XMLOutputParser and PydanticOutputParser Fixes in langchain-core 1.4.9

What it does

langchain-core 1.4.9 patches two output-parser bugs. XMLOutputParser mishandled continuation lines with colons inside tags, rejecting valid XML. PydanticOutputParser fixed how field validation errors propagate, so ValidationError messages surface clearly when a model’s fields cannot be populated from the LLM text.

Why it matters

These parsers back LangChain’s structured output pipeline: every with_structured_output(PydanticModel) or XML-formatted chain uses one. Bugs here silently corrupt data or crash pipelines at scale. The Pydantic fix alone turns cryptic internal errors into field-level detail.

Example

$ Parse a movie review into a Pydantic model, then parse planets as XML.

=== PydanticOutputParser Result ===
Title:  Inception
Rating: 8.5/10
Pros:   ['visually stunning', 'original story']
Cons:   ['complex', 'slow at times']

=== XMLOutputParser Result ===
<Entities>
  <Entity name="Mercury" type="planet" distance="0.39 AU"/>
  <Entity name="Venus" type="planet" distance="0.72 AU"/>
  ...
</Entities>

Common flags

PydanticOutputParser

Parses LLM text into a Pydantic v2 model with full validation

XMLOutputParser

Parses LLM text as XML using a configurable schema

JsonOutputParser

Parses LLM text as JSON into a Pydantic or dataclass model

RetryOutputParser

Wraps any parser and retries on parse failure

with_structured_output(Model)

Wires format instructions and a parser automatically