When there were few of these it made sense to have the extractions be a method on *Dense or other matrix types. However, now (partly due to me) there is a proliferation of these on *Dense. Having just had a look at stat.CC I'm not averse to the approach that is used there with Left and Right. We should consider migrating to this approach (at least for some/most? of the decompositions).
This change would have the advantage of keeping the documentation around the decompositions together for each type of decomposition, would reduce the number of methods on *Dense and would avoid giving the impression that there are blessed decompositions (that packages outside mat64 cannot match).
When there were few of these it made sense to have the extractions be a method on
*Dense
or other matrix types. However, now (partly due to me) there is a proliferation of these on*Dense
. Having just had a look atstat.CC
I'm not averse to the approach that is used there withLeft
andRight
. We should consider migrating to this approach (at least for some/most? of the decompositions).This change would have the advantage of keeping the documentation around the decompositions together for each type of decomposition, would reduce the number of methods on
*Dense
and would avoid giving the impression that there are blessed decompositions (that packages outside mat64 cannot match).