refreshing ghost node values after lss solve

[tbanyai]

Should lss be providing the synchronization of the solution after solve or does it make more sense to call a collective sync outside?

[tlmquintino]

collective sync outside. leave it to the solver to decide.

[tbanyai]

pro: the data has to be copied out from the lss vector and this could be done together with the sync
kontra: this results in of those little built-in inconveniencies which at the end can mess with a developer who doesnt know the "rules of the house"

[tlmquintino]

yep. the principle of least astonishment http://en.wikipedia.org/wiki/Principle_of_least_astonishment

[tbanyai]

gives the advantage that if deltau is always synchronized and u was synchronized initially, then you dont need to synchronize on cf3 side.

[tlmquintino]

Good point!
But I don't consider that a stronger reason than avoid "unexpected" behavior (again PLA).

[wdeconinck]

Agree here with tlmquintino. Unexpected behavior is to be avoided. No big deal for a developer to do the refresh separately.

Consensus?