mateo-berri eba6dbe815 fix(anthropic): keep a valid breakpoint's 1h when it joins one whose ttl drop_params drops ... When breakpoints join into one mark (a tool result's parts, a message and its last block, a tool and its function), the first usable one stayed. A 30m breakpoint counts as usable because drop_params can send it without its ttl, so it took the place of a valid 1h beside it and the joined mark went out at 5m. The first breakpoint Claude takes as sent now stays, and the 30m one is still checked, so it answers 400 without drop_params as before The timing test for settling thousands of marks moves from tests/unit, whose guide bans clock reads, to a CodSpeed benchmark		2026-10-16 11:20:28 +00:00
..
__init__.py	Add CodSpeed performance benchmarks (#23676)	2026-03-14 18:44:36 -07:00
conftest.py	feat(tokenizer): preserve Python defaults with opt-in Rust dispatch (#42174)	2026-09-22 04:41:11 +00:00
test_a2a_benchmarks.py	test(benchmarks): add CodSpeed benchmarks for inference, MCP and A2A hot paths (#31716)	2026-06-30 10:27:12 -07:00
test_benchmarks.py	refactor: expose core private helpers under public names (#44871)	2026-10-07 00:05:27 +00:00
test_inference_benchmarks.py	test(benchmarks): add CodSpeed benchmarks for inference, MCP and A2A hot paths (#31716)	2026-06-30 10:27:12 -07:00
test_mcp_benchmarks.py	test(benchmarks): add CodSpeed benchmarks for inference, MCP and A2A hot paths (#31716)	2026-06-30 10:27:12 -07:00
test_prompt_cache_breakpoint_benchmarks.py	fix(anthropic): keep a valid breakpoint's 1h when it joins one whose ttl drop_params drops	2026-10-16 11:20:28 +00:00