This support package simplifies data persistence between user sessions and software version updates.
The main idea of this package is simplify data migration between versions, and allow to define migration information next to data structure definition.
If You only need to serialize data, then you could use only JSON hooks
import json
from pydantic import BaseModel
from local_migrator import Encoder, object_hook
class SampleModel(BaseModel):
field1: int
field2: str
data = SampleModel(field1=4, field2="abc")
with open("sample.json", "w") as f_p:
json.dump(data, f_p, cls=Encoder)
with open("sample.json") as f_p:
data2 = json.load(f_p, object_hook=object_hook)
assert data == data2
To register this information there is register_class
decorator.
It has 4 parameters:
version
- version of data structuremigration_list
- list of tuple (version
.migration_function
).old_paths
- list of fully qualified python paths to previous class definitions. This is to allow move class during code refactoring.use_parent_migrations
- if True, then parent class migrations will be used.
Lets imagine that we have such code
from local_migrator import Encoder, object_hook
class SampleModel(BaseModel):
field1: int
field_ca_1: str
field_ca_2: float
with open("sample.json", "w") as f_p:
json.dump(data, f_p, cls=Encoder)
But there is decision to move both ca
field to sub structure:
class CaModel(BaseModel):
field_1: str
field_2: float
class SampleModel(BaseModel):
field1: int
field_ca: CaModel
Then with local_migrator
code may look:
from local_migrator import object_hook, register_class
class CaModel(BaseModel):
field_1: str
field_2: float
def ca_migration_function(dkt):
dkt["field_ca"] = CaModel(field1=dkt.pop("field_ca_1"),
field2=dkt.pop("field_ca_2"))
return dkt
@register_class("0.0.1", [("0.0.1", ca_migration_function)])
class SampleModel(BaseModel):
field1: int
field_ca: CaModel
with open("sample.json") as f_p:
data = json.load(f_p, object_hook=object_hook)
Assume that there is decision to rename field1
to id
.
Then code may look:
from local_migrator import object_hook, register_class, rename_key
class CaModel(BaseModel):
field_1: str
field_2: float
def ca_migration_function(dkt):
dkt["field_ca"] = CaModel(field1=dkt.pop("field_ca_1"),
field2=dkt.pop("field_ca_2"))
return dkt
@register_class("0.0.2", [("0.0.1", ca_migration_function), ("0.0.2", rename_key("field1", "id"))])
class SampleModel(BaseModel):
id: int
field_ca: CaModel
with open("sample.json") as f_p:
data = json.load(f_p, object_hook=object_hook)
More examples could be found in examples section of documentation
rename_key(from_key: str, to_key: str, optional=False) -> Callable[[Dict], Dict]
- helper function for rename field migrations.update_argument(argument_name:str)(func: Callable) -> Callable
- decorator to keep backward compatibility by convertingdict
argument to some class base on function type annotation
Contributions are encouraged! Please create pull request or open issue. For PR please remember to add tests and documentation.
This package is originally named nme
but was rename to clarify its purpose.
This package is extracted from PartSeg project for simplify reuse it in another projects.