
String.format(
|
||||
Locale.ROOT,
|
||||
" %-7d %-10s %-12d %-12.2f %-10.1f %d%n",
|
||||
pages,
|
||||
humanBytes(input.length),
|
||||
elapsedMs,
|
||||
elapsedMs / (double) pages,
|
||||
pages * 1000.0 / Math.max(elapsedMs, 1),
|
||||
Math.max(heapDelta, 0)));
|
||||
}
|
||||
System.out.println(report);
|
||||
}
|
||||
|
||||
@Test
|
||||
@DisplayName("breaks conversion down by phase so optimisation has a target")
|
||||
void phaseBreakdown() throws Exception {
|
||||
byte[] input = document(60);
|
||||
|
||||
// Warm up.
|
||||
try (PDDocument warm = Loader.loadPDF(input)) {
|
||||
new TaggedContentExtractor().extract(warm);
|
||||
}
|
||||
|
||||
long parseMs;
|
||||
long extractMs;
|
||||
long analyseMs;
|
||||
long tagMs;
|
||||
List<PageContent> pages;
|
||||
DocumentStructure structure;
|
||||
|
||||
long t0 = System.nanoTime();
|
||||
try (PDDocument document = Loader.loadPDF(input)) {
|
||||
parseMs = ms(t0);
|
||||
|
||||
long t1 = System.nanoTime();
|
||||
pages = new TaggedContentExtractor().extract(document);
|
||||
extractMs = ms(t1);
|
||||
|
||||
long t2 = System.nanoTime();
|
||||
structure = new LayoutAnalyzer().analyse(pages);
|
||||
analyseMs = ms(t2);
|
||||
}
|
||||
|
||||
long t3 = System.nanoTime();
|
||||
try (PDDocument document = Loader.loadPDF(input)) {
|
||||
new PdfUaTagger().tag(document, options());
|
||||
ByteArrayOutputStream out = new ByteArrayOutputStream();
|
||||
document.save(out);
|
||||
}
|
||||
tagMs = ms(t3);
|
||||
|
||||
long t4 = System.nanoTime();
|
||||
var outcome = service.convert(input, options());
|
||||
long totalMs = ms(t4);
|
||||
|
||||
long t5 = System.nanoTime();
|
||||
validation.validate(outcome.pdfBytes(), PdfUaProfile.UA1);
|
||||
long validateMs = ms(t5);
|
||||
|
||||
System.out.printf(
|
||||
Locale.ROOT,
|
||||
"%nPhase breakdown over %d pages (%d blocks)%n"
|
||||
+ " parse %5d ms%n"
|
||||
+ " extract %5d ms (text pass + token scan)%n"
|
||||
+ " analyse %5d ms%n"
|
||||
+ " tag end-to-end %5d ms (includes parse, extract, analyse, inject, write)%n"
|
||||
+ " validate %5d ms (veraPDF)%n"
|
||||
+ " full convert %5d ms (tag + declare + validate)%n",
|
||||
60,
|
||||
structure.getBlocks().size(),
|
||||
parseMs,
|
||||
extractMs,
|
||||
analyseMs,
|
||||
tagMs,
|
||||
validateMs,
|
||||
totalMs);
|
||||
|
||||
assertTrue(pages.size() == 60, "extractor lost pages");
|
||||
}
|
||||
|
||||
@Test
|
||||
@DisplayName("splits the tagging pass into its own sub-phases")
|
||||
void taggingSubPhases() throws Exception {
|
||||
byte[] input = document(60);
|
||||
try (PDDocument warm = Loader.loadPDF(input)) {
|
||||
new TaggedContentExtractor().extract(warm);
|
||||
}
|
||||
|
||||
long extractMs;
|
||||
long analyseMs;
|
||||
long injectMs;
|
||||
long treeMs;
|
||||
long saveMs;
|
||||
|
||||
try (PDDocument document = Loader.loadPDF(input)) {
|
||||
long t = System.nanoTime();
|
||||
List<PageContent> pages = new TaggedContentExtractor().extract(document);
|
||||
extractMs = ms(t);
|
||||
|
||||
t = System.nanoTime();
|
||||
DocumentStructure structure = new LayoutAnalyzer().analyse(pages);
|
||||
analyseMs = ms(t);
|
||||
|
||||
t = System.nanoTime();
|
||||
var injector = new stirling.software.proprietary.pdf.ua.MarkedContentInjector();
|
||||
var byPage =
|
||||
new java.util.LinkedHashMap<
|
||||
Integer, List<stirling.software.proprietary.pdf.ua.StructBlock>>();
|
||||
structure
|
||||
.getBlocks()
|
||||
.forEach(
|
||||
b ->
|
||||
byPage.computeIfAbsent(b.getPageIndex(), k -> new ArrayList<>())
|
||||
.add(b));
|
||||
for (int p = 0; p < document.getNumberOfPages(); p++) {
|
||||
injector.inject(
|
||||
document, document.getPage(p), byPage.getOrDefault(p, List.of()), 0, true);
|
||||
}
|
||||
injectMs = ms(t);
|
||||
|
||||
t = System.nanoTime();
|
||||
new stirling.software.proprietary.pdf.ua.StructTreeWriter()
|
||||
.write(document, structure, PdfUaProfile.UA1);
|
||||
treeMs = ms(t);
|
||||
|
||||
t = System.nanoTime();
|
||||
ByteArrayOutputStream out = new ByteArrayOutputStream();
|
||||
document.save(out);
|
||||
saveMs = ms(t);
|
||||
}
|
||||
|
||||
System.out.printf(
|
||||
Locale.ROOT,
|
||||
"%nTagging sub-phases over 60 pages%n"
|
||||
+ " extract %5d ms%n"
|
||||
+ " analyse %5d ms%n"
|
||||
+ " inject %5d ms%n"
|
||||
+ " struct tree %5d ms%n"
|
||||
+ " save %5d ms%n",
|
||||
extractMs,
|
||||
analyseMs,
|
||||
injectMs,
|
||||
treeMs,
|
||||
saveMs);
|
||||
}
|
||||
|
||||
@Test
|
||||
@DisplayName("memory stays proportional to document size, not quadratic")
|
||||
void memoryScales() throws Exception {
|
||||
List<String> rows = new ArrayList<>();
|
||||
long previousPerPage = 0;
|
||||
boolean blewUp = false;
|
||||
|
||||
for (int pages : new int[] {20, 80, 200}) {
|
||||
byte[] input = document(pages);
|
||||
long before = usedHeap();
|
||||
var outcome = service.convert(input, options());
|
||||
long after = usedHeap();
|
||||
long perPageKb = Math.max(after - before, 0) / 1024 / pages;
|
||||
rows.add(
|
||||
String.format(
|
||||
Locale.ROOT,
|
||||
" %-6d pages in %-9s out %-9s ~%d KB/page retained",
|
||||
pages,
|
||||
humanBytes(input.length),
|
||||
humanBytes(outcome.pdfBytes().length),
|
||||
perPageKb));
|
||||
// Per-page cost should stay roughly flat; a big jump means something accumulates.
|
||||
if (previousPerPage > 0 && perPageKb > previousPerPage * 4 && perPageKb > 200) {
|
||||
blewUp = true;
|
||||
}
|
||||
previousPerPage = Math.max(perPageKb, 1);
|
||||
}
|
||||
System.out.println("\nMemory scaling\n" + String.join("\n", rows));
|
||||
assertTrue(!blewUp, "per-page memory grew superlinearly: " + rows);
|
||||
}
|
||||
|
||||
private static long ms(long startNanos) {
|
||||
return (System.nanoTime() - startNanos) / 1_000_000;
|
||||
}
|
||||
|
||||
private static String humanBytes(int bytes) {
|
||||
return bytes < 1024 * 1024
|
||||
? (bytes / 1024) + " KB"
|
||||
: String.format(Locale.ROOT, "%.1f MB", bytes / 1024.0 / 1024.0);
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user