roughed in cache; elide full dirnames from test results to make them deterministic

aider/sendchat.py

@@ -4,6 +4,7 @@ import json
 import backoff
 import openai
 import requests
+from diskcache import Cache
 from openai.error import (
     APIConnectionError,
     APIError,
@@ -12,6 +13,9 @@ from openai.error import (
     Timeout,
 )
 
+CACHE_PATH = ".aider.send.cache.v1"
+CACHE = Cache(CACHE_PATH)
+
 
 @backoff.on_exception(
     backoff.expo,
@@ -44,10 +48,22 @@ def send_with_retries(model, messages, functions, stream):
     if hasattr(openai, "api_engine"):
         kwargs["engine"] = openai.api_engine
 
+    key = json.dumps(kwargs, sort_keys=True).encode()
+
     # Generate SHA1 hash of kwargs and append it to chat_completion_call_hashes
-    hash_object = hashlib.sha1(json.dumps(kwargs, sort_keys=True).encode())
+    hash_object = hashlib.sha1(key)
+
+    if not stream and key in CACHE:
+        print("hit", key)
+        return hash_object, CACHE[key]
+
+    print("miss", key)
 
     res = openai.ChatCompletion.create(**kwargs)
+
+    if not stream:
+        CACHE[key] = res
+
     return hash_object, res
 
 

benchmark/benchmark.py

@@ -657,7 +657,7 @@ def run_unit_tests(testdir, history_fname):
 
     success = result.returncode == 0
     res = result.stdout
-    res = cleanup_test_output(res)
+    res = cleanup_test_output(res, testdir)
 
     with history_fname.open("a") as fh:
         fh.write(f"```\n{res}\n```")
@@ -667,7 +667,7 @@ def run_unit_tests(testdir, history_fname):
         return res
 
 
-def cleanup_test_output(output):
+def cleanup_test_output(output, testdir):
     # remove timing info, to avoid randomizing the response to GPT
     res = re.sub(
         r"^Ran \d+ tests in \d+\.\d+s$",
@@ -687,6 +687,8 @@ def cleanup_test_output(output):
         res,
         flags=re.MULTILINE,
     )
+
+    res = res.replace(str(testdir), str(testdir.name))
     return res