R devel - Aug 2022 - simplify2array assumes non-NA dim() -- base or method bug?

Hmm. I think I haven't seen NA dims in base R before. R-lang says
> The 'dim' attribute is used to implement arrays.  The content of
> the array is stored in a vector in column-major order and the 'dim'
> attribute is a vector of integers specifying the respective extents of
> the array.  R ensures that the length of the vector is the product of
> the lengths of the dimensions.
And R-intro says:
> A dimension vector is a vector of non-negative integers.
I understand "vector of [non-negative] integers" as excluding
NA_integer_.

Correspondingly:
> a <- 1
> dim(a) <- NA_integer_
Error in dim(a) <- NA_integer_ : the dims contain missing values
> matrix(1, NA_integer_)
Error in matrix(1, NA_integer_) : invalid 'nrow' value (too large or NA)
> array(1, NA_integer_)
Error in array(1, NA_integer_) : negative length vectors are not allowed

There must be many more examples where non-NULL dim() is assumed to
not contain missing values.

simplify2array() aligns with that specification and only needs to check
for numeric (non-NULL) dims at this point and not also !anyNA(c.dim).
So my take is that there is no bug.

Best regards,

	Sebastian Meyer


Am 25.08.22 um 04:51 schrieb Michael Chirico via
R-devel:
> If any component of the dim() returned as c.dim here[1] are missing,
> simplify2array() errors inscrutably (specifically, because the last
&&
> condition is missing):
> 
> Error in if (higher && length(c.dim <- unique(lapply(x, dim)))
== 1 &&  :
>    missing value where TRUE/FALSE needed
> 
> At root here is that dim.tbl_lazy ({dbplyr} package method [2]) very
> intentionally neglects to count the # of rows in the result -- the
> whole point of a lazy table is to avoid calculating full queries
> unless specifically requested, so the # of rows is left as missing,
> i.e., there _is_ a quantity of rows, but the exact number is not
> known.
> 
> That seems to me like a proper usage of NA, and hence this is a
> simplify2array() bug, but I am curious other thoughts here before
> attempting a patch.
> 
> [1]
https://github.com/r-devel/r-svn/blob/2d4f8c283d53ff2c98d92c7b77b11e652297742c/src/library/base/R/sapply.R#L46-L48
> [2]
https://github.com/tidyverse/dbplyr/blob/36b146e36d6d9af215dc48e60862d4b807b9e606/R/tbl-lazy.R#L45-L47
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel

Does this somewhat contradict the requirement that dimensions are non-missing?

ncol(table(letters))
# [1] NA

On the other hand,

dim(table(letters))
# [1] 26

Maybe that's a bug for the ncol() method of table?

On Thu, Aug 25, 2022 at 2:55 AM Sebastian Meyer <seb.meyer at fau.de>
wrote:
>
> Hmm. I think I haven't seen NA dims in base R before. R-lang says
>
> > The 'dim' attribute is used to implement arrays.  The content
of
> > the array is stored in a vector in column-major order and the
'dim'
> > attribute is a vector of integers specifying the respective extents of
> > the array.  R ensures that the length of the vector is the product of
> > the lengths of the dimensions.
>
> And R-intro says:
>
> > A dimension vector is a vector of non-negative integers.
>
> I understand "vector of [non-negative] integers" as excluding
NA_integer_.
>
> Correspondingly:
>
> > a <- 1
> > dim(a) <- NA_integer_
>
> Error in dim(a) <- NA_integer_ : the dims contain missing values
>
> > matrix(1, NA_integer_)
>
> Error in matrix(1, NA_integer_) : invalid 'nrow' value (too large
or NA)
>
> > array(1, NA_integer_)
>
> Error in array(1, NA_integer_) : negative length vectors are not allowed
>
> There must be many more examples where non-NULL dim() is assumed to
> not contain missing values.
>
> simplify2array() aligns with that specification and only needs to check
> for numeric (non-NULL) dims at this point and not also !anyNA(c.dim).
> So my take is that there is no bug.
>
> Best regards,
>
>         Sebastian Meyer
>
>
> Am 25.08.22 um 04:51 schrieb Michael Chirico via R-devel:
> > If any component of the dim() returned as c.dim here[1] are missing,
> > simplify2array() errors inscrutably (specifically, because the last
&&
> > condition is missing):
> >
> > Error in if (higher && length(c.dim <- unique(lapply(x,
dim))) == 1 &&  :
> >    missing value where TRUE/FALSE needed
> >
> > At root here is that dim.tbl_lazy ({dbplyr} package method [2]) very
> > intentionally neglects to count the # of rows in the result -- the
> > whole point of a lazy table is to avoid calculating full queries
> > unless specifically requested, so the # of rows is left as missing,
> > i.e., there _is_ a quantity of rows, but the exact number is not
> > known.
> >
> > That seems to me like a proper usage of NA, and hence this is a
> > simplify2array() bug, but I am curious other thoughts here before
> > attempting a patch.
> >
> > [1]
https://github.com/r-devel/r-svn/blob/2d4f8c283d53ff2c98d92c7b77b11e652297742c/src/library/base/R/sapply.R#L46-L48
> > [2]
https://github.com/tidyverse/dbplyr/blob/36b146e36d6d9af215dc48e60862d4b807b9e606/R/tbl-lazy.R#L45-L47
> >
> > ______________________________________________
> > R-devel at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-devel