fix: PDF/A-3 metadata and ICC profile validation errors

[chrizzellu]

Summary

Fixes PDF/A-3B validation errors reported by the official Factur-X validator at https://services.fnfe-mpe.org/. PDFs generated by embedInPdf now pass both PDF and XML validation. Relates to #110.

Root Cause

The XMP metadata stream contained xmlns:about="" (an empty XML namespace declaration), which is forbidden by the XML Namespace specification. This caused every strict XML parser — including the PDF/A validator — to reject the entire XMP document. The validator then:

1. Could not read pdfaid:part=3, so it defaulted to testing as PDF/A-1b
2. Reported all XMP metadata fields as null
3. Failed EmbeddedFiles checks (forbidden in PDF/A-1, allowed in PDF/A-3)

This single typo (xmlns:about instead of rdf:about) was the root cause of all reported PDF validation errors.

Changes

1. Fix invalid xmlns:about="" → rdf:about="" (formatter.ts)

The Factur-X rdf:Description block for fx:DocumentType etc. used @xmlns:about which generates xmlns:about="" — an empty namespace, invalid XML. Changed to @rdf:about which generates the correct rdf:about="" attribute.

This was the critical fix that made the entire XMP parseable.

2. Add dc:title and dc:description to XMP metadata (formatter.ts)

The Info dictionary fields Title and Subject (set via setTitle/setSubject) had no corresponding XMP entries, causing metadata mismatch errors. Now included with correct rdf:Alt / xml:lang="x-default" structure per XMP spec.

3. Fix dc:creator to use metadata.author (formatter.ts)

dc:creator was set to metadata.creator, but per PDF/XMP spec:

• dc:creator must match Info /Author (= metadata.author)
• xmp:CreatorTool must match Info /Creator (= metadata.creator)

Falls back to metadata.creator when author is not provided.

4. Encode XMP as UTF-8 bytes (formatter.ts)

The XMP string was passed directly to context.stream() with Length: xmp.length. JavaScript's .length counts UTF-16 code units, but the PDF stream contains UTF-8 bytes. The \uFEFF BOM (1 JS char → 3 UTF-8 bytes) caused a length mismatch that made validators fail to read the stream. Now uses TextEncoder().encode() for correct byte-level handling.

5. Add N: 3 to ICC profile stream (formatter.ts)

The sRGB ICC profile stream was missing the required N entry (number of color components), causing: "The N entry (value null) does not match the number of components in the embedded ICC profile (color space RGB)"

6. Load PDF with updateMetadata: false (create.ts)

PDFDocument.load() defaults to updateMetadata: true, which auto-sets Producer = "pdf-lib" and Creator = "pdf-lib" in the Info dictionary before embedInPdf can set the correct values. Now disabled to prevent stale metadata.

7. Save with useObjectStreams: false (create.ts)

Forces traditional cross-reference tables instead of xref streams for maximum PDF/A compatibility.

Validation Results

Before (from issue #110):

PDF: Not a PDF/A-3 (13 errors)
XML: valid

After:

PDF: flavour=3b, isCompliant=true, assertions=[] (0 errors)
XML: valid (32 rules fired, 0 failed, 8 XRechnung-specific notices)

Tested with Puppeteer-generated PDFs validated against the official Factur-X validator.

Test plan

• Existing tests pass (2 pre-existing XSD validation failures in formatter.test.ts unrelated — require Java)
• Generated PDFs validated at https://services.fnfe-mpe.org/ — PDF/A-3B compliant, XML valid
• Tested with BASIC profile, real invoice data, EU reverse charge scenario

[vercel]

@chrizzellu is attempting to deploy a commit to the better-auth-extended Team on Vercel.

A member of the Team first needs to authorize it.

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/jslno/node-zugferd/@node-zugferd/api@111

npm i https://pkg.pr.new/jslno/node-zugferd@111

commit: 1c883fc

[cubic-dev-ai]

No issues found across 2 files