Impact & reversibility of disabling "Use semantic models across workspaces"

[nixonba2]

Hello,

After reading the documenation I still don't understand precisely what disabling "Use semantic models across workspaces" will do in my tenant, so I don't know how best to prepare. I know we want this disabled for information security reasons, but I want to ensure a smooth transition for the users. I considered briefly switching it off to run some tests then switch it back on, but I cannot establish if it is strongly reversible. My questions are:

1. In general, does this setting only prevent the creation of new cross-workspace content, or does it block access to (or delete) existing reports in shared workspaces that are connected to semantic models in other workspaces?
2. Does this setting apply to "My Workspace" workspaces in the same way as shared workspaces? That is, will it block access to (or delete) reports in users' "My Workspace" that are connected to semantic models in other workspaces?
3. Does this setting override share links that grant users access to content in workspaces of which they are not a member?
4. Is this setting strongly reversible or does turning it off remove content or permissions that would not be restored were it to be turned back on?

Thank you,

[nandic]

@nixonba2 i find this link much more useful: https://learn.microsoft.com/en-us/power-bi/connect-data/service-datasets-across-workspaces

In general, if you want to have 1 central dataset and want to allow report creators from multiple teams to create reports based on that central datasets and publish them to different workspaces, you need to enable this option.

Otherwise, you would need to have the same dataset in multiple workspaces and each of them needs to be refreshed.

I haven't seen any issues with this option, i find sharing dataset across workspaces very useful for centralized reporting. 
Refresh is taking only once and if there are any changes, you make them only on central dataset.
Otherwise, you would need to fix same things on multiple datasets.

Cheers,
Nemanja Andic