What feature would you like to see?

TL;DR

• XMLAdapter only handles flat key/value XML.
• Nested models, lists or any repeated tags raise AdapterParseError.
• Root cause: a greedy one–level regex (<(\w+)>(.*?)</\1>) with no recursion.
• Effect: Structured-data extraction use-cases are broken on DSPy.
• I am preparing a PR that rewrites the parser to perform proper XML traversal; details & full analysis are linked below.

Minimal Reproduction

from typing import List
from pydantic import BaseModel
import dspy

class Person(BaseModel):
    name: str
    age: int

class PersonExtraction(dspy.Signature):
    text: str = dspy.InputField()
    person: Person = dspy.OutputField()

sig = PersonExtraction()
lm_output = """
<person>
  <name>John Doe</name>
  <age>30</age>
</person>
"""

dspy.XMLAdapter(sig).parse(lm_output)   # Raises AdapterParseError

Expected result

{"person": Person(name="John Doe", age=30)}

Actual result

AdapterParseError: Failed to parse LM response …

Environment
DSPy            : 3.0.0b1
Python          : 3.11.
OS              : macOS (Apple Silicon)

Impact

• Any signature with nested Pydantic models or List[...] cannot be parsed.
• Few-shot examples generated by format_field_with_value teach the LM an incorrect JSON-in-XML format, compounding errors.
• Users are pushed to JSONAdapter even when XML would be preferable.

Proposed Fix

Replace the one–level regex with an XML traversal (e.g., xml.etree.ElementTree) that:

• Recursively walks the tree,
• Builds Python containers (dict, list) matching the signature,
• Delegates scalar conversion to parse_value.
• Update format_field_with_value to emit canonical XML rather than JSON-in-XML.

Add unit tests covering:

• Nested models,
• Repeated tags → lists,
• Mixed scalar & container fields.

I have an implementation that passes these tests and will open a draft PR once this issue is acknowledged.

Would you like to contribute?

• Yes, I'd like to help implement this.
• No, I just want to request it.

Additional Context

No response