R devel - Nov 2023 - saveRDS()/readRDS() on environments

Dear r-devel,

I was surprised to see that saveRDS() and readRDS() work quite well with
environments, see below:

```
z <- 3 # in global env
y <- new.env()
y$a <- 1
x <- new.env(parent = y)
x$b <- 2
saveRDS(x, "x.RDS")

# in a new session
x <- readRDS("x.RDS")
y <- parent.env(x)
x$b
#> [1] 2
y$a
#> [1] 1
parent.env(y)
#> <environment: R_GlobalEnv>
z
#> Error: object 'z' not found
```
We know that environments by definition cannot be cloned without
compromises, and we see above that compromises were made: we did save the
first parent, but didn't save or at least restore the global env. I
don't
see this documented in `?ReadRDS`, did I miss something ?

I would like to know what rules are followed, if I had to guess I would
think that parents are saved until a "special environment" is reached
(such
as global env, empty env, package envs, namespaces, imports, datasets), but
I'd like clarifications on this point.

Thanks,

Antoine

	[[alternative HTML version deleted]]

On Wed, 15 Nov 2023 19:49:00 +0100
Antoine Fabri <antoine.fabri at gmail.com> wrote:
> I would like to know what rules are followed, if I had to guess I
> would think that parents are saved until a "special environment"
is
> reached (such as global env, empty env, package envs, namespaces,
> imports, datasets), but I'd like clarifications on this point.
This is what R Internals has to say:
>> Serialization in R needs to take into account that objects may
>> contain references to environments, which then have enclosing
>> environments and so on. (Environments recognized as package or name
>> space environments are saved by name.
>> Some objects are written as if they were SEXPTYPEs: such
>> pseudo-SEXPTYPEs cover R_NilValue, R_EmptyEnv, R_BaseEnv,
>> R_GlobalEnv, R_UnboundValue, R_MissingArg and R_BaseNamespace. 
>> Environments are treated in several ways: as we have seen, some are
>> written as specific pseudo-SEXPTYPEs. Package and namespace
>> environments are written with pseudo-SEXPTYPEs followed by the name.
>> ?Normal? environments are written out as ENVSXPs with an integer
>> indicating if the environment is locked followed by the enclosure,
>> frame, ?tag? (the hash table) and attributes. 
<http://mirrors.dotsrc.org/cran/doc/manuals/R-ints.html#Serialization-Formats>

(I would normally link to cran.r-project.org, but they are undergoing
maintenance until tomorrow.)

To summarise, ordinary, anonymous environments (including those
originating from a call frame that a closure may inherit!) are saved.
Package-related environments are referenced by name, and so is the
global environment.

-- 
Best regards,
Ivan