Martin Maechler wrote:
>
> Your patch is basically only affecting the default
> method = "pearson". For (most) other cases, 'y = NULL' would
> still remain *the* way to save computations, unless we'd start
> to use an R-level equivalent [which I think does not exist] of
> your C trick (DATAPTR(x) == DATAPTR(y)).
>
>
yes, my patch was constrained to the c code, but i don't think it would
be particularly difficult to fix the relevant r-level code as well. i
did think about it, but didn't want to invest more time in this until
(or unless) someone would respond. (thanks for the response.)
> Also, for S- and R- backcompatibility reasons, we'd need to
> continue allowing y = NULL (as your patch would, too),
only in its current for -- indeed, the (unimplemented) intention was to
detach from the old misdesign, and fix everything so that y=x by default
anywhere.
> so
> currently I think this whole idea -- as slick as it is, I
> learned something! --
> does not make sense applying here.
>
i think it does, because the current state is somewhat funny, including
both the difference in performance between var(x) and var(x,x) (with x
being a matrix), and the respective comment in ?var.
> > the attached patch suggests modifications to src/main/cov.c and
> > src/library/stats/man/cor.Rd.
>
> BTW: since you didn't (and shouldn't , because of method != "pearson" !)
> change the R code,
i would suggest it be done, though.
> the docs \usage{.} part should not have been
> changed either !
>
indeed, the change in the docs didn't match what i *have* actually fixed
in the code.
> and as I mentioned: using 'y = NULL' in the function call must
>
*MUST* ?
> continue to work, hence should also be documented as
> possibility
> ==> the docs would not really become more clear, I think
>
no, of course, without the change in r code having the docs say y=x by
default would be a nonsense. but again, this was a start, not a
complete modification (and i admit i failed to acknowledge this).
vQ
______________________________________________
R-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
