feat: add as_generic_chat_history function to convert any Context to … by akihikokuroda · Pull Request #1007 · generative-computing/mellea

akihikokuroda · 2026-05-04T18:59:54Z

Misc PR

Type of PR

Bug Fix
New Feature
Documentation
Other

Description

Link to Issue: Fix: fix: as_chat_history is outdated and functionality doesn't match it's name #932

Add as_generic_chat_history() function to handle heterogeneous contexts with
configurable formatter. Unlike as_chat_history(), this gracefully handles
arbitrary component types by converting them to strings, with optional logging
for unknown types.

as_generic_chat_history(ctx, formatter=None) converts any Context to Messages
Default formatter logs warnings for unknown component types
Support custom formatter parameter for domain-specific type handling
Handle ModelOutputThunk, CBlock, and arbitrary components seamlessly
CBlock → Message(role="user"), MOT → Message(role="assistant")
Add 7 comprehensive tests covering all conversion scenarios

Testing

Tests added to the respective file if code was changed
New code has 100% coverage if code as added
Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

Attribution

AI coding assistants used: claude

…Message Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

github-actions · 2026-05-04T19:00:27Z

The PR description has been updated. Please fill out the template for your PR to be reviewed.

psschwei · 2026-05-05T13:59:21Z

+                    return c.parsed_repr
+                # Use value if parsed_repr is None or not a Message
+                content = (
+                    str(c.value) if c.parsed_repr is None else formatter(c.parsed_repr)


I think we also need to consider the case if c.parsed_repr is a string (I believe that's what a MOT will return after a completed CBlock action (?) )

quick way to test:

mot = ModelOutputThunk(value='reply text') mot.parsed_repr = 'reply text' as_generic_chat_history(ChatContext().add(Message('user','hi')).add(mot)) # → WARNING: Unknown component type str ...

Yes, it was an issue. I fix it. Thanks!

should we add a test case for it too?

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

psschwei · 2026-05-05T18:18:43Z

there's a typo in the PR title: s/ ant / any /

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

planetf1 · 2026-05-06T07:25:42Z

+
+def as_generic_chat_history(
+    ctx: Context, formatter: Callable[[Any], str] | None = None
+) -> list[Message]:


Minor / style: Small typing issue here — both _default_formatter and the formatter parameter use Any, which the project convention asks us to avoid unless an external library forces it (nothing does here). object is the right type: it accepts everything, is covariant, and keeps mypy happy with callers that pass typed formatters.

def _default_formatter(obj: object) -> str: ... formatter: Callable[[object], str] | None = None

The test helpers already use obj: object, so it is just the library code that needs updating.

planetf1 · 2026-05-06T07:25:52Z

+                # Use value if parsed_repr is None or some other type
+                content = (
+                    str(c.value) if c.parsed_repr is None else formatter(c.parsed_repr)
+                )


Two things worth noting on this line — one minor, one more significant.

Minor / style: When parsed_repr is a dict or similar, the default formatter will log "Unknown component type dict" — but dict is not an unexpected component here, it is a parsed form of a ModelOutputThunk we already recognise. That message will be confusing to anyone monitoring logs. A dedicated log line would be clearer:

_logger.warning( "ModelOutputThunk.parsed_repr is %s, not a Message; falling back to value.", type(c.parsed_repr).__name__, )

More significant: If both parsed_repr and value are None (a ModelOutputThunk that has not been evaluated yet), str(None) silently produces the literal string "None" in the chat history with no warning logged. A backend receiving that will have no indication anything went wrong. Worth guarding explicitly:

if c.value is None: raise ValueError("ModelOutputThunk has no value and no parsed_repr — was it evaluated?")

planetf1 · 2026-05-06T07:26:00Z

+                return Message(role="assistant", content=content)
+            case CBlock():
+                return Message(role="user", content=str(c))
+            case _:


More significant: One tricky edge case: ImageBlock inherits from CBlock (see core/base.py:79), so it will fall into this arm and produce Message(role="user", content=str(image_block)) — meaning a raw base64 PNG string ends up in the chat history as a text message. This is a reachable code path if a context with image content is passed in, and the corruption would be invisible downstream.

A simple guard fixes it:

case CBlock(): if type(c) is not CBlock: return Message(role="user", content=formatter(c)) return Message(role="user", content=str(c))

That way any CBlock subclasses not explicitly handled go through the formatter instead.

planetf1

Generally looks good - just came across 2 possible corruption issues. 1 is an edge case, the other affects images. 2 other minors.

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

Co-authored-by: Paul Schweigert <paul@paulschweigert.com>

markstur

No blockers from me unless you agree to make formatter more general purpose.

Bob had some suggestions below. I'd probably do #2, but optional/FYI --

Suggestions (Optional Improvements)

Type Narrowing: Consider using typing.TypeGuard for _to_msg to help type checkers understand the return type better.
Performance: The nested _to_msg function is recreated on every call. Consider making it a module-level function if performance matters.
Consistency with as_chat_history(): Both functions raise the same generic Exception for non-linear contexts. Consider extracting this to a shared helper or using a custom exception type.

markstur · 2026-05-08T18:01:02Z

+    Returns:
+        List of ``Message`` objects in conversation order.
+
+    Raises:


also:
ValueError: If a ModelOutputThunk has neither value nor parsed_repr.

markstur · 2026-05-08T18:02:08Z

@@ -0,0 +1,85 @@
+# pytest: unit


I thought we default to unit and don't put this in, but at the moment I'm not sure if that applies to docs/examples.

pytest: unit in docs/examples/as_generic_chat_history.py is correct per AGENTS.md — examples require this opt-in comment to be collected.

markstur · 2026-05-08T18:52:13Z

+
+    def _to_msg(c: CBlock | Component | ModelOutputThunk) -> Message:
+        match c:
+            case Message():


as a naive user I expected to be able to format a Message.
I might be totally wrong about the use case, but hard-coding a "format=" to be so specialized seems odd. Why doesn't the formatter work on the prepared final message/content instead of only certain types of content.

Formatting apply to the Message content itself would be a breaking change and might not be the intended use case.

markstur · 2026-05-08T18:53:28Z

+
+    all_ctx_events = ctx.as_list()
+    if all_ctx_events is None:
+        raise Exception("Trying to cast a non-linear history into a chat history.")


nit: maybe a ValueError

markstur · 2026-05-08T19:01:58Z

+    assert len(history) == 2
+    assert history[0].role == "user"
+    assert history[0].content == "inline content"
+


Bob suggested test (needs testing):

Missing test: The CBlock subclass path (lines 347-348) isn't explicitly tested. The test at line 357-365 uses a plain CBlock, not a subclass.

Recommendation: Add a test:

def test_as_generic_chat_history_with_cblock_subclass(): """CBlock subclasses use the formatter.""" class CustomCBlock(CBlock): pass ctx = ChatContext() ctx = ctx.add(CustomCBlock("custom content")) history = as_generic_chat_history(ctx) assert len(history) == 1 assert history[0].role == "user"

The test is added.

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

feat: add as_generic_chat_history function to convert ant Context to …

dce6efc

…Message Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

akihikokuroda requested a review from a team as a code owner May 4, 2026 18:59

akihikokuroda requested review from markstur and planetf1 May 4, 2026 18:59

github-actions Bot added the enhancement New feature or request label May 4, 2026

psschwei reviewed May 5, 2026

View reviewed changes

review comment

afbb19a

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

akihikokuroda changed the title ~~feat: add as_generic_chat_history function to convert ant Context to …~~ feat: add as_generic_chat_history function to convert any Context to … May 5, 2026

review comments

6a5b6ae

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

planetf1 reviewed May 6, 2026

View reviewed changes

planetf1 requested changes May 6, 2026

View reviewed changes

review comments

e055bdd

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

akihikokuroda requested review from planetf1 and psschwei May 6, 2026 13:24

psschwei reviewed May 6, 2026

View reviewed changes

Comment thread mellea/stdlib/components/chat.py Outdated

Update mellea/stdlib/components/chat.py

019cdd3

Co-authored-by: Paul Schweigert <paul@paulschweigert.com>

akihikokuroda requested a review from psschwei May 6, 2026 19:04

psschwei approved these changes May 6, 2026

View reviewed changes

markstur reviewed May 8, 2026

View reviewed changes

akihikokuroda requested a review from markstur May 9, 2026 00:29

review comments

630d14d

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

Conversation

akihikokuroda commented May 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Misc PR

Type of PR

Description

Testing

Attribution

Uh oh!

github-actions Bot commented May 4, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

psschwei commented May 5, 2026

Uh oh!

planetf1 May 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

planetf1 May 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

planetf1 May 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

planetf1 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

markstur left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

akihikokuroda commented May 4, 2026 •

edited

Loading

planetf1 May 6, 2026 •

edited

Loading

planetf1 May 6, 2026 •

edited

Loading

planetf1 May 6, 2026 •

edited

Loading