core: fix trim_messages token_counter detection for lambdas and postponed annotations

[Alvin Tang (alvinttang)]

Description

Fixes a bug where trim_messages misclassifies per-message token_counter callables as per-list counters, causing AttributeError: 'list' object has no attribute 'content' at runtime.

Root cause

The detection logic used annotation is BaseMessage (identity check) to determine if token_counter accepts a single BaseMessage or a list. This fails for:

• Lambdas — no annotation available
• Unannotated functions — annotation is inspect.Parameter.empty
• String annotations — annotation is a string like "BaseMessage", not the class
• Subclass annotations — e.g. msg: HumanMessage — HumanMessage is not BaseMessage
• Postponed annotations — from __future__ import annotations turns all annotations into strings

Fix

Added a _is_per_message_counter() helper that:

1. Uses typing.get_type_hints() to resolve string and postponed annotations
2. Uses issubclass() instead of identity to handle subclass annotations
3. Falls back to probing with a dummy message for lambdas and unannotated callables

Before

# All of these fail with AttributeError
trim_messages(messages, max_tokens=5, token_counter=lambda msg: len(msg.content.split()))
trim_messages(messages, max_tokens=5, token_counter=counter_no_annotation)
trim_messages(messages, max_tokens=5, token_counter=counter_with_subclass_annotation)

After

All forms work correctly, including len as a list counter.

Issue

Fixes #35629

Tests

Verified with the reproduction script from the issue — all five cases now pass. List counters (len, annotated list functions) continue to work correctly.

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 13 untouched benchmarks
⏩ 23 skipped benchmarks¹

---

_{Comparing alvinttang:fix/trim-messages-callable-counter (d44e9cb) with master (29134dc)²}

Footnotes

1. 23 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on master (532b014) during the generation of this report, so 29134dc was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

[GAUTAM V DATLA (gautamvarmadatla)]

Hi! Creator of the issue here :)

Thanks for working on this. However, I already have a PR open to fix it from when I originally opened the issue. Plelase try to avoid duplicates

[ccurme (ccurme)]

Duplicated with #35630.

[Ethan T. (gambletan)]

Sorry about this — I should have checked for existing PRs before submitting. I didn't realize you already had #35630 open for the same issue. Won't happen again. Thanks for the fix!