Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Compare the upgraded studio.db by the rows the old build wrote, not its bytes (#11678)

Daniel Han0c8babaf91e5
diff --git a/studio/backend/tests/multi_account/legacy_probe.py b/studio/backend/tests/multi_account/legacy_probe.py
index 78cd37f42ac..57046dd6755 100644
--- a/studio/backend/tests/multi_account/legacy_probe.py
+++ b/studio/backend/tests/multi_account/legacy_probe.py
@@ -14,6 +14,7 @@ from .seed import (
PASSWORD,
SENTINEL,
THREAD_ID,
+ legacy_studio_rows,
old_auth_row,
seed_legacy_install,
)
@@ -23,6 +24,7 @@ def main() -> None:
home = Path(os.environ["UNSLOTH_STUDIO_HOME"])
original = seed_legacy_install(home)
auth_row = old_auth_row(home / "auth" / "auth.db")
+ studio_rows = legacy_studio_rows(home / "studio.db")
application = importlib.import_module("main")
for name, payload in original.items():
assert (home / name).read_bytes() == payload, f"Import changed {name}"
@@ -66,10 +68,15 @@ def main() -> None:
assert login.json()["access_token"]
finally:
client.close()
+ # auth.db and studio.db gain columns when a current build opens them, so both are compared by what
+ # the old build wrote rather than byte for byte. Everything else must be untouched.
for name, payload in original.items():
- if name != "auth/auth.db":
+ if name not in ("auth/auth.db", "studio.db"):
assert (home / name).read_bytes() == payload, f"Owner read changed {name}"
assert old_auth_row(home / "auth" / "auth.db") == auth_row
+ assert (
+ legacy_studio_rows(home / "studio.db") == studio_rows
+ ), "Owner read changed legacy studio.db rows"
assert not (home / "accounts").exists()
print(json.dumps({"preserved_files": len(original), "owner_login": True}))
diff --git a/studio/backend/tests/multi_account/seed.py b/studio/backend/tests/multi_account/seed.py
index dffd51b714c..252b956440c 100644
--- a/studio/backend/tests/multi_account/seed.py
+++ b/studio/backend/tests/multi_account/seed.py
@@ -129,6 +129,43 @@ def seed_legacy_install(home: Path) -> dict[str, bytes]:
}
+def legacy_studio_columns() -> dict[str, tuple[str, ...]]:
+ """Every table in the frozen studio.db schema, with the columns it had then."""
+ with closing(sqlite3.connect(":memory:")) as conn:
+ conn.executescript(
+ Path(__file__).with_name("legacy_studio_schema.sql").read_text(encoding = "utf-8")
+ )
+ tables = [
+ row[0]
+ for row in conn.execute(
+ "SELECT name FROM sqlite_master WHERE type = 'table' AND name NOT LIKE 'sqlite_%' "
+ "ORDER BY name"
+ )
+ ]
+ return {
+ table: tuple(row[1] for row in conn.execute(f'PRAGMA table_info("{table}")'))
+ for table in tables
+ }
+
+
+def legacy_studio_rows(path: Path) -> dict[str, list[tuple]]:
+ """Every row of every legacy table, read through its legacy columns only.
+
+ A current build adds columns to studio.db as it opens it (`ALTER TABLE ... ADD COLUMN`), which
+ changes the file's bytes but not a byte of what the old build wrote. This is the part an upgrade
+ has to preserve, and the part an old build reading its own named columns would see. A dropped or
+ renamed legacy column raises here rather than comparing equal.
+ """
+ rows = {}
+ with closing(sqlite3.connect(f"file:{path}?mode=ro", uri = True)) as conn:
+ for table, columns in legacy_studio_columns().items():
+ quoted = ", ".join(f'"{column}"' for column in columns)
+ rows[table] = sorted(
+ conn.execute(f'SELECT {quoted} FROM "{table}"').fetchall(), key = repr
+ )
+ return rows
+
+
def old_auth_row(path: Path) -> tuple:
with closing(sqlite3.connect(path)) as conn:
return conn.execute(