pdf-accessibility-checker

# Download installer
curl -L -o verapdf.zip https://software.verapdf.org/releases/1.30/verapdf-greenfield-1.30.2-installer.zip
unzip verapdf.zip

# The zip ships an IzPack installer, not a ready-to-run binary
verapdf-greenfield-1.30.2/verapdf-install ./auto-install.xml
<install-dir>/bin/verapdf --version

verapdf --flavour ua1 --format json out.pdf > vera-report.json

import json, subprocess

def test_pdf_passes_pdf_ua_1():
    result = subprocess.run(
        ["verapdf", "--flavour", "ua1", "--format", "json", "out.pdf"],
        capture_output=True, text=True,
    )
    report = json.loads(result.stdout)

    # Iterate jobs[0].validationResult
    job = report["report"]["jobs"][0]
    failed = job["validationResult"]["details"]["failedRules"]

    assert failed == 0, f"PDF/UA validation failed: {job['validationResult']['details']}"

import pikepdf

def test_pdf_is_tagged():
    pdf = pikepdf.open("out.pdf")
    catalog = pdf.Root
    assert "/StructTreeRoot" in catalog, "PDF lacks StructTreeRoot (untagged)"
    mark_info = catalog.get("/MarkInfo", {})
    # Use bool(), not `is True`: pikepdf can return a pikepdf.Object wrapper
    # rather than the singleton Python True, so `is True` is always False
    # even when /Marked is set (per [pikepdf objects]).
    assert bool(mark_info.get("/Marked", False)), "PDF MarkInfo /Marked is false"

def test_pdf_has_title_and_lang():
    pdf = pikepdf.open("out.pdf")
    info = pdf.docinfo

    title = info.get("/Title")
    assert title and str(title).strip(), "PDF has no /Info /Title"

    lang = pdf.Root.get("/Lang")
    assert lang and str(lang) in ("en", "en-US", "en-GB", "de", "fr"), f"PDF /Lang invalid: {lang}"

def test_all_images_have_alt():
    pdf = pikepdf.open("out.pdf")
    structure_root = pdf.Root["/StructTreeRoot"]

    # The structure tree is NESTED: each element's /K can hold child
    # elements to arbitrary depth, so /Figure tags live below /Document,
    # /Sect, /P, etc. A top-level-only scan misses nested figures and
    # passes vacuously. Walk every /K recursively. Each /Figure must carry
    # an /Alt (or /ActualText) per [WCAG PDF1].
    untagged_images = []

    def walk(node):
        if isinstance(node, pikepdf.Array):
            for child in node:
                walk(child)
            return
        if not isinstance(node, pikepdf.Dictionary):
            return  # marked-content id (int) or other leaf
        if node.get("/S") == "/Figure":
            alt = node.get("/Alt")
            if not alt or not str(alt).strip():
                untagged_images.append(node)
        kids = node.get("/K")
        if kids is not None:
            walk(kids)

    walk(structure_root.get("/K"))
    assert untagged_images == [], f"Images without /Alt: {len(untagged_images)}"

def test_reading_order_matches_visual_order():
    # TEMPLATE: the title and section labels below are placeholders for an
    # invoice layout. Replace them with the expected reading-order landmarks
    # for your own document before using this test.
    text_per_page = extract_text_with_structure_order("out.pdf")

    # First page should start with the document's leading landmark.
    assert text_per_page[0].startswith("Invoice #")  # replace per document
    # Each expected landmark appears on its page, in reading order.
    expected = ["Bill To", "Items", "Total", "Footer"]  # replace per document
    for i, section in enumerate(expected):
        assert section in text_per_page[i] or i == 0

assert title, "PDF18: Document title required (WCAG 2.1)"

- name: PDF/UA conformance
  run: |
    for pdf in out/*.pdf; do
      # jq -e sets the exit status from the result: false/null gives exit 1,
      # so the gate actually fails on violations. Plain jq exits 0 regardless
      # of the printed boolean (per [jq manual]), so `|| exit 1` never fired.
      verapdf --flavour ua1 --format json "$pdf" \
        | jq -e '.report.jobs[0].validationResult.details.failedRules == 0' > /dev/null || exit 1
    done