CONCERN: Is it safe to have attributes on environments?

Note to self: Is it safe to have attributes on environments? It might not be, cf. https://github.com/wch/R6/issues/64 and https://bugs.r-project.org/bugzilla3/show_bug.cgi?id=16441.

UPDATE: I'm pretty sure I went through this discussion back in ~2001/2 for the R.oo package. The Object:s in R.oo basically uses:

``` r
x <- NA
attr(x, "envir") <- new.env()
class(x) <- c("Object")
```

such that attributes are _not_ attached to the environment. However, I'm not sure if this is related.