Workflow hashes are not consistent when blocks are the same but mixed

[GhislainJ]

**Note: for support questions, please use https://nextcloud.dessia.tech/call/hr9z9bif

• I'm submitting a ...

  • bug report
  • feature request

• What is the current behavior?

Workflow hash computes differently when its blocks sequence is "shuffled"

• If the current behavior is a bug, please provide the steps to reproduce and if possible a minimal demo of the problem Avoid reference to other packages

Build twice the exact same workflow by only changing order of blocks. Compute hash, results are different.

Steps to reproduce :

from dessia_common.workflow import Workflow, InstantiateModel, ModelMethod, Pipe
from dessia_common.forms import Generator
from dessia_common.typings import MethodType

instantiate_block = InstantiateModel(model_class=Generator, name="Generator")

method_type = MethodType(class_=Generator, name="generate")
generate_block = ModelMethod(method_type=method_type)

pipes = [Pipe(instantiate_block.outputs[0], generate_block.inputs[0])]
output = generate_block.outputs[0]

blocks1 = [instantiate_block, generate_block]
workflow1 = Workflow(blocks=blocks1, pipes=pipes, output=output)

blocks2 = [generate_block, instantiate_block]
workflow2 = Workflow(blocks=blocks2, pipes=pipes, output=output)

print("Hash Workflow1 :", hash(workflow1))
print("Hash Workflow2 :", hash(workflow2))

Result :

Hash Workflow1 : 89
Hash Workflow2 : 88

• What is the expected behavior?

Such workflows should be considered equal

• What is the motivation / use case for changing the behavior?

Frontend's workflow builder will really be subject to it, as someone would add blocks not in any particular order.

• Possible fixes
  This looks like it comes from the ouput hash. We sum its variable_indices elements which are bound to block index.
  Should we add a hash computation for Variables ?

• Please tell us about your environment:

  • branch: dev/master
  • commit:
  • python version: python3.9

[GhislainJ]

Actually, we should use graphs for eq/hash, instead of evaluating hash for Variables as proposed