[graalvm-users] Setting attributes inconsistencies
Lukas Stadler
lukas.stadler at oracle.com
Mon Jul 9 09:01:02 PDT 2018
R vectors are passed around as immutable values, so that in theory every operation will create a new copy of the vector.
E.g., the assignment function `names<-` returns a vector with a modified "names" attribute, so that "names(vect) <- nams)" is equivalent to "vect <- `names<-`(vect, nams)".
Since creating new vectors all the time would be prohibitively expensive, R engines internally keep some form of "reference count" to see whether the vector is accessible from outside the current scope.
In the first call to `names<-` in your example, the vector is only referenced by the "vect" symbol, so that it can be modified in place.
If you print the vector, the (internally quite complex) printing code will force the vector into a "shared" state, which requires a copy-on-write the next time it is modified, which in your case is the call to `dim<-`.
This call will then create a new copy of the vector with the dimensions set, but leave the original one (in "vect") unchanged.
Long story short, the update functions (`xyz<-`) shouldn't be called directly in R, and (as Stepan said) FastR should mark all objects as "shared" when they escape the R context.
- Lukas
> On 09.07.2018, at 15:45, Rodrigo Botafogo <rodrigo.a.botafogo at gmail.com> wrote:
>
> Hello,
>
> Thanks for all the answers. It just did a bunch of test and it seems to me now to either be a bug in Gnu R or just a very strange behavior that I don't understand either about R. I will report it to the Gnu R user groups and see what they tell me. The important point here is that Graalvm is doing exactly the same as Gnu R.
>
> For the curious, here what happens in GNU R. Bellow, two examples. The only difference between the two is that I commented a 'print' on the second code:
>
> Example 1:
>
> > vect = c(1, 2, 3, 4, 5, 6);
> > nams = c("a", "b", "c", "d", "e", "f");
> > `names<-`(vect, nams);
> a b c d e f
> 1 2 3 4 5 6
> > print(vect);
> a b c d e f
> 1 2 3 4 5 6
> > dims = c(3, 2);
> > `dim<-`(vect, dims);
> [,1] [,2]
> [1,] 1 4
> [2,] 2 5
> [3,] 3 6
> > print(vect);
> a b c d e f
> 1 2 3 4 5 6
>
> As we can see, the vect at the end does not have the dimension property.
>
> Example 2
>
> > vect = c(1, 2, 3, 4, 5, 6);
> > nams = c("a", "b", "c", "d", "e", "f");
> > `names<-`(vect, nams);
> a b c d e f
> 1 2 3 4 5 6
> > # print(vect);
> > dims = c(3, 2);
> > `dim<-`(vect, dims);
> [,1] [,2]
> [1,] 1 4
> [2,] 2 5
> [3,] 3 6
> > print(vect);
> [,1] [,2]
> [1,] 1 4
> [2,] 2 5
> [3,] 3 6
>
> In this example, the first 'print(vect)' is commented out. When printing vect on the last line, the dimension is set.
>
>
> On Fri, Jul 6, 2018 at 7:03 PM Benoit Daloze <eregontp at gmail.com <mailto:eregontp at gmail.com>> wrote:
> It seems to me it's as Stepan said.
> R of course does not reassign Ruby variables.
> And most R operations might or not mutate the receiver in place, with
> the general model of mutating only if it's unobservable
> (it is observable here, `vec` is actually a second reference to the
> initial vector).
>
> On Fri, Jul 6, 2018 at 11:31 PM, Chris Seaton <chris.seaton at oracle.com <mailto:chris.seaton at oracle.com>> wrote:
> > Maybe this is something to do with the `dim<-` method? I don’t know anything about R but I guess that’s called when you do `names(something) =`. Can you try making a simple test case that just tests that these kind of methods can be called correctly? Maybe something about it messes up the arguments?
> >
> > Chris
> >
> >> On 6 Jul 2018, at 17:41, Rodrigo Botafogo <rodrigo.a.botafogo at gmail.com <mailto:rodrigo.a.botafogo at gmail.com>> wrote:
> >>
> >> Hello,
> >>
> >> I'm trying to set attributes to an r object using interop. For some attributes everything works fine and for others the attribute is not set.
> >>
> >> Here what I'm trying in R:
> >>
> >> R.eval(<<-R)
> >> vect = c(1, 2, 3, 4, 5, 6);
> >> names(vect) = c("a", "b", "c", "d", "e", "f");
> >> print(vect);
> >> dim(vect) = c(3, 2);
> >> print(vect)
> >> R
> >>
> >> The first print show the names set:
> >>
> >> a b c d e f
> >> 1 2 3 4 5 6
> >>
> >> and the second print shows the dimension change. Changing the "dim"
> >> attribute erases the names attribute:
> >> [,1] [,2]
> >> [1,] 1 4
> >> [2,] 2 5
> >> [3,] 3 6
> >>
> >> Now, doing this using interop:
> >>
> >> # pp to allow printing: already reported bug
> >> pp = Polyglot.eval("R", "function(x) print(x)")
> >>
> >> vec = Polyglot.eval("R", "c").call(1, 2, 3, 4, 5, 6)
> >> nams = Polyglot.eval("R", "c").call("a", "b", "c", "d", "e", "f")
> >> Polyglot.eval("R", "`names<-`").call(vec, nams)
> >> pp.call(vec)
> >>
> >> this prints as expected the vector with the names attribute:
> >>
> >> a b c d e f
> >> 1 2 3 4 5 6
> >>
> >> dims = Polyglot.eval("R", "c").call(3, 2)
> >> Polyglot.eval("R", "`dim<-`").call(vec, dims)
> >> pp.call(vec)
> >>
> >> a b c d e f
> >> 1 2 3 4 5 6
> >>
> >> method dim<- does not seem to have worked. There is no error message or anything after calling `dim<-`
> >>
> >> Is this a bug or did I do something wrong here? Its strange that some attributes work and others don't.
> >>
> >>
> >>
> >>
> >> --
> >> Rodrigo Botafogo
> >>
> >> _______________________________________________
> >> GraalVM-Users mailing list
> >> GraalVM-Users at oss.oracle.com <mailto:GraalVM-Users at oss.oracle.com>
> >> https://oss.oracle.com/mailman/listinfo/graalvm-users <https://oss.oracle.com/mailman/listinfo/graalvm-users>
> >
> >
> > _______________________________________________
> > GraalVM-Users mailing list
> > GraalVM-Users at oss.oracle.com <mailto:GraalVM-Users at oss.oracle.com>
> > https://oss.oracle.com/mailman/listinfo/graalvm-users <https://oss.oracle.com/mailman/listinfo/graalvm-users>
>
>
> --
> Rodrigo Botafogo
> Integrando TI ao seu negócio
> 21-3010-4802/11-3010-1802
> _______________________________________________
> GraalVM-Users mailing list
> GraalVM-Users at oss.oracle.com <mailto:GraalVM-Users at oss.oracle.com>
> https://oss.oracle.com/mailman/listinfo/graalvm-users <https://oss.oracle.com/mailman/listinfo/graalvm-users>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://oss.oracle.com/pipermail/graalvm-users/attachments/20180709/d1c8e90e/attachment-0001.html
More information about the GraalVM-Users
mailing list