

Hi all,
I came across an issue in using the Matrix package which made it that I
could only subset Matrices using the numeric class, but could not using
integers. Steps to reproduce the problem:
library(Matrix)
# this class has *nothing* to do with Matrix
setClass("MyClass",
representation(myvalue = "numeric"),
prototype(myvalue = NA_real_))
# this class also has nothing to do with Matrix
setClassUnion("NumOrMyClass", c("numeric", "MyClass"))
# this is to demonstrate creating and subsetting a specific Matrix
m < new("dgCMatrix", i = c(2L, 0L, 1L, 2L, 0L, 1L), p = c(0L, 1L, 2L, 4L,
4L, 6L),
Dim = c(3L, 5L), Dimnames = list(NULL, NULL), x = c(2, 1, 2, 1, 2, 1),
factors = list())
# this next line fails:
m[1:2, 1:2]
The last line fails consistently with R 3.6.1 and Matrix 1.217 on a
variety of operating systems and computers (tested with ubuntu 19.10 &
windows 10). I'll note at this point that without the class union I defined
above, the rest of the code works perfectly fine.
There are 2 workarounds I've come up with. Firstly, one could change the
last line to
m[as.numeric(1:2), as.numeric(1:2)]
or as.numeric(seq()) or c(1, 2), etc.
The other workaround is to change setClassUnion("NumOrMyClass",
c("numeric", "MyClass")) to setClassUnion("NumOrMyClass", c("numeric",
"double", "integer", "MyClass")), which I believe should be the recommended
workaround here.
The underlying issue, though, is that I would not have expected *anything*
that I write to get in the way of Matrix; in fact, the classes that I
defined have nothing to do with Matrix at all. So, I have some fears that
this might be a bigger problem, possibly either in Methods or in Base.
1. Can others confirm that this is in fact an issue as I believe it to be?
2. If it is a legitimate issue, have others seen it manifested working with
other packages that are not Matrix? In this case, I believe that the class
union I defied is somehow interfering with the "index" class union in
Matrix, R/AllClass.R, line 809
3. Why should the class union I defined interfere with the inner workings
of a separate package?
4. Is this a bug in Base or Methods?
Thank you for your time!
Sincerely,
Ezra
[[alternative HTML version deleted]]
______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/rdevel


>>>>> Ezra Tucker
>>>>> on Mon, 11 Nov 2019 21:47:41 +0000 writes:
> Hi all,
>
> I came across an issue in using the Matrix package which made it that I
> could only subset Matrices using the numeric class, but could not using
> integers. Steps to reproduce the problem:
>
> library(Matrix)
>
> # this class has *nothing* to do with Matrix
> setClass("MyClass",
> representation(myvalue = "numeric"),
> prototype(myvalue = NA_real_))
> # this class also has nothing to do with Matrix
> setClassUnion("NumOrMyClass", c("numeric", "MyClass"))
>
> # this is to demonstrate creating and subsetting a specific Matrix
> m < new("dgCMatrix", i = c(2L, 0L, 1L, 2L, 0L, 1L), p = c(0L, 1L, 2L, 4L,
> 4L, 6L),
> Dim = c(3L, 5L), Dimnames = list(NULL, NULL), x = c(2, 1, 2, 1, 2, 1),
> factors = list())
> # this next line fails:
> m[1:2, 1:2]
>
> The last line fails consistently with R 3.6.1 and Matrix 1.217 on a
> variety of operating systems and computers (tested with ubuntu 19.10 &
> windows 10). I'll note at this point that without the class union I defined
> above, the rest of the code works perfectly fine.
>
> There are 2 workarounds I've come up with. Firstly, one could change the
> last line to
> m[as.numeric(1:2), as.numeric(1:2)]
> or as.numeric(seq()) or c(1, 2), etc.
>
> The other workaround is to change setClassUnion("NumOrMyClass",
> c("numeric", "MyClass")) to setClassUnion("NumOrMyClass", c("numeric",
> "double", "integer", "MyClass")), which I believe should be the recommended
> workaround here.
>
> The underlying issue, though, is that I would not have expected *anything*
> that I write to get in the way of Matrix; in fact, the classes that I
> defined have nothing to do with Matrix at all. So, I have some fears that
> this might be a bigger problem, possibly either in Methods or in Base.
>
> 1. Can others confirm that this is in fact an issue as I believe it to be?
yes, others have confirmed this is an issue .. (unfortunately
not in this Rdevel thread).
Thank you, Ezra, very much for your helpful report, including a
simple reproducible example.
> 2. If it is a legitimate issue, have others seen it manifested working with
> other packages that are not Matrix? In this case, I believe that the class
> union I defied is somehow interfering with the "index" class union in
> Matrix, R/AllClass.R, line 809
> 3. Why should the class union I defined interfere with the inner workings
> of a separate package?
There is no good reason ...
> 4. Is this a bug in Base or Methods?
This is a bug in "base R", in package 'methods'.
The R core team had taken the issue up, already two weeks ago,
but unfortunately did not get to address this in a definitive
way. ==> I'll remind us about it !
Martin Maechler
ETH Zurich and R Core
> Thank you for your time!
>
> Sincerely,
> Ezra
______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/rdevel


I've been working on this and will hopefully get a fix checked in today.
On Wed, Nov 27, 2019 at 3:53 AM Martin Maechler
< [hidden email]> wrote:
>
> >>>>> Ezra Tucker
> >>>>> on Mon, 11 Nov 2019 21:47:41 +0000 writes:
>
> > Hi all,
> >
> > I came across an issue in using the Matrix package which made it that I
> > could only subset Matrices using the numeric class, but could not using
> > integers. Steps to reproduce the problem:
> >
> > library(Matrix)
> >
> > # this class has *nothing* to do with Matrix
> > setClass("MyClass",
> > representation(myvalue = "numeric"),
> > prototype(myvalue = NA_real_))
> > # this class also has nothing to do with Matrix
> > setClassUnion("NumOrMyClass", c("numeric", "MyClass"))
> >
> > # this is to demonstrate creating and subsetting a specific Matrix
> > m < new("dgCMatrix", i = c(2L, 0L, 1L, 2L, 0L, 1L), p = c(0L, 1L, 2L, 4L,
> > 4L, 6L),
> > Dim = c(3L, 5L), Dimnames = list(NULL, NULL), x = c(2, 1, 2, 1, 2, 1),
> > factors = list())
> > # this next line fails:
> > m[1:2, 1:2]
> >
> > The last line fails consistently with R 3.6.1 and Matrix 1.217 on a
> > variety of operating systems and computers (tested with ubuntu 19.10 &
> > windows 10). I'll note at this point that without the class union I defined
> > above, the rest of the code works perfectly fine.
> >
> > There are 2 workarounds I've come up with. Firstly, one could change the
> > last line to
> > m[as.numeric(1:2), as.numeric(1:2)]
> > or as.numeric(seq()) or c(1, 2), etc.
> >
> > The other workaround is to change setClassUnion("NumOrMyClass",
> > c("numeric", "MyClass")) to setClassUnion("NumOrMyClass", c("numeric",
> > "double", "integer", "MyClass")), which I believe should be the recommended
> > workaround here.
> >
> > The underlying issue, though, is that I would not have expected *anything*
> > that I write to get in the way of Matrix; in fact, the classes that I
> > defined have nothing to do with Matrix at all. So, I have some fears that
> > this might be a bigger problem, possibly either in Methods or in Base.
> >
> > 1. Can others confirm that this is in fact an issue as I believe it to be?
>
> yes, others have confirmed this is an issue .. (unfortunately
> not in this Rdevel thread).
>
> Thank you, Ezra, very much for your helpful report, including a
> simple reproducible example.
>
> > 2. If it is a legitimate issue, have others seen it manifested working with
> > other packages that are not Matrix? In this case, I believe that the class
> > union I defied is somehow interfering with the "index" class union in
> > Matrix, R/AllClass.R, line 809
>
> > 3. Why should the class union I defined interfere with the inner workings
> > of a separate package?
>
> There is no good reason ...
>
> > 4. Is this a bug in Base or Methods?
>
> This is a bug in "base R", in package 'methods'.
>
> The R core team had taken the issue up, already two weeks ago,
> but unfortunately did not get to address this in a definitive
> way. ==> I'll remind us about it !
>
> Martin Maechler
> ETH Zurich and R Core
>
>
> > Thank you for your time!
> >
> > Sincerely,
> > Ezra
>
> ______________________________________________
> [hidden email] mailing list
> https://stat.ethz.ch/mailman/listinfo/rdevel
Michael Lawrence
Senior Scientist, Bioinformatics and Computational Biology
Genentech, A Member of the Roche Group
Office +1 (650) 2257760
[hidden email]
Join Genentech on LinkedIn  Twitter  Facebook  Instagram  YouTube
______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/rdevel

