core: speedup writing to cache by using qmark style sql rather than n…

…amed

```
before
src/cachew/tests/test_cachew.py::test_many[gc_off-1000000] [INFO    2023-09-13 03:31:29,980 cachew __init__.py:778 ] cachew.tests.test_cachew:test_many.<locals>.iter_data: wrote   1000000 objects to   cachew (sqlite /tmp/pytest-of-karlicos/pytest-18/test_many_gc_off_1000000_0/test_many)
test_many: initial write to cache took 4.5s
test_many: cache size is 72.904704Mb
[INFO    2023-09-13 03:31:30,273 cachew __init__.py:641 ] cachew.tests.test_cachew:test_many.<locals>.iter_data: loading 1000000 objects from cachew (sqlite /tmp/pytest-of-karlicos/pytest-18/test_many_gc_off_1000000_0/test_many)
test_many: reading from cache took 2.9s
```

after:
```
src/cachew/tests/test_cachew.py::test_many[gc_off-1000000] [INFO    2023-09-14 20:50:55,287 cachew __init__.py:791 ] cachew.tests.test_cachew:test_many.<locals>.iter_data: wrote   1000000 objects to   cachew (sqlite /tmp/pytest-of-karlicos/pytest-80/test_many_gc_off_1000000_0/test_many)
test_many: initial write to cache took 3.3s
test_many: cache size is 72.904704Mb
[INFO    2023-09-14 20:50:55,479 cachew __init__.py:654 ] cachew.tests.test_cachew:test_many.<locals>.iter_data: loading 1000000 objects from cachew (sqlite /tmp/pytest-of-karlicos/pytest-80/test_many_gc_off_1000000_0/test_many)
test_many: reading from cache took 2.8s
```

src/cachew/__init__.py

@@ -46,6 +46,7 @@ def orjson_dumps(*args, **kwargs):  # type: ignore[misc]
 import appdirs
 import sqlalchemy
 from sqlalchemy import Column, Table, event, text
+from sqlalchemy.dialects import sqlite
 
 from .logging_helper import makeLogger
 from .marshall.cachew import CachewMarshall, build_schema
@@ -630,8 +631,8 @@ def cachew_wrapper(
 
             def cached_items():
                 rows = conn.execute(table_cache.select())
-                for row in rows:
-                    j = orjson_loads(row[0])
+                for (blob,) in rows:
+                    j = orjson_loads(blob)
                     obj = marshall.load(j)
                     yield obj
 
@@ -730,19 +731,18 @@ def missing_keys(cached: List[str], wanted: List[str]) -> Optional[List[str]]:
             # at this point we're guaranteed to have an exclusive write transaction
 
             datas = func(*args, **kwargs)
-            column_names = [c.name for c in table_cache_tmp.columns]
-            insert_into_table_cache_tmp = table_cache_tmp.insert()
+            # uhh. this gives a huge speedup for inserting
+            # since we don't have to create intermediate dictionaries
+            insert_into_table_cache_tmp_raw = str(table_cache_tmp.insert().compile(dialect=sqlite.dialect(paramstyle='qmark')))
+            # I also tried setting paramstyle='qmark' in create_engine, but it seems to be ignored :(
+            # idk what benefit sqlalchemy gives at this point, seems to just complicate things
 
             chunk: List[Any] = []
+
             def flush() -> None:
                 nonlocal chunk
                 if len(chunk) > 0:
-                    # TODO optimize this, we really don't need to make extra dicts here just to insert
-                    chunk_dict = [
-                        dict(zip(column_names, row))
-                        for row in chunk
-                    ]
-                    conn.execute(insert_into_table_cache_tmp, chunk_dict)
+                    conn.exec_driver_sql(insert_into_table_cache_tmp_raw, [(c,) for c in chunk])
                     chunk = []
 
             total_objects = 0
@@ -755,8 +755,8 @@ def flush() -> None:
                     return
 
                 dct = marshall.dump(obj)
-                j = orjson_dumps(dct)
-                chunk.append((j,))
+                blob = orjson_dumps(dct)
+                chunk.append(blob)
                 if len(chunk) >= chunk_by:
                     flush()
             flush()

src/cachew/tests/test_cachew.py

@@ -23,7 +23,7 @@
 
 from .. import cachew, get_logger, NTBinder, CachewException, settings
 
-from .utils import running_on_ci
+from .utils import running_on_ci, gc_control
 
 
 logger = get_logger()
@@ -284,9 +284,10 @@ class TE2(NamedTuple):
 
 # you can run one specific test (e.g. to profile) by passing it as -k to pytest
 # e.g. -k 'test_many[500000-False]'
-@pytest.mark.parametrize('count', [100_000, 500_000, 1_000_000])
-def test_many(count: int, tmp_path: Path) -> None:
-    if count > 100_000 and running_on_ci:
+@pytest.mark.parametrize('count', [99, 500_000, 1_000_000])
+@pytest.mark.parametrize('gc_on', [True, False], ids=['gc_on', 'gc_off'])
+def test_many(count: int, tmp_path: Path, gc_control) -> None:
+    if count > 99 and running_on_ci:
         pytest.skip("test would be too slow on CI, only meant to run manually")
     # should be a parametrized test perhaps
     src = tmp_path / 'source'