Thank you hugely. Your suggestion nailed it. I was *sure* that I
had tried doing
environment(bar) <- new.env()
prior to doing the assignment and that it didn't help. I was also sure
that in AssetPricing I done assignments into environments that *weren't*
new. Wrong on both counts. :-( Jeez, it's tough being mentally
handicapped!
Thanks very much for delving into this and for looking at the code in
AssetPricing more carefully and perspicaciously than I did.
cheers,
Rolf
On 09/07/18 11:42, Martin Morgan wrote:
On 07/08/2018 07:23 PM, Duncan Murdoch wrote:
On 08/07/2018 6:57 PM, Rolf Turner wrote:
Recently I experimented with assigning a variable within the environment
of a function in a package that I am developing. Slightly more
explicitly:
In a function "foo()" in the package, I have lines like:
big <- 42
assign("big",big,envir=environment(bar))
where "bar()" is another function in the package.
This causes an error to be thrown:
Error in assign("big", big, envir = environment(bar)) :
cannot add bindings to a locked environment
The reason that I am puzzled about this is that I used this syntax
repeatedly in my CRAN package "AssetPricing". In this latter package
the syntax seems to work flawlessly. No errors are thrown. No mention
of "locked environments".
Why would environments get "locked" in the package that I am currently
fooling about with, but not in the AssetPricing package? Is there
anything that I can do to keep environments from getting "locked"?
It may be the timing. When R is installing a package, it executes all
of the code in the .R files (which has been consolidated into one
file, but that doesn't matter). This produces all of the objects and
functions in the package.
Then it locks the environment, so that things don't change.
Then it compiles the functions. If they were allowed to change, then
the compilation would have to be redone.
So perhaps AssetPricing makes the change in the .R file, processed
before locking, and your new package does it in a function, executed
by the user after locking.
AssetPricing does, at least some of the time, in the body of a function
(the indentation is confusing)
environment(vupdate) <- new.env()
environment(scrG) <- new.env()
environment(initx) <- new.env()
environment(cev) <- new.env()
#
assign("dS",dS,envir=environment(vupdate))
assign("dS",dS,envir=environment(scrG))
assign("dS",dS,envir=environment(initx))
assign("dS",dS,envir=environment(cev))
so that assignment is to the unlocked new.env(). Presumably the new code
assigns to bar's original environment, which is the (locked, once the
package is loaded) package name space.
Martin Morgan
Or maybe not.
Duncan Murdoch
Can anyone suggest where I should start looking for differences in the
nature of the two packages?
I could possibly give more detail, but I have no idea as to what might
be relevant.
One difference is that AssetPricing involves no dynamically loaded
Fortran code, whereas the package that I am currently fooling about with
*does* involve such code. (But neither "foo()" not "bar()" make direct
calls to .Fortran().)
Grateful for any insights.
______________________________________________
R-package-devel@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-package-devel
