Can fixest auto-Inverse-Variance weight the FE estimate?

[giuliogcantone]

set.seed(1018)
tibble(X = rnorm(1000),
       Y = X + rnorm(1000),
       G = sample(c("A","B"),1000,replace=T)
) %>% fixest::feols(data = .,
            fml = Y ~ X | G,
            se =  "hetero",
            weights = ???) %>%
  tidy %>%
  mutate(across(where(is.numeric),round,3))

Is there a way to instruct fixest to do this:

• compute the ("hetero"?) standard error within groups "A" and "B"
• power it to -2
• rescale
• that is the Inverse Variance Weighting (IVW) for each element of "A", and "B"

or any result of the estimate for X that is coherent with IVW?

[giuliogcantone]

UPDATE:

Someone suggested me this code:

nlme::gls(
model = Y ~ X + G - 1,
weights=varIdent(form = ~ 1 | G),
data = .)

I cannot evaluate gls() because I never worked with GLS before. What I noticed is that nlme::gls() accept a costructor function varIdent() as weights, so my guess is if fixest do the same, and which one should be the function.