Functions with envir = parent.frame()

[krlmlr]

Compare the following two statements:

"file.rda" %>% load
load("file.rda")

The former will load the data into a temporary environment; the reason is that load is declared as

load <- function (file, envir = parent.frame(), verbose = FALSE)

The same happens with assign and get.

This is at least unexpected, I'm not sure if we should call it a bug. Worse, I'm not sure if this issue can be solved at all in a robust way.

Similar to #32.

[byzheng]

I have the same problem. For my digging, function "pipe" do evaluate "load" and load all variables into the new environment, but this code below returns result with visible = FALSE (L43 in pipe.R).

result <- withVisible(eval(quote(`_fseq`(`_lhs`)), env, env))

So pipe returns results with invisible (L53 in pipe.R)

invisible(result[["value"]])

Not sure how to fix it.

[smbache]

This is one of those things that keep popping up; on the one hand I am not a fan of side-effect functions like e.g. load, and typically I wrap it in a function which returns whatever is loaded; in general depending "too much" on the environments around is not really good practice. That being said, I know it would be nice to solve this, but to be honest, I am not sure there is an easy fix for it. I'll keep it in mind though...

[byzheng]

Thanks for your reply. Andri and Richard Scriven at Stackoverflow provided a solution and explanation:
http://stackoverflow.com/questions/27053935

[hadley]

Generally, I think it's better to use saveRDS()/writeRDS rather than save()/load() because of how they mess around with environments.

[krlmlr]

This issue affects all functions with the construct envir = parent.frame(). ???envir lists more than 800 entries.

Compare (from an interactive prompt):

> parent.frame()
<environment: R_GlobalEnv>
> 1 %>% parent.frame
<environment: 0x4157240>

How about something like the following:

pipe.frame <- function() parent.frame(7)
"b" %>% assign(3, envir = pipe.frame())
b
## [1] 3

The 7 is an implementation detail, consistency could be assured with a unit test. (The example is just a placeholder for any usage of parent.frame in a function call.

I have looked, but haven't found a way to make parent.frame() return the "correct" value from within a pipe.

[hadley]

I think the best we can do is to provide a convenient way to get the right environment, i.e. #171.

I'm now almost certain that that's no way to implement the pipe that would preserve parent.frame(), without sacrificing properties that we care about more.

[iagogv3]

I've just found another problem with the save function, which is envir = parent.frame() too, which seems to be related to this issue.

If I save a data frame with pipe:

library(magrittr)
mtcars %>% save(file = "test.rdata")

And later, when I don't have it in the Global Env, I try to load with

load("test.rdata")

it does not load anything. If I save it without pipe, all works as usual.