Object properties not persisted in DataLab when modified from a proxy (RemoteProxy / Web API)

[deuterium33]

Describe the bug
Modifying Image object properties, such as (dx, dy, x0, y0), from a RemoteProxy call updates a copy of the object but does not modify the object in DataLab.

To Reproduce

Run the following code:

import numpy as np
data = np.zeros([800,800])
from datalab.control.proxy import RemoteProxy
proxy = RemoteProxy()
proxy.add_image('zeros', data)
o = proxy.get_object()
dx, dy = 2, 2
o.set_uniform_coords(dx, dy)
print([o.dx,o.dy])
# Returns [2.0, 2.0]
o.dx, o.dy = 3, 3 #same thing
print([o.dx,o.dy])
# Returns [3.0, 3.0]

In DataLab, the object keeps its original properties (here, dx=1 and dy=1) despite changes:

https://www.pasteboard.co/vJef0vtufGlt.png

Expected behavior
The object properties are updated when changed from the RemoteProxy, as working currently when doing it from the console:

o = dl.get_object()
o.dx = 2.

https://www.pasteboard.co/XFRcKv-a4L3v.png

Installation information

• DataLab installation type "Python package"
• DataLab v1.1.0

Additional context
Running the RemoteProxy code from Spyder 6.1.3

[PierreRaybaut]

Thanks for the detailed report and for the clear reproduction steps.

This is indeed a real issue, but the root cause is a bit subtle:

• dl.get_object() from the DataLab internal console returns the actual in-memory object stored in the application, so modifying o.dx, o.dy, etc. updates the object directly.
• RemoteProxy.get_object() goes through the remote control layer (XML-RPC), which serializes the object and returns a deserialized copy on the client side.
• As a consequence, changing attributes on the object returned by RemoteProxy.get_object() only modifies the local copy in the client process, not the object stored inside DataLab.

So the current behavior is consistent with the implementation, but it is misleading from a user perspective and should be fixed or at least made explicit.

The proper fix is to add an explicit set_object() method to the remote proxy API.

[deuterium33]

Dear Pierre,

Thank you in return for the prompt answer!

Yes, I was not sure whether to call it a "bug" because I realized it was an expected behavior due to the deserialization. I would also understand the practicality of keeping things desynchronized by default and not running an automatic update each time the local copy is modified. Then, one may want to update the real DataLab object by specifically running a set_object(modified_object) or update_object(modified_object) from the RemoteProxy API. It would be great!

Cheers,
Mathieu

[PierreRaybaut]

Hey @deuterium33, this is done. You may test it if you want.
It will be part the forthcoming v1.2.0 release (scheduled for next Monday).

[deuterium33]

Excellent! Thank you!

The only issue I encountered was a ValueError popping up in the console after running proxy.set_object(obj_modified):
ValueError: Computed item 'xmin' is read-only

To avoid that, I suppose we could edit the set_object definition in gui/panel/base.py as follows:

from guidata.dataset.datatypes import ComputedProp
for item in existing._items:  # pylint: disable=protected-access
    name = item.get_name()
    if not name.startswith("_") and not isinstance(item.get_prop("data", "computed", None), ComputedProp):
        setattr(existing, name, getattr(obj, name))

Cheers,
Mathieu

[PierreRaybaut]

Excellent! Thank you!

The only issue I encountered was a ValueError popping up in the console after running proxy.set_object(obj_modified): ValueError: Computed item 'xmin' is read-only

To avoid that, I suppose we could edit the set_object definition in gui/panel/base.py as follows:

from guidata.dataset.datatypes import ComputedProp
for item in existing._items:  # pylint: disable=protected-access
    name = item.get_name()
    if not name.startswith("_") and not isinstance(item.get_prop("data", "computed", None), ComputedProp):
        setattr(existing, name, getattr(obj, name))

Cheers, Mathieu

Indeed, I've missed this specific case of read-only properties: this is now fixed.
Please do not hesitate to test further.
And thanks again!

Cheers,
Pierre

[deuterium33]

I tested, and it works well!

To also get a refresh of the properties panel after running set_object, I would just add self.objview.item_selection_changed() after self.objview.update_tree() in this method.

Cheers,
Mathieu

[PierreRaybaut]

Excellent suggestion, done!