fix(weave): support google-adk 2.3.0+ trace_inference_result signature

[fixit-agent]

Problem

Weave's Google ADK integration is incompatible with google-adk >= 2.3.0 (including 2.4.0). ADK prepended a leading invocation_context argument to trace_inference_result in 2.3.0:

ADK version	Signature
≤ 2.2.0	trace_inference_result(span, llm_response)
≥ 2.3.0	trace_inference_result(invocation_context, span, llm_response)

_wrap_trace_inference_result still declared and called the original with the old 2-arg form (original(span, llm_response)), so once Weave's patch is applied on ADK ≥ 2.3.0, every ADK inference raises TypeError. The test shard installs google-adk>=2.2.0, which now resolves to 2.4.0.

Fix

weave/integrations/google_adk/google_adk_sdk.py:

• The wrapper now accepts *args and forwards them verbatim to the same-version original — so it calls ADK with exactly the arguments ADK handed it, regardless of shape.
• It then reads the span and response off the tail two positions (args[-2], args[-1]). ADK kept span and llm_response in the same relative order and only prepended invocation_context, so tail-based extraction is correct for both the 2-arg and 3-arg forms. ADK always calls this function positionally, so *args captures every case.
• _TraceInferenceResult is broadened to the open Callable[..., None], matching the existing _TraceToolCall pattern already used in this file for ADK's cross-release signature churn.

This keeps compatibility with ADK >= 1.36 — no version sniffing required.

Testing

• Added test_trace_inference_result_supports_3arg_signature — drives the wrapper with the new 3-arg shape, asserts the leading arg is forwarded to the original and the span is still enriched from the trailing response.
• The existing 2-arg partial-chunk test still passes, confirming backward compatibility.
• End-to-end proof: the google_adk shard installs google-adk 2.4.0 (the exact 3-arg version from the report), and TestAdkRunnerE2E::test_emits_all_otel_attributes — which drives a real ADK InMemoryRunner through a full inference loop — passes. On the old code this test raised TypeError on every inference.

15 passed in 20.19s   # nox tests-3.12(shard='google_adk')

Lint (ruff check) passes on both changed files.

---

🤖 Opened by Fixit
Requested by: Emmanuel Turlay
Slack thread: https://weightsandbiases.slack.com/archives/C0APFJB2H5E/p1783722806526949
View session: https://fixit.withwandb.com/session/fix-6a51734a-f9df9fc3

[codecov]

Codecov Report

❌ Patch coverage is 33.33333% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
weave/integrations/google_adk/google_adk_sdk.py	33.33%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!