findGlobals(~NULL) errors

[DavisVaughan]

Simplest possible reprex:

library(globals)

globals::findGlobals(~NULL)
#> Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") {: missing value where TRUE/FALSE needed

More complex reprex with some details. This came up when running revdeps for furrr. The baguette package failed because it has quosure objects (which are formulas) that get searched by getGlobalsAndPackages() and one of the quosures had NULL in it.

library(globals)
library(rlang)

x <- quo(NULL)

globals::findGlobals(x)
#> Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") {: missing value where TRUE/FALSE needed

# Oh...
x[-1]
#> Warning: Subsetting quosures with `[` is deprecated as of rlang 0.4.0
#> Please use `quo_get_expr()` instead.
#> This warning is displayed once per session.
#> Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") {: missing value where TRUE/FALSE needed

# The problem:
# rlang:::`[.quosure` hands of to NextMethod
# which gets us into stats:::`[.formula`
# which fails in the second half of the `if()` statement there
stats:::`[.formula`
#> function (x, i) 
#> {
#>     ans <- NextMethod("[")
#>     if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") {
#>         class(ans) <- "formula"
#>         environment(ans) <- environment(x)
#>     }
#>     ans
#> }
#> <bytecode: 0x7f9a455f8a28>
#> <environment: namespace:stats>

Using [ and [[ on quosures are both soft deprecated in rlang, so perhaps the formula could be unclassed at some point? Not sure. Unfortunately I think this will block a furrr release as well since there isn't anything that I'd change in baguette to fix it. :/

[DavisVaughan]

CC @lionel-. Not sure if you knew that quo(NULL)[-1] would error like this (even though that behavior is soft deprecated)

[lionel-]

@lionel-. Not sure if you knew that quo(NULL)[-1] would error like this (even though that behavior is soft deprecated)

It looks like it's coming from the formula method to which [.quosure delegates.

[HenrikBengtsson]

For the purpose of fixing this in 'globals', this bug affects also:

globals::findGlobals(~NULL)
globals::findGlobals(NULL~rhs)
globals::findGlobals(NULL~NULL)

but not

globals::findGlobals(lhs~NULL)

[HenrikBengtsson]

This happens because internally we end up doing:

> e <- ~ NULL
> str(e)
Class 'formula'  language ~NULL
  ..- attr(*, ".Environment")=<environment: R_GlobalEnv> 
> length(e)
[1] 2
> e[-1]
Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") { : 
  missing value where TRUE/FALSE needed
> traceback()
2: `[.formula`(e, -1)
1: e[-1]

where

> stats:::`[.formula`
function (x, i) 
{
    ans <- NextMethod("[")
    if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") {
        class(ans) <- "formula"
        environment(ans) <- environment(x)
    }
    ans
}
<bytecode: 0x55a10bdca238>
<environment: namespace:stats>

where ans becomes:

Browse[2]> str(ans)
 language NULL()
Browse[2]> length(ans)
[1] 1
Browse[2]> ans[[1L]]
NULL
Browse[2]> as.character(ans[[1L]])
character(0)
Browse[2]> as.character(ans[[1L]])[1L]
[1] NA
Browse[2]> as.character(ans[[1L]])[1L] == "~"
[1] NA
Browse[2]> length(ans) == 0L || as.character(ans[[1L]])[1L] == "~"
[1] NA

Hmm... I wonder if this is a bug in stats:::[.formula. This bug is not only when we try to drop the first element but also when we attempt to access any NULL element in a formula using [();

> e <- ~ NULL
> length(e)
[1] 2
> str(as.list(e))
List of 2
 $ : symbol ~
 $ : NULL
 - attr(*, "class")= chr "formula"
 - attr(*, ".Environment")=<environment: R_GlobalEnv>
> e[2]
Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") { : 
  missing value where TRUE/FALSE needed

> e <- NULL ~ NULL
> length(e)
[1] 3
List of 3
 $ : symbol ~
 $ : NULL
 $ : NULL
 - attr(*, "class")= chr "formula"
 - attr(*, ".Environment")=<environment: R_GlobalEnv>
> e[2]
Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") { : 
  missing value where TRUE/FALSE needed
> e[3]
Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") { : 
  missing value where TRUE/FALSE needed

> e <- NULL ~ rhs
> length(e)
[1] 3
> str(as.list(e))
List of 3
 $ : symbol ~
 $ : NULL
 $ : symbol rhs
 - attr(*, "class")= chr "formula"
 - attr(*, ".Environment")=<environment: R_GlobalEnv>
> e[1]
`~`()
> e[2]
Error in if (length(ans) == 0L || as.character(ans[[1L]])[1L] == "~") { : 
  missing value where TRUE/FALSE needed
> e[3]
rhs()

I might be able to work around it in findGlobals() - need to think about it.

[HenrikBengtsson]

I rewrote the code so it avoids using the problematic e[-1] on formula objects to instead uses e[[kk]] on formula objects;

         stop_if_not(identical(e[[1]], as.symbol("~")))
-        expr <- e[-1]
-        for (kk in seq_along(expr)) {
-          globals <- find_globals_ordered(expr = expr[[kk]], envir = w$env)
+        stop_if_not(length(e) >= 2L, identical(e[[1]], as.symbol("~")))
+        for (kk in 2:length(e)) {
+          globals <- find_globals_ordered(expr = e[[kk]], envir = w$env)

[HenrikBengtsson]

I've posted 'Bug in stats:::[.formula: (~ NULL)[2] gives Error ... missing value where TRUE/FALSE needed' to R-devel on 2020-08-14.