Hi, this is a minor problem in solve.default but I didn't see anyone

mention it. The function does not give "correct" error-handling if given

non-square matrix, LINPACK=F, and matrix has names attributes (either

row or column names).

Demo:

##"correct" error handling of non-square matrix

> temp<-diag(1,5)[,1:4]

> solve(temp,LINPACK=F)

Error in solve.default(diag(1, 5)[, 1:4], LINPACK = F) :

'A' (5 x 4) must be square

#indeciferable error handling when same matrix is given names attributes

> temp<-diag(1,5)[,1:4]

> rownames(temp)<-as.character(1:5)

> colnames(temp)<-as.character(1:4)

> solve(temp,LINPACK=F)

Error in "rownames<-"(`*tmp*`, value = c("1", "2", "3", "4")) :

length of 'dimnames' [1] not equal to array extent

I think the problem is that if LINPACK=F, then the error handling of the

argument "a" is not done until after creating a matrix "b", when none

was given. Then, the function creates a square identity matrix for "b",

to which it tries to assign row and column names of "a". Of course,

since "a" is not square, they are not of the correct length.

Best,

Elizabeth

______________________________________________

[hidden email] mailing list

https://stat.ethz.ch/mailman/listinfo/r-devel