R devel - Mar 2008 - Losing attributes in data.frame() (PR#10873)

Folks:

Problem: [<-.data,frame()  is losing attributes under certain curcumstances
shown below.

I think this is a bug, at least in documentation, as I was unable to find
explicit documentation of the behavior. Indeed, the only documentation I
found told me attributes are preserved. Here is a detailed description:
> df <- data.frame(a=1:3, b=letters[1:3])
> attr(df,"foo") <- 10
> df
  a b
1 1 a
2 2 b
3 3 c
> attributes(df)
$names
[1] "a" "b"

$row.names
[1] 1 2 3

$class
[1] "data.frame"

$foo
[1] 10

## But now we add a new column using matrix-type indexing
> df[,"c"] <- 10:12
> df
  a b  c
1 1 a 10
2 2 b 11
3 3 c 12
> attributes(df)
$names
[1] "a" "b" "c"

$row.names
[1] 1 2 3

$class
[1] "data.frame"
> ## Note that "foo" attribute is lost; however...
> df <- data.frame(a=1:3, b=letters[1:3])
> attr(df,"foo") <- 10
> df[["c"]] <- 10:12
> attributes(df)
$names
[1] "a" "b" "c"

$row.names
[1] 1 2 3

$foo
[1] 10

$class
[1] "data.frame"

## The foo attribute has been kept this time.

## Details:
> sessionInfo()
R version 2.6.2 (2008-02-08) 
i386-pc-mingw32 

locale:
LC_COLLATE=English_United States.1252;LC_CTYPE=English_United
States.1252;LC_MONETARY=English_United
States.1252;LC_NUMERIC=C;LC_TIME=English_United States.1252

attached base packages:
[1] graphics  grDevices datasets  grid      tcltk     stats     utils
methods   base     

other attached packages:
[1] svSocket_0.9-5    svIO_0.9-5        svIDE_0.9-5       R2HTML_1.58
svMisc_0.9-5      lme4_0.99875-9   
[7] Matrix_0.999375-4 lattice_0.17-6   
> 

Cheers,

Bert Gunter
Genentech Nonclinical Statistics

This is a delicate balance.  The crucial part (which you didn't mention) 
is that this only happens if you add a column by df[, "c"] <- value
; if
you replace the contents of a column then attributes are preserved.

Clearly some attributes like names are invalidated when you extend a list, 
and others may or may not be.  Some parts of R copy all attributes which 
are not obviously invalidated (see mostattributes<-) and others only copy 
those that are obviously valid (and the balance has swung over the years 
towards the first camp).

I'm in the camp that says we are too generous in copying over attributes, 
but here I will make [<- consistent with [[<- and $<-.

Note that e.g. df[, 1:2] drops the attribute, whereas df[,] does not, 
something inherited from lists (and vectors):
> x <- structure(list(a=1,b=2), foo=10)
> attributes(x[1:2])
$names
[1] "a" "b"
> attributes(x[])
$names
[1] "a" "b"

$foo
[1] 10



On Mon, 3 Mar 2008, gunter.berton at gene.com wrote:
>
> Folks:
>
> Problem: [<-.data,frame()  is losing attributes under certain
curcumstances
> shown below.
>
> I think this is a bug, at least in documentation, as I was unable to find
> explicit documentation of the behavior. Indeed, the only documentation I
> found told me attributes are preserved.
I don't think you can expect such detailed documentation -- given how 
little use is made of the documentation we do write no one is going to 
volunteer to write up (and maintain the write up of) such details.
> Here is a detailed description:
>
>> df <- data.frame(a=1:3, b=letters[1:3])
>> attr(df,"foo") <- 10
>> df
>  a b
> 1 1 a
> 2 2 b
> 3 3 c
>> attributes(df)
> $names
> [1] "a" "b"
>
> $row.names
> [1] 1 2 3
>
> $class
> [1] "data.frame"
>
> $foo
> [1] 10
>
> ## But now we add a new column using matrix-type indexing
>
>> df[,"c"] <- 10:12
>> df
>  a b  c
> 1 1 a 10
> 2 2 b 11
> 3 3 c 12
>> attributes(df)
> $names
> [1] "a" "b" "c"
>
> $row.names
> [1] 1 2 3
>
> $class
> [1] "data.frame"
>
>> ## Note that "foo" attribute is lost; however...
>
>
>> df <- data.frame(a=1:3, b=letters[1:3])
>> attr(df,"foo") <- 10
>> df[["c"]] <- 10:12
>> attributes(df)
> $names
> [1] "a" "b" "c"
>
> $row.names
> [1] 1 2 3
>
> $foo
> [1] 10
>
> $class
> [1] "data.frame"
>
> ## The foo attribute has been kept this time.
>
> ## Details:
>
>> sessionInfo()
> R version 2.6.2 (2008-02-08)
> i386-pc-mingw32
>
> locale:
> LC_COLLATE=English_United States.1252;LC_CTYPE=English_United
> States.1252;LC_MONETARY=English_United
> States.1252;LC_NUMERIC=C;LC_TIME=English_United States.1252
>
> attached base packages:
> [1] graphics  grDevices datasets  grid      tcltk     stats     utils
> methods   base
>
> other attached packages:
> [1] svSocket_0.9-5    svIO_0.9-5        svIDE_0.9-5       R2HTML_1.58
> svMisc_0.9-5      lme4_0.99875-9
> [7] Matrix_0.999375-4 lattice_0.17-6
>>
>
>
> Cheers,
>
> Bert Gunter
> Genentech Nonclinical Statistics
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595