Feature request: ability to forward warnings and messages

[sebffischer]

In mlr3, we are storing messages and warnings that are thrown during model training in logging objects so that users can afterwards investigate them. This is currently not possible with mirai, because it does not forward warnings and messages.

Would it be possible to implement bindings $warnings and $messages of a mirai after it was evaluated. Making it opt-in would also mean there is no overhead in the default scenario.

[shikokuchuo]

You can achieve something like that (in the non-error case) with something like the below:

with_conditions <- function(expr) {
  cnds <- NULL
  list(
    value = withCallingHandlers(expr, condition = function(cnd) cnds <<- c(cnds, list(cnd))),
    conditions = cnds
  )
}

m <- mirai::mirai(
  with_conditions({
    a <- as.integer("a")
    c(a, 2) == 1:3
  }),
  with_conditions = with_conditions
)
m[]
#> $value
#> [1]   NA TRUE   NA
#> 
#> $conditions
#> $conditions[[1]]
#> <simpleWarning in withCallingHandlers(expr, condition = function(cnd) cnds <<- c(cnds,     list(cnd))): NAs introduced by coercion>
#> 
#> $conditions[[2]]
#> <simpleWarning in c(a, 2) == 1:3: longer object length is not a multiple of shorter object length>

^{Created on 2025-08-22 with reprex v2.1.1}

To implement this in mirai would require a deeper integration, and I've not decided on the semantics yet.

[xwanner]

Not sure if it is sufficient in your case, but I use output = T when creating my daemons. It forwards all outputs to the main R process (in my case Shiny)

mirai::daemons(3, output = TRUE, autoexit = FALSE)
mirai::mirai( message("Hello"))