Migrating from write_dill to write_json¶

write_dill and read_dill are deprecated and will be removed in a future release. This page explains how to migrate to the replacement write_json / read_json API.

Quick substitution¶

The call signatures are the same — both accept a file path or a file-like object:

# Before (deprecated)
comp.write_dill('comp.dill')
comp2 = Computation.read_dill('comp.dill')

# After
comp.write_json('comp.json')
comp2 = Computation.read_json('comp.json')

Key differences¶

Lambdas must be replaced (or opt in to dill)¶

write_dill serialized lambdas via pickle. write_json raises SerializationError if a node's function is a lambda. The cleanest fix is to replace lambdas with module-level functions:

# Before — works with write_dill, fails with write_json
comp.add_node('b', lambda a: a + 1)

# After — works with write_json
def increment(a):
    return a + 1

comp.add_node('b', increment)

If refactoring is impractical, there are two escape hatches:

Option 1 — serialize the value only (function is lost, node cannot be re-run after load):

comp.add_node('b', lambda a: a + 1, serialize=False)

Option 2 — use ComputationSerializer(use_dill_for_functions=True) (function is preserved as a dill blob, re-computation works after load):

from loman import ComputationSerializer

s = ComputationSerializer(use_dill_for_functions=True)
comp.write_json('comp.json', serializer=s)
comp2 = Computation.read_json('comp.json', serializer=s)

The same serializer instance must be used for both write and read. The dill blob is not portable across Python versions.

File-like objects must be text-mode¶

write_dill used binary mode. write_json uses text mode:

import io

# Before
buf = io.BytesIO()
comp.write_dill(buf)

# After
buf = io.StringIO()
comp.write_json(buf)

Custom types¶

If your computation holds values of types that are not handled by the default serializer (anything beyond Python scalars, lists, dicts, NumPy arrays, and Pandas DataFrames/Series), you need to register a custom transformer. See the Serializing Computations page for an example.