Yes. I'm not sure I fully understand unionfs yet, but it seems to me that it is preferable to using symbolic links for whole subtrees (a la extrootfs) for the following reason.

With the symbolic link approach, the risk you take is that something important about the contents/layout of the linked subtree may change in the next OS upgrade. After such an upgrade, when you run the script to recreate the links pointing to your MMC, you will effectively back out those important changes, and this may result in some part of the OS not working properly.

With the unionfs approach, there is still a similar risk, but only at the level of individual files, not of whole subtrees. (For example if you install something that causes a modified version of the dpkg status file to be written to your MMC, shadowing the one on the internal memory, and then an OS upgrade changes something in the shipped status file, you risk missing that change.)

So in summary, even though there is risk both ways, it seems to me that there's less scope for problematic conflicts with unionfs than there is with symbolic links.

Does that all sound correct, or have I missed some other factors here that would modify the risks?

Finally, a plain question. Does the unionfs approach require that the root filesystem is initially mounted (i.e. during boot) as part of a unionfs, or is it possible to convert the existing, mounted root filesystem to unionfs later on? (I'd be interested to see the precise commands, if someone can post them.)

Regards - Neil