PDF 2.0 bundle: mechanisms and validation

Core 0.7 extends the PDF 2.0 foundation with the companion specifications distributed by the PDF Association. The mechanisms remain explicit PDF objects. Applications choose their file version, algorithms, namespaces, metadata, relationships and validation policy. There is no global registry or implicit conversion of an external file.

Scope by document

Document	Executable mechanisms	Validation and limits
ISO 32000-2:2020 EC3	Edition-specific base dictionary inventory, text encodings, associated files, R=6, namespaces, document parts, encrypted wrappers, DSS and document timestamps	306 dictionary tables and 1,709 entry objects; selected EC3 overlays and explicit semantic limits remain separate.
ISO/TS 32001:2022 EC3	SHA3-256/384/512 and fixed 64-byte SHAKE256; explicit ISO extension declaration and signature digest objects	Independent digest/CMS checks; EC3 deletion of clause 5.1.3 is respected rather than extending the struck-out signature-reference rule.
ISO/TS 32002:2022 EC3	Named P-256/P-384/P-521 and Brainpool curves, Ed25519/Ed448, exact curve/digest combinations and CMS construction	Independent verification of 23 supported curve/digest signatures. Certificate trust and revocation remain application policies.
ISO/TS 32003:2023	AES-GCM/AESV4 with V=6/R=7, authentication tag, per-key nonce sequence and R=6 credential algorithms	pyHanko decrypts native files; altered ciphertext is rejected. R=6 AES-CBC remains a separate selection.
ISO/TS 32004:2024	KDFSalt, permission bit 13, HKDF, AES key wrap, CMS AuthenticatedData, autonomous and signature-attached MAC	pyHanko verifies actual initial, incremental and linearized files; MAC output has an exact DER length without padding.
ISO/TS 32005:2023	Qualified roles, standard/custom namespaces, role-map resolution, common PDF 2.0 roles and table 5 inclusion rules	Full source matrix is reproducible. Published issue resolutions are selected explicitly; source ambiguities are not silently guessed.
PDF Declarations	Declaration/ClaimData objects, dates and optional PDF/A XMP extension schema	Declarations are author assertions with a precise metadata scope; they are not audit certificates.
WTPDF 1.0.0	Explicit reuse/accessibility audit levels, structural, metadata, annotation and font-use checks	Reports distinguish detected failures from pending semantic and external font-program checks. No automatic PDF/UA-2 certification is claimed.
Application Note 001	`UseBlackPtComp` with exactly `ON`, `OFF` or `Default`	PDF 2.0 version and value checks, including raw graphics-state resources without a Type entry. The output device performs colour management.
Application Note 002	Associated files on information objects, marked content, encrypted payload descriptor and document wrapper	Effective graph checks cover discoverability, relationships, hidden collection and the unencrypted outer document.
Application Note 003	Object metadata streams, correct resource scope, named marked-content properties, document parts and DPM	Native XMP serialization is escaped UTF-8 XML. Opaque caller-supplied XML and content bytes require their own validation.

The base ISO 32000-2 catalogue inventories 440 tables and models 306 dictionary tables with 1,709 fixed or variable entries. Source continuation pages and extraction corrections are traceable. This coverage and these companion mechanisms do not establish complete conformance to every clause of the base standard. The PDF 2.0 guide and coverage guide retain that distinction. The optional reader's supported input scope is unchanged; the writer's new mechanisms do not turn it into a PDF 2.0 reader.

Algorithms are selected independently

ModernStandardSecurityHandler constructs R=6 AES-CBC or R=7 AES-GCM credentials from explicitly prepared user and owner passwords. ModernFileEncryption binds the credentials, file key and cryptographic operation. Pdf20Permissions keeps the permission word and the MAC requirement consistent. An ISO extension is an array member under Extensions.ISO_; declaring level 32004 does not implicitly declare levels 32001, 32002 or 32003.

PdfMacTokenComposer binds the actual file encryption key and the 32-byte KDFSalt. The writer reserves an exact hexadecimal slot, fixes the byte range, hashes the final serialized bytes and inserts the DER token without shifting offsets. Incremental writing preserves earlier bytes and requires a fresh MAC operation for the new revision. The same rules apply to linearized output.

$mac = new PdfMacTokenComposer($encryption->key, $salt, new SystemRandomSource());
$revision = (new FileWriter(encryption: $encryption, mac: $mac))->revision($file);

$nextMac = new PdfMacTokenComposer($encryption->key, $salt, new SystemRandomSource());
$next = (new IncrementalWriter(mac: $nextMac))->serialize(
    new IncrementalUpdate($revision, $explicitDelta),
);

For a signed revision, select macSignature: $pending on the writer and pass mac: $mac to PreparedSignature. The MAC becomes a CMS unsigned attribute bound to both the PDF byte range and the actual SignerInfo signature octets. A standalone MAC cannot be carried forward unchanged when signing changes the reserved signature bytes. Capacity includes the attached MAC; overflow requires a new reservation and a new signature.

PHP's OpenSSL extension handles the supported RSA/ECDSA primitives. Ed25519 and Ed448 use an explicitly selected OpenSSL executable through the dedicated provider because PHP 8.3 does not expose the needed EdDSA operation. This provider receives the executable and key explicitly; selecting other algorithms does not start a process. See the signature and integrity tutorial.

Metadata and declarations have an owner

XmpPacket composes properties, RDF bags/sequences, localized alternatives and declarations. MetadataStream writes the XML stream; MetadataEntry links it indirectly to its PDF owner. A declaration in document metadata concerns the document. The same declaration on an object or marked sequence has that narrower scope. Metadata properties do not automatically become visible page text.

Named marked-content properties resolve in the page, Form XObject, tiling pattern or Type 3 font resource scope. Emission facts are frozen with the actual serialized stream, so an unstaged builder change cannot rewrite an earlier revision's meaning. Direct inline metadata stream references are rejected.

DocumentPart, DocumentPartRoot and DocumentPartMetadata represent document parts and their data. Native graph validation checks hierarchy, page ranges and the bidirectional page-to-part association. An encrypted payload uses a separate outer document, an associated file with EncryptedPayload relationship, an EP descriptor and a hidden collection. The outer document remains readable without decrypting its inner PDF. The tutorials explain metadata and parts and the encrypted wrapper.

Structure rules and WTPDF auditing

StructureNamespace, StructureRole and NamespaceRoleMap make namespace identity part of the role. NamespaceRoleResolver follows qualified mappings and rejects unresolved targets and cycles. The effective document must be PDF 2.0 when these mechanisms are used, including retained incremental objects.

StructureInclusionRules applies the source matrix to real parent/child roles, including the PDF 1.7-only exemption and mixed-namespace rules. Corrections 336 and 440 are an explicit correction selection. The retained source's literal c occurrence is preserved in the source inventory; the selected published resolution supplies its interpretation.

WtPdfAudit produces findings for the selected conformance level. Its report does not equate "no detected failure" with "conformant": meaningful reading order, appropriate alternate descriptions, faithful Unicode mapping and external font-program facts still need evidence. The namespace tutorial shows a concrete report next to the actual tagged PDF.

Preserve signatures with timestamps and validation data

PendingDocumentTimestamp owns a /DocTimeStamp dictionary with the ETSI.RFC3161 SubFilter. Rfc3161TimestampBackend sends an explicit provider the actual document imprint and nonce, then checks the returned token, authority binding and CMS signature. LocalTimestampAuthority is an explicitly selected offline provider; it performs no network request and establishes no public trust.

DocumentSecurityStore contains indirect certificate, CRL and OCSP streams. ValidationRelatedInformation requires a successful caller-selected validation policy and identifies the exact earlier signature Contents, including padding. The graph rejects VRI introduced before its signature or evidence absent from the matching DSS arrays. A later timestamp covers the appended evidence too.

ISO 32000-2 excludes document timestamps and DSS from DocMDP evaluation. DocumentTimestampRevisionPolicy proves a narrow maintenance delta against all frozen previous objects before allowing it under P1: pages, content, widgets and ordinary signatures cannot change alongside it. FieldMDP and additional field locks remain binding. The timestamp tutorial explains three actual revisions; the independent oracle also checks certified, encrypted and linearized histories.

Geospatial measurements retain their coordinate system

GeographicCoordinateSystem and ProjectedCoordinateSystem identify an explicit EPSG code or WKT string. GeospatialControlPoints pairs local and world positions and derives the matching LPTS/GPTS arrays. GeospatialMeasure, Viewport and Viewports attach those measurements to their actual page, Form or image owner. PreferredDisplayUnits, ProjectedCoordinateSystemMatrix and PointDataCollection complete the display preferences, affine calculation and point-cloud objects.

Native and effective-graph validation checks contexts, finite values, matching dimensions, normalized bounds and version requirements. It does not fetch a GIS registry or convert projections. The geospatial tutorial explains two independently declared sample regions and their serialized data.

StructureDestination identifies a real member of the structure tree and an explicit DestinationView. Page resolution follows the first content item in declared child order, with a first-page fallback when no content item qualifies. GoTo actions and named destination dictionaries preserve their page fallback alongside SD. Remote and embedded forms use explicit byte-string IDs without opening another file. Effective-graph checks preserve these relationships in retained revisions. The structure-destination tutorial shows semantic links and their actual PDF objects.

Independent checks

The PHP suite runs under PHP 8.3, 8.4 and 8.5. The companion checks use separate implementations:

pyHanko validates password/public-key AES-CBC and AES-GCM files, standalone and attached MACs, historical revisions and tampering rejection.
cryptography and ASN.1 tooling verify curve/digest signatures and MAC tokens.
pyHanko and OpenSSL verify RFC3161 tokens, DSS/VRI, certification and every signed historical prefix rather than only the last file revision.
qpdf validates the emitted structures and independently decrypts the example encrypted payload; pypdf and an XML parser inspect the metadata and relations.
The TS 32005 source matrix is regenerated from all its table pages and checked against the committed resource and PHP representation.

These checks establish the tested behaviour. They do not promise that arbitrary application input is safe or that an author-supplied conformance claim is true.

Retained originals and provenance

All eleven originals retain their supplied names and bytes under resources/reference-documents/ in the core repository. The source manifest records fingerprints, byte counts, physical pages and correction collections. They are linked from the site rather than copied into Composer installations or duplicated as large documentation assets.

Guide source at the documented commit ↗