Aaron Lun

Just spent the last hour trying that. It's even worse than I expected, I can't override the `setAs` even if I wanted. Consider the following example (same as suggested above):...

While that happens... > This was at least 7-8 years ago so I don't remember the details but I think that having the `rowRanges()` getter slap the `elementMetadata` onto the...

Yes, I remember it well, there was all that `SummarizedExperiment0` business. Gee, 8 years. I was but a callow grad student. Brings back memories.

Same for `type=` for `rowQuantiles` and `colQuantiles`, though in this case, it seems like it would be best to respond to that choice; you might just copy the loop over...

Similarly, `colRanks` has `preserve.shape=` instead of `preserveShape=`. Why not put these arguments in the generic so that they are guaranteed to be consistent across methods?

An expedient solution might be to swap the order of `preserveShape` and `dim.` in `.default_colRanks`.

Seems like the choice is between that of developer-level surprise, where we have to make sure that the call inside `.default_colRanks` swaps the arguments; or user-level surprise, where people might...

I'll just mention that I've been obsessively using `center=` with some precomputed row means under the assumption that it was more efficient. But upon inspecting the code, it's not always...

FYI the BumpyMatrix class may be helpful here, for storing non-scalar elements of the same type/class in each cell of a matrix.

FWIW I eventually debugged it by modifying the **DelayedArray** source to eliminate the `try()` altogether in `DelayedOp-class.R`. Just slapping `res` onto the message wasn't enough, unfortunately, as I needed the...