Objects of class "expression" don't seem to be searched recursively

[DavisVaughan]

It seems like x should be identified as a potential global here:

expressions <- parse(text = "x")
expressions
#> expression(x)

# Why isn't `x` found?
globals::findGlobals(expressions)
#> character(0)

# We can find it like this:
expressions[[1]]
#> x
globals::findGlobals(expressions[[1]])
#> [1] "x"

# Or like this
list_of_expressions <- list(quote(x))
list_of_expressions
#> [[1]]
#> x

globals::findGlobals(list_of_expressions)
#> [1] "x"

Originally reported in futureverse/furrr#209

[DavisVaughan]

It is worth noting that future_map() fails, but future_lapply() works. That only works because of the fact that future_lapply() converts the input to a list with as.list() before the globals are searched - which gives us something like list_of_expressions above that findGlobals() can search correctly.

I still think the problem is that findGlobals() can't search "expression" types recursively

library(future.apply)
#> Loading required package: future
library(furrr)
plan(multisession, workers = 2)

x <- 0

y <- parse(text = "x")
y
#> expression(x)

future_map(y, eval)
#> Error in ...furrr_fn(...): object 'x' not found

future_lapply(y, eval)
#> [[1]]
#> [1] 0

# Because future_lapply() does this ahead of time:
# if (!is.vector(X) || is.object(X)) 
#   X <- as.list(X)

# Which gives us the list form that findGlobals() correctly searches:
as.list(y)
#> [[1]]
#> x

^{Created on 2021-10-25 by the reprex package (v2.0.1)}

[abdellah19jan]

Hello,

Thank you very much for your reply!

Unfortunately the solution you suggested does not work in my case.

I've tried to formulate a simplified example of the code I am running. It looks like this:

library(furrr)
library(tidyverse)

plan(multisession, workers = 2)

fun <- function(arg) arg + 1
dic <- c("mod" = "fun")

f <- function(x) invoke_map(dic[[x]], 2)

map("mod", f)
future_map("mod", f)

Here are the solutions you suggested (see code below):

1 - Pass the function object as an argument rather than its name as a string: Same result, it works with map but not with future_map.
2 - Use future_lapply instead of future_map: it gives the same error as future_map.

library(future.apply)

f <- function(x) invoke_map(eval(parse(text = dic[[x]])), 2)

map("mod", f)
future_map("mod", f)
future_lapply("mod", f)

The original code I am running is much more complicated than the example above, but I believe if there is a solution that makes the example work, then it will work also for my code.

[abdellah19jan]

Hello,

Is there a solution in the meantime to work around this problem ?

Thanks a lot for your help !