feat: complete pre-ai local completion v3

central/app/ai/batch_runner.py

@@ -0,0 +1,55 @@
+"""Offline-safe, ID-correlated provider batch execution."""
+
+from dataclasses import dataclass, field
+
+from .provider import ProviderTransportError
+
+
+@dataclass
+class BatchRunResult:
+    results_by_work_item_id: dict[str, dict] = field(default_factory=dict)
+    errors: dict[str, str] = field(default_factory=dict)
+    provider_calls: int = 0
+    retries: int = 0
+    usage: dict = field(default_factory=dict)
+
+
+def run_batches(items, provider, batch_size: int, max_retries: int = 2, *, scenario: str | None = None) -> BatchRunResult:
+    if batch_size <= 0:
+        raise ValueError("batch_size must be positive")
+    if max_retries < 0:
+        raise ValueError("max_retries must not be negative")
+
+    result = BatchRunResult()
+    for start in range(0, len(items), batch_size):
+        batch = list(items[start:start + batch_size])
+        expected_ids = [str(item.work_item_id) for item in batch]
+        answers = None
+        for attempt in range(max_retries + 1):
+            result.provider_calls += 1
+            try:
+                kwargs = {} if scenario is None else {"scenario": scenario}
+                answers = provider.select_batch(batch, **kwargs)
+                break
+            except ProviderTransportError as error:
+                if attempt == max_retries:
+                    for work_item_id in expected_ids:
+                        result.errors[work_item_id] = str(error)
+                else:
+                    result.retries += 1
+            except Exception as error:
+                for work_item_id in expected_ids:
+                    result.errors[work_item_id] = str(error)
+                break
+        if answers is None:
+            continue
+        answer_ids = [str(answer.get("work_item_id")) for answer in answers]
+        if len(answer_ids) != len(expected_ids) or set(answer_ids) != set(expected_ids) or len(set(answer_ids)) != len(answer_ids):
+            error = "provider response did not contain exactly one answer per work_item_id"
+            for work_item_id in expected_ids:
+                result.errors[work_item_id] = error
+            continue
+        result.results_by_work_item_id.update({str(answer["work_item_id"]): answer for answer in answers})
+
+    result.usage = dict(getattr(provider, "usage", {}) or {})
+    return result