[Bioc-devel] Import BSgenome class without attaching BiocGenerics (and others)?
There's never a need to attach a package to satisfy the dependencies of another package. That would defeat the purpose of namespaces. The three coercion functions gets at the heart of the S3/S4 mess. setAs() provides a dynamic coercion mechanism, so is useful for as(x, class) when class can be anything. as.data.frame() is an S3 generic defined by the base package, so every package sees it. Something promotes as.data.frame() to an S4 generic, but only packages that import the generic can see it. That likely excludes the vast majority of CRAN packages. Thus, we define an S3 method for calls to the S3 generic. The S4 generic will fall back to the S3 methods; however, it will first check all S4 methods. Defining as.data.frame,Vector() defends against the definition of a method above it, such as as.data.frame,Annotated(), which would intercept dispatch to the S3 as.data.frame.Vector(). Michael On Fri, Sep 6, 2019 at 10:08 AM Bhagwat, Aditya
<Aditya.Bhagwat at mpi-bn.mpg.de> wrote:
Thanks Kasper and Michael.
The importClassesFrom sounds like something that would allow an attachment-free S4 class import, will check them out.
Michael, the current auto-attach is causing 66 namespace clashes ? not feeling very comfortable about that, so trying to avoid them.
I also think there?s something about S4 coercion that I don?t yet fully understand.
For instance: the S4Vectors package has three different versions of a S4Vectors::Vector -> data.frame coercer. Why? Any useful pointers?
setAs("Vector", "data.frame", function(from) as.data.frame(from))
as.data.frame.Vector <- function(x, row.names=NULL, optional=FALSE, ...) {
as.data.frame(x, row.names=NULL, optional=optional, ...)
}
setMethod("as.data.frame", "Vector",
function(x, row.names=NULL, optional=FALSE, ...)
{
x <- as.vector(x)
as.data.frame(x, row.names=row.names, optional=optional, ...)
})
From: Kasper Daniel Hansen <kasperdanielhansen at gmail.com>
Sent: Freitag, 6. September 2019 17:49
To: Michael Lawrence <lawrence.michael at gene.com>
Cc: Bhagwat, Aditya <Aditya.Bhagwat at mpi-bn.mpg.de>; bioc-devel at r-project.org
Subject: Re: [Bioc-devel] Import BSgenome class without attaching BiocGenerics (and others)?
There are
importMethodsFrom(PACKAGE, NAME_OF_METHODS)
importClassesFrom(PACKAGE, NAME_OF_METHODS)
to help with selective importing S4 methods/classes. See section 1.5.6 of WRE.
On Fri, Sep 6, 2019 at 9:24 AM Michael Lawrence via Bioc-devel <bioc-devel at r-project.org> wrote:
It sounds like you are trying to defer loading of namespaces in order
to save time when they are unnecessary? That's probably going to end
up a losing battle.
On Fri, Sep 6, 2019 at 5:47 AM Bhagwat, Aditya
<Aditya.Bhagwat at mpi-bn.mpg.de> wrote:
Thank you Michael,
Appreciate your time for helping me fill the gaps in my understanding of the S4 flow :-).
It all started when I defined (in my multicrispr package) the S4 coercer :
methods::setAs( "BSgenome",
"GRanges",
function(from) as(GenomeInfoDb::seqinfo(from), "GRanges")
When building, I noticed the message
in method for 'coerce' with signature '"BSgenome","GRanges"': no definition for class "BSgenome"
So, I added
BSgenome <- methods::getClassDef('BSgenome', package = 'BSgenome')
That loads all these dependencies.
From your answer, I understand that there is currently no alternative to loading all these dependencies.
I guess because these dependencies are needed to provide for all required S4 methods for the BSgenome class, am I right?
Is there a way to define a methods::setAs without loading the class definition?
Aditya
________________________________________ From: Michael Lawrence [lawrence.michael at gene.com] Sent: Friday, September 06, 2019 1:09 PM To: Bhagwat, Aditya Cc: bioc-devel at r-project.org Subject: Re: [Bioc-devel] Import BSgenome class without attaching BiocGenerics (and others)? The way to keep a "clean namespace" is to selectively import symbols into your namespace, not to import _nothing_ into your namespace. Otherwise, your code will fill with namespace qualifications that distract from what is more important to communicate: the intent of the code. And no, there's no way to define method signatures using anything other than simple class names. It would be interesting to explore alternative ways of specifying method signatures. One way would be if every package exported a "class reference" (class name with package attribute, at least) for each of its classes. Those could be treated like any other exported object, and referenced via namespace qualification. It would require major changes to the methods package but that should probably happen anyway to support disambiguation when two packages define a class of the same name. It would be nice to get away from the exportClasses() and importClasses() stuff. File that under the "rainy year" category. Michael On Fri, Sep 6, 2019 at 3:39 AM Bhagwat, Aditya <Aditya.Bhagwat at mpi-bn.mpg.de> wrote: Dear Bioc devel, Is it possible to import the BSgenome class without attaching BiocGenerics (to keep a clean namespace during the development of multicrispr<https://gitlab.gwdg.de/loosolab/software/multicrispr>). BSgenome <- methods::getClassDef('BSgenome', package = 'BSgenome') (Posted earlier on BioC support<https://support.bioconductor.org/p/124442/> and redirected here following Martin's suggestion) Thankyou :-) Aditya [[alternative HTML version deleted]] _______________________________________________ Bioc-devel at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/bioc-devel -- Michael Lawrence Scientist, Bioinformatics and Computational Biology Genentech, A Member of the Roche Group Office +1 (650) 225-7760 michafla at gene.com Join Genentech on LinkedIn | Twitter | Facebook | Instagram | YouTube
-- Michael Lawrence Scientist, Bioinformatics and Computational Biology Genentech, A Member of the Roche Group Office +1 (650) 225-7760 michafla at gene.com Join Genentech on LinkedIn | Twitter | Facebook | Instagram | YouTube
_______________________________________________ Bioc-devel at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/bioc-devel -- Best, Kasper
Michael Lawrence Scientist, Bioinformatics and Computational Biology Genentech, A Member of the Roche Group Office +1 (650) 225-7760 michafla at gene.com Join Genentech on LinkedIn | Twitter | Facebook | Instagram | YouTube