R devel - Jan 2015 - Inspect a "delayed" assigned whose value throws an error?

On Mon, Jan 26, 2015 at 12:24 PM, Hadley Wickham <h.wickham at gmail.com>
wrote:
> If it was any other environment than the global, you could use substitute:
>
> e <- new.env()
> delayedAssign("foo", stop("Hey!"), assign.env = e)
> substitute(foo, e)
>
> delayedAssign("foo", stop("Hey!"))
> substitute(foo)
Hmm... interesting and odd.

Unfortunately, this doesn't seem to help for reaching into the
namespace of hgu133a.db and inspecting 'hgu133aPFAM', e.g.
> library("hgu133a.db")
> substitute(hgu133aPFAM, env=ns)
Error: hgu133aPFAM is defunct. Please use select() if you need access to PFAM
  or PROSITE accessions.
> evalq(substitute(hgu133aPFAM), envir=ns)
Error: hgu133aPFAM is defunct. Please use select() if you need access to PFAM
  or PROSITE accessions.
> evalq(substitute(hgu133aPFAM, env=ns), envir=ns)
Error: hgu133aPFAM is defunct. Please use select() if you need access to PFAM
  or PROSITE accessions.

Thanks,

Henrik

>
> Hadley
>
> On Mon, Jan 26, 2015 at 12:53 PM, Henrik Bengtsson <hb at
biostat.ucsf.edu> wrote:
>> Hi, I got an interesting programming challenge:
>>
>> How do you inspect an object which is assigned via delayedAssign() and
>> that throws an error as soon as it is "touched" (=the value
is
>> evaluated)?  Is it possible?
>>
>>
>> MINIMAL EXAMPLE:
>>
>> $ R --vanilla
>>> delayedAssign("foo", stop("Hey!"))
>>
>> (If you find this minimal example silly/obvious, please skip down to
>> the real example at the end)
>>
>>> foo
>> Error: Hey!
>>
>>> str(foo)
>> Error in str(foo) : Hey!
>> In addition: Warning message:
>> In str(foo) : restarting interrupted promise evaluation
>>
>>> mode(foo)
>> Error in mode(foo) : Hey!
>> In addition: Warning message:
>> In mode(foo) : restarting interrupted promise evaluation
>>
>>> .Internal(inspect(foo))
>> Error: Hey!
>> In addition: Warning message:
>> restarting interrupted promise evaluation
>>
>>> traceback()
>> 1: stop("Hey!")
>>
>> Is there anyway I can inspect this object using the R API without
>> evaluating the value in the delayed assignment?  Is it possible to
>> test if this is a delayed assigned or not?
>>
>>
>> BACKGROUND:
>> The background to this is where I have a function in the R.oo package
>> that scans namespaces for functions with a certain class attribute.
>> For this I use is.function() and inherits() to inspect each object.
>> An aroma.affymetrix user reported on a problem that boiled down to the
>> following:
>>
>> # source("http://bioconductor.org/biocLite.R");
biocLite("hgu133a.db")
>>> library("hgu133a.db")
>>> is.function(hgu133aPFAM)
>> Error: hgu133aPFAM is defunct. Please use select() if you need access
to PFAM
>>   or PROSITE accessions.
>>> .Internal(inspect(hgu133aPFAM))
>>
>>> traceback()
>> 3: stop(paste(msg, collapse = ""), call. = FALSE, domain =
NA)
>> 2: .Defunct(msg = msg)
>> 1: (function ()
>>    {
>>        if (grepl("PFAM", x)) {
>>            bimapName <- paste0(prefix, "PFAM")
>>        }
>>        else {
>>            bimapName <- paste0(prefix, "PROSITE")
>>        }
>>        x <- dc[[bimapName]]
>>        msg = wmsg(paste0(bimapName, " is defunct. ",
"Please use select() if you
>>  need access to PFAM or PROSITE accessions. \n"))
>>        if (interactive()) {
>>            .Defunct(msg = msg)
>>        }
>>    })()
>>
>> My immediate solution is to perform those tests using tryCatch(), but
>> this is interesting, because this function is such that the error is
>> only thrown in interactive() sessions, i.e. the following works:
>>
>> $ Rscript -e "hgu133a.db::hgu133aPFAM"
>> [...]
>> NULL
>>
>> This is probably also why none of my aroma.affymetrix system tests
>> caught this.  Without tracing the source code behind, which seems
>> quite nested, the above is why I believe the assignment is
"delayed";
>> traceback() shows a body source code, the object evaluates to
>> different things depending on interactive().
>>
>> /Henrik
>>
>> ______________________________________________
>> R-devel at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
>
>
>
> --
> http://had.co.nz/

>>>>> Henrik Bengtsson <hb at biostat.ucsf.edu>
>>>>>     on Mon, 26 Jan 2015 12:41:48 -0800 writes:
    > On Mon, Jan 26, 2015 at 12:24 PM, Hadley Wickham <h.wickham at
gmail.com> wrote:
    >> If it was any other environment than the global, you could use
substitute:
    >> 
    >> e <- new.env()
    >> delayedAssign("foo", stop("Hey!"), assign.env =
e)
    >> substitute(foo, e)
    >> 
    >> delayedAssign("foo", stop("Hey!"))
    >> substitute(foo)

    > Hmm... interesting and odd.

    > Unfortunately, this doesn't seem to help for reaching into the
    > namespace of hgu133a.db and inspecting 'hgu133aPFAM', e.g.

    >> library("hgu133a.db")

    >> substitute(hgu133aPFAM, env=ns)
    > Error: hgu133aPFAM is defunct. Please use select() if you need access
to PFAM
    > or PROSITE accessions.

    >> evalq(substitute(hgu133aPFAM), envir=ns)
    > Error: hgu133aPFAM is defunct. Please use select() if you need access
to PFAM
    > or PROSITE accessions.

    >> evalq(substitute(hgu133aPFAM, env=ns), envir=ns)
    > Error: hgu133aPFAM is defunct. Please use select() if you need access
to PFAM
    > or PROSITE accessions.

this *is* interesting..

Note that shortly after delayedAssign() was introduced into R,
we had

  R : Copyright 2005, The R Foundation for Statistical Computing
  Version 2.2.1  (2005-12-20 r36812)
  ISBN 3-900051-07-0

  ............

  > delayedAssign("x", pi^2) ; substitute(x)
  pi^2
  > 

so it also worked with the globalenv;  but that feature already
disappeared with 

  R : Copyright 2006, The R Foundation for Statistical Computing
  Version 2.3.0 (2006-04-24)

Almost surely as an inadvertent side effect of something else.

-----------------------------------

--> Quiz (even though it's not Friday) :

The help page for delayedAssign() contains a somewhat more
sophisticated example of 'promise's, and BTW an example that
worked also in *much* earlier version of R, probably in all
versions of R {I checked  R 1.0.1, probably the oldest still
running on my current Linux desktop} :

e <- (function(x, y = 1, z) environment())(1+2, "y", {cat("
HO! "); pi+2})

which gives an environment full of promises

Now, the following also works alread in R 1.0.1 :
> gete <- function(e)
   lapply(lapply(ls(env=e), as.name), 
         function(n) eval(substitute(substitute(X, e), list(X=n))))
> gete(e)
[[1]]
1 + 2

[[2]]
[1] "y"

[[3]]
{
    cat(" HO! ")
    pi + 2
}
> 
In newer versions of R, you can use  'ls(e)' instead of
'ls(env=e)', and we have bquote() to make it a tad shorter:

gete2 <- function(e)
   lapply(lapply(ls(e), as.name), function(n) eval(bquote(substitute(.(n), e))))

Apart from that (and from not using spaces and from silly things
such as 'L = lapply', and using L),
can anyone find a shorter (or "nicer") way to achieve the same
as gete(), or  gete2(), i.e., return a list with the promise expressions and
not evaluating e ?

You are not allowed to use 'x', 'y', and 'z' explicitly
of
course, because here,
  gete0 <- function(e) list(substitute(x,e), substitute(y,e),
substitute(z,e))
would be slightly shorter.

Martin

On Tue, Jan 27, 2015 at 8:56 AM, Martin Maechler
<maechler at lynne.stat.math.ethz.ch> wrote:
>>>>>> Henrik Bengtsson <hb at biostat.ucsf.edu>
>>>>>>     on Mon, 26 Jan 2015 12:41:48 -0800 writes:
>
>     > On Mon, Jan 26, 2015 at 12:24 PM, Hadley Wickham <h.wickham at
gmail.com> wrote:
>     >> If it was any other environment than the global, you could use
substitute:
>     >>
>     >> e <- new.env()
>     >> delayedAssign("foo", stop("Hey!"),
assign.env = e)
>     >> substitute(foo, e)
>     >>
>     >> delayedAssign("foo", stop("Hey!"))
>     >> substitute(foo)
>
>     > Hmm... interesting and odd.
>
>     > Unfortunately, this doesn't seem to help for reaching into the
>     > namespace of hgu133a.db and inspecting 'hgu133aPFAM', e.g.
>
>     >> library("hgu133a.db")
>
>     >> substitute(hgu133aPFAM, env=ns)
>     > Error: hgu133aPFAM is defunct. Please use select() if you need
access to PFAM
>     > or PROSITE accessions.
>
>     >> evalq(substitute(hgu133aPFAM), envir=ns)
>     > Error: hgu133aPFAM is defunct. Please use select() if you need
access to PFAM
>     > or PROSITE accessions.
>
>     >> evalq(substitute(hgu133aPFAM, env=ns), envir=ns)
>     > Error: hgu133aPFAM is defunct. Please use select() if you need
access to PFAM
>     > or PROSITE accessions.
>
> this *is* interesting..
>
> Note that shortly after delayedAssign() was introduced into R,
> we had
>
>   R : Copyright 2005, The R Foundation for Statistical Computing
>   Version 2.2.1  (2005-12-20 r36812)
>   ISBN 3-900051-07-0
>
>   ............
>
>   > delayedAssign("x", pi^2) ; substitute(x)
>   pi^2
>   >
>
> so it also worked with the globalenv;  but that feature already
> disappeared with
>
>   R : Copyright 2006, The R Foundation for Statistical Computing
>   Version 2.3.0 (2006-04-24)
>
> Almost surely as an inadvertent side effect of something else.
The substitute docs have: "If it is an ordinary variable, its value is
substituted, unless env is .GlobalEnv in which case the symbol is left
unchanged."

So it's definitely a deliberate change. This commit looks related:
https://github.com/wch/r-source/commit/182d197094f8fcecf6ec8f14bfc7c69a02ac251d.
There Duncan commented "For historical reasons, don't substitute in
R_GlobalEnv", which suggests that the behaviour you see in 2.2.1 might
have been a bug.

Hadley


-- 
http://had.co.nz/