setMethod sometimes fails to set package slot when signature has trailing 'ANY'

[Martin Morgan]

after this

   setClass("A")

   setGeneric("bar", function(x, y) standardGeneric("bar"))
   setMethod(bar, signature(x="A", y="A"), function(x, y) {})
   setMethod(bar, signature(x="A", y="ANY"), function(x, y) {})

the method for signature c("A", "A") contains package information

 > str(getMethod(bar, c("A", "A")))
Formal class 'MethodDefinition' [package "methods"] with 4 slots
   ..@ .Data  :function (x, y)
   ..@ target :Formal class 'signature' [package "methods"] with 3 slots
   .. .. ..@ .Data  : chr [1:2] "A" "A"
   .. .. ..@ names  : chr [1:2] "x" "y"
   .. .. ..@ package: chr [1:2] ".GlobalEnv" ".GlobalEnv"
   ..@ defined:Formal class 'signature' [package "methods"] with 3 slots
   .. .. ..@ .Data  : chr [1:2] "A" "A"
   .. .. ..@ names  : chr [1:2] "x" "y"
   .. .. ..@ package: chr [1:2] ".GlobalEnv" ".GlobalEnv"
   ..@ generic: atomic [1:1] bar
   .. ..- attr(*, "package")= chr ".GlobalEnv"

but the method for signature c("A", "ANY") does not

 > str(getMethod(bar, c("A", "ANY")))
Formal class 'MethodDefinition' [package "methods"] with 4 slots
   ..@ .Data  :function (x, y)
   ..@ target :Formal class 'signature' [package "methods"] with 3 slots
   .. .. ..@ .Data  : chr [1:2] "A" "ANY"
   .. .. ..@ names  : chr [1:2] "x" "y"
   .. .. ..@ package: chr [1:2] "" ""
   ..@ defined:Formal class 'signature' [package "methods"] with 3 slots
   .. .. ..@ .Data  : chr [1:2] "A" "ANY"
   .. .. ..@ names  : chr [1:2] "x" "y"
   .. .. ..@ package: chr [1:2] "" ""
   ..@ generic: atomic [1:1] bar
   .. ..- attr(*, "package")= chr ".GlobalEnv"

This does not occur when the order of setMethod calls is reversed. The
reason is that the package information added when
methods/R/Methods.R:552 calls matchSignature are stripped (because
matchSignature returns a vector with trailing 'ANY' classes removed) by
the call to .matchSigLength at line 603.

The context for this is the thread starting at

https://stat.ethz.ch/pipermail/bioconductor/2012-August/047241.html

where a complicated package dependency and attempt to redefine an
existing method leads to

Error : package slot missing from signature for generic 'sampleNames<-'
and classes LumiBatch, ANY
cannot use with duplicate class names (the package may need to be
re-installed)

Martin
--
Computational Biology / Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N.
PO Box 19024 Seattle, WA 98109

Location: Arnold Building M1 B861
Phone: [phone]

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

[Martin Morgan]

On 08/07/2012 02:43 PM, John Chambers wrote:
> Good catch.  This should have been fixed in r-devel, revision 60192.  If
> nothing bad results in other tests, we'll port it to the current patched
> version.

Thank you for the fix. Martin

>
> Thanks,
>    John
>
> On 8/5/12 4:31 PM, Martin Morgan wrote:
>> after this
>>
>>   setClass("A")
>>
>>   setGeneric("bar", function(x, y) standardGeneric("bar"))
>>   setMethod(bar, signature(x="A", y="A"), function(x, y) {})
>>   setMethod(bar, signature(x="A", y="ANY"), function(x, y) {})
>>
>> the method for signature c("A", "A") contains package information
>>
>> > str(getMethod(bar, c("A", "A")))
>> Formal class 'MethodDefinition' [package "methods"] with 4 slots
>>   ..@ .Data  :function (x, y)
>>   ..@ target :Formal class 'signature' [package "methods"] with 3 slots
>>   .. .. ..@ .Data  : chr [1:2] "A" "A"
>>   .. .. ..@ names  : chr [1:2] "x" "y"
>>   .. .. ..@ package: chr [1:2] ".GlobalEnv" ".GlobalEnv"
>>   ..@ defined:Formal class 'signature' [package "methods"] with 3 slots
>>   .. .. ..@ .Data  : chr [1:2] "A" "A"
>>   .. .. ..@ names  : chr [1:2] "x" "y"
>>   .. .. ..@ package: chr [1:2] ".GlobalEnv" ".GlobalEnv"
>>   ..@ generic: atomic [1:1] bar
>>   .. ..- attr(*, "package")= chr ".GlobalEnv"
>>
>> but the method for signature c("A", "ANY") does not
>>
>> > str(getMethod(bar, c("A", "ANY")))
>> Formal class 'MethodDefinition' [package "methods"] with 4 slots
>>   ..@ .Data  :function (x, y)
>>   ..@ target :Formal class 'signature' [package "methods"] with 3 slots
>>   .. .. ..@ .Data  : chr [1:2] "A" "ANY"
>>   .. .. ..@ names  : chr [1:2] "x" "y"
>>   .. .. ..@ package: chr [1:2] "" ""
>>   ..@ defined:Formal class 'signature' [package "methods"] with 3 slots
>>   .. .. ..@ .Data  : chr [1:2] "A" "ANY"
>>   .. .. ..@ names  : chr [1:2] "x" "y"
>>   .. .. ..@ package: chr [1:2] "" ""
>>   ..@ generic: atomic [1:1] bar
>>   .. ..- attr(*, "package")= chr ".GlobalEnv"
>>
>> This does not occur when the order of setMethod calls is reversed. The
>> reason is that the package information added when
>> methods/R/Methods.R:552 calls matchSignature are stripped (because
>> matchSignature returns a vector with trailing 'ANY' classes removed)
>> by the call to .matchSigLength at line 603.
>>
>> The context for this is the thread starting at
>>
>> https://stat.ethz.ch/pipermail/bioconductor/2012-August/047241.html
>>
>> where a complicated package dependency and attempt to redefine an
>> existing method leads to
>>
>> Error : package slot missing from signature for generic 'sampleNames<-'
>> and classes LumiBatch, ANY
>> cannot use with duplicate class names (the package may need to be
>> re-installed)
>>
>> Martin
>

--
Computational Biology / Fred Hutchinson Cancer Research Center
1100 Fairview Ave. N.
PO Box 19024 Seattle, WA 98109

Location: Arnold Building M1 B861
Phone: [phone]

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel

[John Chambers-2]

Good catch.  This should have been fixed in r-devel, revision 60192.  If
nothing bad results in other tests, we'll port it to the current patched
version.

Thanks,
   John

On 8/5/12 4:31 PM, Martin Morgan wrote:

> after this
>
>   setClass("A")
>
>   setGeneric("bar", function(x, y) standardGeneric("bar"))
>   setMethod(bar, signature(x="A", y="A"), function(x, y) {})
>   setMethod(bar, signature(x="A", y="ANY"), function(x, y) {})
>
> the method for signature c("A", "A") contains package information
>
> > str(getMethod(bar, c("A", "A")))
> Formal class 'MethodDefinition' [package "methods"] with 4 slots
>   ..@ .Data  :function (x, y)
>   ..@ target :Formal class 'signature' [package "methods"] with 3 slots
>   .. .. ..@ .Data  : chr [1:2] "A" "A"
>   .. .. ..@ names  : chr [1:2] "x" "y"
>   .. .. ..@ package: chr [1:2] ".GlobalEnv" ".GlobalEnv"
>   ..@ defined:Formal class 'signature' [package "methods"] with 3 slots
>   .. .. ..@ .Data  : chr [1:2] "A" "A"
>   .. .. ..@ names  : chr [1:2] "x" "y"
>   .. .. ..@ package: chr [1:2] ".GlobalEnv" ".GlobalEnv"
>   ..@ generic: atomic [1:1] bar
>   .. ..- attr(*, "package")= chr ".GlobalEnv"
>
> but the method for signature c("A", "ANY") does not
>
> > str(getMethod(bar, c("A", "ANY")))
> Formal class 'MethodDefinition' [package "methods"] with 4 slots
>   ..@ .Data  :function (x, y)
>   ..@ target :Formal class 'signature' [package "methods"] with 3 slots
>   .. .. ..@ .Data  : chr [1:2] "A" "ANY"
>   .. .. ..@ names  : chr [1:2] "x" "y"
>   .. .. ..@ package: chr [1:2] "" ""
>   ..@ defined:Formal class 'signature' [package "methods"] with 3 slots
>   .. .. ..@ .Data  : chr [1:2] "A" "ANY"
>   .. .. ..@ names  : chr [1:2] "x" "y"
>   .. .. ..@ package: chr [1:2] "" ""
>   ..@ generic: atomic [1:1] bar
>   .. ..- attr(*, "package")= chr ".GlobalEnv"
>
> This does not occur when the order of setMethod calls is reversed. The
> reason is that the package information added when
> methods/R/Methods.R:552 calls matchSignature are stripped (because
> matchSignature returns a vector with trailing 'ANY' classes removed)
> by the call to .matchSigLength at line 603.
>
> The context for this is the thread starting at
>
> https://stat.ethz.ch/pipermail/bioconductor/2012-August/047241.html
>
> where a complicated package dependency and attempt to redefine an
> existing method leads to
>
> Error : package slot missing from signature for generic 'sampleNames<-'
> and classes LumiBatch, ANY
> cannot use with duplicate class names (the package may need to be
> re-installed)
>
> Martin

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel