Do the methods all_bases() and all_penalties() really make sense?

[hriebl]

Just a minor concern here, but if a smooth really comprises multiple "sub-smoothes", wouldn't the user usually want to retrieve all matrices belonging to one sub-smooth together? Does it make sense to "tear apart" bases and penalties across sub-smoothes?

[jobrachem]

I don't see it really as "tearing apart" the bases and penalties, but rather as outputting them from mgcv with as little processing as possible, since mgcv outputs them individually from mgcv::smoothCon.

At the same time, outputting one big basis and one big penalty would probably be convenient in many cases. I thought about implementing that, but decided against it in the current version mainly because I don't have a full overview of all potential smooth configurations and edge cases that mgcv might output, so I considered the current version to be a bit safer. But of course that does not mean that anything really speaks against it, it might be straightforward.