Making workspaces immutable?

Question

Currently workspaces are mutable:

import pyhf
spec = {
    "channels": [
        {
            "name": "Signal Region",
            "samples": [
                {
                    "data": [10],
                    "modifiers": [
                        {"data": [1], "name": "staterror_Signal-Region", "type": "staterror"},
                        {"data": None, "name": "Signal strength",  "type": "normfactor"},
                    ],
                    "name": "Signal",
                }
            ],
        }
    ],
    "measurements": [
        {
            "config": {
                "parameters": [{"name": "staterror_Signal-Region", "fixed": True, "inits": [1.1]}],
                "poi": "Signal strength",
            },
            "name": "fit",
        }
    ],
    "observations": [{"data": [475], "name": "Signal Region"}],
    "version": "1.0.0",
}
ws = pyhf.Workspace(spec)
model = ws.model()
print(model.config.suggested_init())  # -> [1.1, 1.0]
spec["measurements"][0]["config"]["parameters"][0]["inits"] = [1.2]
print(model.config.suggested_init())  # -> [1.1, 1.0]
model = ws.model()
print(model.config.suggested_init())  # -> [1.2, 1.0]

Models on the other hand are immutable. What do you think about making workspaces immutable as well? While I do not have a concrete example in mind, I imagine it could prevent some difficult-to-debug scenarios where the workspace changes without the user noticing.

Relevant Issues and Pull Requests

none

[lukasheinrich]

I thiink a simple change like

spec_copy = copy.deepcopy(spec)
super(Workspace, self).__init__(spec_copy, channels=spec_copy['channels'])

should suffice

[matthewfeickert]

@lukasheinrich Sorry hadn’t taken the time to read all the other replies when I copy pasted that in. That would be great.