R devel - May 2018 - Dispatch mechanism seems to alter object before calling method on it

Hi,

This was quite unexpected:

   setGeneric("foo", function(x) standardGeneric("foo"))

   setMethod("foo", "vector", identity)

   foo(matrix(1:12, ncol=3))
   # [1]  1  2  3  4  5  6  7  8  9 10 11 12

   foo(array(1:24, 4:2))
   # [1]  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 
22 23 24

If I define a method for array objects, things work as expected though:

   setMethod("foo", "array", identity)

   foo(matrix(1:12, ncol=3))
   #      [,1] [,2] [,3]
   # [1,]    1    5    9
   # [2,]    2    6   10
   # [3,]    3    7   11
   # [4,]    4    8   12

So, luckily, I have a workaround.

But shouldn't the dispatch mechanism stay away from the business of
altering objects before passed to it?

Thanks,
H.

-- 
Herv? Pag?s

Program in Computational Biology
Division of Public Health Sciences
Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N, M1-B514
P.O. Box 19024
Seattle, WA 98109-1024

E-mail: hpages at fredhutch.org
Phone:  (206) 667-5791
Fax:    (206) 667-1319

My understanding is that array (or any other structure) does not
"simply" inherit from vector, because structures are not vectors in
the strictest sense. Basically, once a vector gains attributes, it is
a structure, not a vector. The methods package accommodates this by
defining an "is" relationship between "structure" and
"vector" via an
"explicit coerce", such that any "structure" passed to a
"vector"
method is first passed to as.vector(), which strips attributes. This
is very much by design.

Michael


On Tue, May 15, 2018 at 5:25 PM, Herv? Pag?s <hpages at fredhutch.org>
wrote:
> Hi,
>
> This was quite unexpected:
>
>   setGeneric("foo", function(x) standardGeneric("foo"))
>
>   setMethod("foo", "vector", identity)
>
>   foo(matrix(1:12, ncol=3))
>   # [1]  1  2  3  4  5  6  7  8  9 10 11 12
>
>   foo(array(1:24, 4:2))
>   # [1]  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22
23
> 24
>
> If I define a method for array objects, things work as expected though:
>
>   setMethod("foo", "array", identity)
>
>   foo(matrix(1:12, ncol=3))
>   #      [,1] [,2] [,3]
>   # [1,]    1    5    9
>   # [2,]    2    6   10
>   # [3,]    3    7   11
>   # [4,]    4    8   12
>
> So, luckily, I have a workaround.
>
> But shouldn't the dispatch mechanism stay away from the business of
> altering objects before passed to it?
>
> Thanks,
> H.
>
> --
> Herv? Pag?s
>
> Program in Computational Biology
> Division of Public Health Sciences
> Fred Hutchinson Cancer Research Center
> 1100 Fairview Ave. N, M1-B514
> P.O. Box 19024
> Seattle, WA 98109-1024
>
> E-mail: hpages at fredhutch.org
> Phone:  (206) 667-5791
> Fax:    (206) 667-1319
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>

On 05/15/2018 09:13 PM, Michael Lawrence wrote:
> My understanding is that array (or any other structure) does not
> "simply" inherit from vector, because structures are not vectors
in
> the strictest sense. Basically, once a vector gains attributes, it is
> a structure, not a vector. The methods package accommodates this by
> defining an "is" relationship between "structure" and
"vector" via an
> "explicit coerce", such that any "structure" passed to
a "vector"
> method is first passed to as.vector(), which strips attributes. This
> is very much by design.
It seems that the problem is really with matrices and arrays, not
with "structures" in general:

   f <- factor(c("z", "x", "z"),
levels=letters)
   m <- matrix(1:12, ncol=3)
   df <- data.frame(f=f)
   x <- structure(1:3, titi="A")

Only the matrix looses its attributes when passed to a "vector"
method:

   setGeneric("foo", function(x) standardGeneric("foo"))
   setMethod("foo", "vector", identity)

   foo(f)     # attributes are preserved
   # [1] z x z
   # Levels: a b c d e f g h i j k l m n o p q r s t u v w x y z

   foo(m)     # attributes are stripped
   # [1]  1  2  3  4  5  6  7  8  9 10 11 12

   foo(df)    # attributes are preserved
   #   f
   # 1 z
   # 2 x
   # 3 z

   foo(x)     # attributes are preserved
   # [1] 1 2 3
   # attr(,"titi")
   # [1] "A"

Also if structures are passed to as.vector() before being passed to
a "vector" method, shouldn't as.vector() and foo() be equivalent
on
them? For 'f' and 'x' they're not:

   as.vector(f)
   # [1] "z" "x" "z"

   as.vector(x)
   # [1] 1 2 3

Finally note that for factors and data frames the "vector" method gets
selected despite the fact that is( , "vector") is FALSE:

   is(f, "vector")
   # [1] FALSE

   is(m, "vector")
   # [1] TRUE

   is(df, "vector")
   # [1] FALSE

   is(x, "vector")
   # [1] TRUE

Couldn't we recognize these problems as real, even if they are by
design? Hopefully we can all agree that:
- the dispatch mechanism should only dispatch, not alter objects;
- is() and selectMethod() should not contradict each other.

Thanks,
H.
> 
> Michael
> 
> 
> On Tue, May 15, 2018 at 5:25 PM, Herv? Pag?s <hpages at
fredhutch.org> wrote:
>> Hi,
>>
>> This was quite unexpected:
>>
>>    setGeneric("foo", function(x)
standardGeneric("foo"))
>>
>>    setMethod("foo", "vector", identity)
>>
>>    foo(matrix(1:12, ncol=3))
>>    # [1]  1  2  3  4  5  6  7  8  9 10 11 12
>>
>>    foo(array(1:24, 4:2))
>>    # [1]  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21
22 23
>> 24
>>
>> If I define a method for array objects, things work as expected though:
>>
>>    setMethod("foo", "array", identity)
>>
>>    foo(matrix(1:12, ncol=3))
>>    #      [,1] [,2] [,3]
>>    # [1,]    1    5    9
>>    # [2,]    2    6   10
>>    # [3,]    3    7   11
>>    # [4,]    4    8   12
>>
>> So, luckily, I have a workaround.
>>
>> But shouldn't the dispatch mechanism stay away from the business of
>> altering objects before passed to it?
>>
>> Thanks,
>> H.
>>
>> --
>> Herv? Pag?s
>>
>> Program in Computational Biology
>> Division of Public Health Sciences
>> Fred Hutchinson Cancer Research Center
>> 1100 Fairview Ave. N, M1-B514
>> P.O. Box 19024
>> Seattle, WA 98109-1024
>>
>> E-mail: hpages at fredhutch.org
>> Phone:  (206) 667-5791
>> Fax:    (206) 667-1319
>>
>> ______________________________________________
>> R-devel at r-project.org mailing list
>>
https://urldefense.proofpoint.com/v2/url?u=https-3A__stat.ethz.ch_mailman_listinfo_r-2Ddevel&d=DwIFaQ&c=eRAMFD45gAfqt84VtBcfhQ&r=BK7q3XeAvimeWdGbWY_wJYbW0WYiZvSXAJJKaaPhzWA&m=gynT4YhbmVKZhnX4srXlCWZZRyVBMXG211CKgftdEs0&s=_I0aFHQVnXdBfB5kTLg9TxK_2LHdSuaB6gqZwSx1orQ&e>>
-- 
Herv? Pag?s

Program in Computational Biology
Division of Public Health Sciences
Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N, M1-B514
P.O. Box 19024
Seattle, WA 98109-1024

E-mail: hpages at fredhutch.org
Phone:  (206) 667-5791
Fax:    (206) 667-1319