Inferno with binomial glm with "Wilkinson-Rogers" format and visreg package

Let do a binomial glm with cbind(success, failure)

First, I prepare the data:

factor <- rnorm(7, 10, 2)

dta <- data.frame(p=c(2, 7, 8, 9, 10, 18, 7),
                  n=c(20, 20, 20, 20, 20, 20, 20))

dta cannot be used directly; it must be a matrix, not a data.frame:

g <- glm(dta ~ factor, 

         family = binomial(link = "logit"))

# Error in model.frame.default(formula = dta ~ factor, drop.unused.levels = TRUE) : 

#     type (list) incorrect pour la variable 'dta'

g <- glm(as.matrix(dta) ~ factor, 

         family = binomial(link = "logit"))

coef(g)

 (Intercept)       factor 

-0.767901152 -0.006062598 

Great, it works.

But if you try to plot the effects with visreg package, it produced an error:

library(visreg)

visreg(g, xvar ="factor")

# Error in dimnames(x) <- dn : 

#  la longueur de 'dimnames' [1] n'est pas égale à l'étendue du tableau

The error is because the first term of the formula is not understood correctly; it must be a simple term.
For example, here it does not work either:
g <- glm(cbind(p=dta$p, n=dta$n) ~ factor, 
         family = binomial(link = "logit"))

But here, it works:

n <- dta$n

p <- dta$p

g <- glm(cbind(p=p, n=n) ~ factor, 

         family = binomial(link = "logit"))

library(visreg)

visreg(g, xvar ="factor", scale = "response", ylim=c(0, 1))

Thanks to Breheny, Patrick J <patrick-breheny@uiowa.edu> author of the visreg package for helping me to solve this problem.