Add rbind and cbind methods for mmMatrix

[agrueneberg]

rbind will be easy for rmmMatrix, cbind for cmmMatrix. The other ones will be tricky and probably involve creating new chunks.

[agrueneberg]

Another option, instead of manually adding chunks to an existing mmMatrix, would be to create a whole new one that combines all input parameters of rbind or cbind. This would allow for both functions to be easily implemented and for multiple mmMatrix arguments, and also result in leaner chunking (our main overhead).

[agrueneberg]

We also have to decide whether to get the current storage location of an mmMatrix via an attribute of the mmMatrix (which impacts portability as we cannot always guarantee accuracy of that value since an mmMatrix can be loaded directly using load) or as a separate parameter (which would be unusual for an *bind method).

[agrueneberg]

Looks like additional parameters can be added to the *bind methods. vmode, folderOut, nChunks, and dimorder would be nice.

[agrueneberg]

Here's an initial draft for rbind for rmmMatrix:

rbind.rmmMatrix<-function(...,vmode='byte',folderOut=NULL,nChunks=NULL,dimorder=c(2,1),deparse.level=1){
    allargs<-list(...)
    alldims<-unique(sapply(allargs,function(arg){
        dims<-dim(arg)
        if(is.null(dims)){
            c(1, length(arg))
        }else{
            dims
        }
    }))
    numcols<-unique(alldims[2,])
    if(length(numcols)>1){
        stop('all params need the same number of columns')
    }
    numrows<-sum(alldims[1,])
    mmmatrix<-rmmMatrix(nrow=numrows,ncol=numcols,vmode=vmode,folderOut=folderOut,nChunks=nChunks,dimorder=dimorder)
    currow<-1
    for(arg in 1:ncol(alldims)){
        argrows<-alldims[1,arg]
        for(argrow in 1:argrows){
            if(argrows>1){
                mmmatrix[currow,]<-allargs[[arg]][argrow,]
            }else{
                mmmatrix[currow,]<-allargs[[arg]]
            }
            currow<-currow+1
        }
    }
    mmmatrix
}

Let me know what you think. In the meantime I will time it out for larger matrices. There is are some possible performance improvements when copying more than one row at the same time.

[agrueneberg]

I'll reopen this in the LinkedMatrix repository.