[BUG] DICOM channels with a filter/transformer fail with `IllegalStateException: state:EXPECT_FRAG` on private undefined-length sequences

[jessesaga]

Describe the bug

Any DICOM-typed connector that has a filter or transformer attached fails to process messages containing an undefined-length sequence under a private tag that dcm4che2 resolves to VR UN (implicit VR, tag not in its dictionary). Common real-world producers: FujiFILM (0029,E131) and Siemens MEDCOM (0029,1140). Messages with no filter/transformer pass through fine, because they never go through the XML round trip.

com.mirth.connect.donkey.model.message.MessageSerializerException: Error converting XML to DICOM
Caused by: java.lang.IllegalStateException: state:EXPECT_FRAG
        at org.dcm4che2.io.ContentHandlerAdapter.startElement(...)
        at com.mirth.connect.plugins.datatypes.dicom.DICOMSerializer.fromXML(...)

Root cause

dcm4che2 2.0.29's XML round trip is asymmetric:

• DicomInputStream reads a UN element with undefined length whose items hold dataset elements and represents it as SQ in memory.
• toXML uses the streaming SAXWriter, which emits the wire VR: <tag0029E131 vr="UN" len="-1"> with structured <item> children.
• fromXML hands that to ContentHandlerAdapter, which only enters its sequence state for vr="SQ" and otherwise expects fragments; the first <attr> inside the item throws state:EXPECT_FRAG.

dcm4che2 has had no release since 2015, so this cannot be fixed at the library.

To Reproduce

Programmatic, no fixtures needed (this is what the accompanying test does):

1. Build a BasicDicomObject, putNestedDicomObject(0x0029E131, item) with any element in the item, write it as Implicit VR Little Endian.
2. new DICOMSerializer().toXML(base64) — note the output carries vr="UN" with an <item> child.
3. fromXML(thatXml) throws.

Or in a running engine: DICOM Listener → any transformer step → send a Fuji or Siemens CR/CT object carrying a private sequence.

Expected behavior

fromXML accepts the XML toXML produced, and the re-encoded DICOM is identical to what the engine emits without a transformer.

Proposed fix

Reader-side, in DICOMSerializer.fromXML's existing DOM pre-pass (the one that renames <tagXXXXXXXX> back to <attr>): when an element's <item> children contain renamed <attr> elements, set vr="SQ" before the SAX parse. That is the same content-based decision DicomInputStream makes for the binary form. toXML output is unchanged, so transformer scripts see identical XML; fragment containers (encapsulated pixel data) are untouched; every input that parses today produces an identical DicomObject; and the rewrite is a no-op if a future dcm4che emits SQ itself.

Verified byte-for-byte that the transformer round trip now produces the same DICOM as the no-transformer pass-through path, for implicit and explicit VR LE. PR to follow.

Environment

• OIE main at 423df18 (also reproduces on 4.6.0-based builds)
• dcm4che2 2.0.29 as bundled