noSpecials for propto

[mmcgillycuddy]

Hi Ben,

I'm not sure if this is the problem, but I am getting a problem with the formula in glmmTMB. Let me know if this should be in the glmmTMB issues. I think when I have a propto structure in the formula it doesn't get rid of propto with the noSpecials function.

> noSpecials(sub_specials(y ~ x + homdiag(1 + x | spp)), delete = F)
y ~ x + (1 + x + spp)

> noSpecials(sub_specials(y ~ x + propto(1 + x | spp)), delete = F)
y ~ x + propto(1 + x + spp)

[bbolker]

Yes, the problem is here; there's a hard-coded list of specials. I had already flagged this as a potential issue; have to think about how to pass the information between packages (without having to update in both places), and without some incredibly tedious process of passing information down through a stack of function calls ...

I think we should be able to do this by passing specials = c(names(.valid_covstruct), "s") as an argument ...

[mmcgillycuddy]

Yes, that worked. Thanks Ben.

[bbolker]

Arguably findReTrmClasses() shouldn't use a hard-coded list at all ...? Use an empty list by default and throw a warning/error if user doesn't provide anything?

[mebrooks]

Sorry for the delay. I merged the pull request in glmmTMB (mentioned above) that makes it possible to pass known covstructs to reformulas.