Bug when using internal tuning and encapsulation

[sebffischer]

library(mlr3torch)
#> Loading required package: mlr3
#> Loading required package: mlr3pipelines
#> Loading required package: torch
library(mlr3)
l = lrn("classif.debug", validate = 0.3, early_stopping = TRUE)
l$train(tsk("iris"))
l$internal_valid_scores
#> $acc
#> [1] 0.3111111
l$encapsulate("callr", lrn("classif.featureless"))
l$train(tsk("iris"))
#> INFO  [16:55:35.461] [mlr3] Calling train method of fallback 'classif.featureless' on task 'iris' with 105 observations {learner: <LearnerClassifFeatureless/LearnerClassif/Learner/R6>}
l$internal_valid_scores
#> named list()

^{Created on 2025-02-04 with reprex v2.1.1}

The issue is that the model might be in a marshaled state when the internal tuning values are being extracted