fix: handle lambda, postponed annotations, and subclass types in trim_messages token_counter

[atian8179]

Problem

trim_messages breaks when token_counter is a lambda, uses from __future__ import annotations, or has a BaseMessage subclass annotation (e.g. HumanMessage).

The root cause is the annotation check:

if ... .annotation is BaseMessage:

This uses identity comparison (is) which only matches the exact BaseMessage class. It fails for:

1. Lambdas — parameters have inspect.Parameter.empty annotation
2. Postponed annotations (from __future__ import annotations) — annotation is the string 'BaseMessage'
3. Subclass annotations (e.g. HumanMessage) — HumanMessage is BaseMessage is False

In all cases, the function incorrectly treats the callable as a list-level counter, causing a TypeError.

Solution

Extract the detection logic into a _is_per_message_counter() helper that handles all three cases:

• inspect.Parameter.empty → assume per-message (lambdas, bare defs)
• str annotation containing "message" → postponed annotations
• type that is a subclass of BaseMessage → subclass annotations

Fixes #35629

[codspeed-hq]

Merging this PR will improve performance by 39.56%

⚠️ 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 improved benchmarks
⏩ 23 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	WallTime	test_import_time[tool]	587.4 ms	474.1 ms	+23.9%
⚡	WallTime	test_async_callbacks_in_sync	25.6 ms	18.3 ms	+39.56%
⚡	WallTime	test_import_time[LangChainTracer]	485.8 ms	407.6 ms	+19.19%
⚡	WallTime	test_import_time[RunnableLambda]	537.2 ms	441.4 ms	+21.71%
⚡	WallTime	test_import_time[HumanMessage]	279.1 ms	235.4 ms	+18.55%
⚡	WallTime	test_import_time[Runnable]	538 ms	443.9 ms	+21.21%
⚡	WallTime	test_import_time[InMemoryVectorStore]	650.8 ms	536.3 ms	+21.36%
⚡	WallTime	test_import_time[ChatPromptTemplate]	682.8 ms	542.4 ms	+25.89%
⚡	WallTime	test_import_time[PydanticOutputParser]	583 ms	473.2 ms	+23.21%
⚡	WallTime	test_import_time[Document]	198 ms	166.3 ms	+19.1%
⚡	WallTime	test_import_time[CallbackManager]	341.3 ms	282.4 ms	+20.85%
⚡	WallTime	test_import_time[InMemoryRateLimiter]	186 ms	154.5 ms	+20.38%
⚡	WallTime	test_import_time[BaseChatModel]	573.8 ms	478.8 ms	+19.83%

---

_{Comparing atian8179:fix/trim-messages-lambda-counter (e97b42f) 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 (27651d9) 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 issue here :) Thanks for working on this, however i already have a PR open to fix this when i opened the issue. Please try to avoid duplicates :)

[ccurme (ccurme)]

Duplicated with #35630.