R devel - Feb 2006 - Bug in Sweave? -- scoping problem? (PR#8615)

I have found a strange scoping problem in Sweave.  The following Rnw
file doesn't produce the same output in Sweave as it does if I produce
an R file using Stangle and execute that:

\documentclass[12pt]{article}
\begin{document}
<<R>>election <- data.frame(A=1:3, B=9:7, C=rep(0,3))
partytotal <- rep(0, ncol(election))
for (i in 1:ncol(election)) {
      partytotal[i] <- sum(election[,i], na.rm=TRUE)
}
@

<<fig=true,width=12,height=6,echo=true>>names(partytotal) <-
names(election)
partytotal <- sort(partytotal)
partytotal
dotchart(partytotal)
@
\end{document}

The data.frame "election" contains columns which are vote counts for 3
parties, with rows being 3 electoral districts.   The column totals are
6 for A, 24 for B, and 0 for C.  I calculate those totals in a vector
partytotal, assign the names to its entries, and then sort it.

The strange thing is that while the value in partytotal is output 
correctly as

\begin{Sinput}
  > names(partytotal) <- names(election)
  > partytotal <- sort(partytotal)
  > partytotal
\end{Sinput}
\begin{Soutput}
   C  A  B
   0  6 24
\end{Soutput}
\begin{Sinput}
  > dotchart(partytotal)
\end{Sinput}

but the dotchart contains the wrong values:  it shows sorted
values, but not sorted names, which is also what is left over in my
workspace after the Sweave call:

  > partytotal
   A  B  C
   0  6 24

These tests were run in R 2.2.1, but I see the same thing in today's 
R-devel.

Duncan Murdoch

G'day Duncan,
>>>>> "DM" == murdoch  <murdoch at stats.uwo.ca>
writes:
    DM> I have found a strange scoping problem in Sweave.  [...]

    DM> The strange thing is that while the value in partytotal is
    DM> output correctly as [...]

    DM> but the dotchart contains the wrong values: it shows sorted
    DM> values, but not sorted names, [...]
No bug, but a feature. :)
(But then, a well known software producer seems to use `feature' as a
euphemism for `bug', so perhaps I shouldn't call it a feature.)

There was recently a discussion on r-help (?) about Sweave producing
different output in the text and plot when random numbers were
generated, and it seems as if you have run into the same trap:

      Code in chunks that produce pictures is executed several times.
      First, to produce the output in the text.  And then once more
      for *each* format in which the figure has to be produced.  I.e.,
      if you want a PDF and a PostScript version of the figure, the
      code is executed a total of three times.

All instances of this feature reported so far involved commands that
produced random numbers and the poster was surprised that the output
in the text and the figures differed (and that the two figures were
different).

In your case the first execution of the code assigns the names to
partytotal, sorts partytotal and produces the data.  On the next
execution, when the picture is produced, partytotal is already sorted
but you reassign the names.  Then the partytotal is sorted again and
the plot is produced.  But that re-assigning of names lead to the
disconnect between values and names.

Hope this helps.

Cheers,

        Berwin

========================== Full address ===========================Berwin A
Turlach                      Tel.: +61 (8) 6488 3338 (secr)
School of Mathematics and Statistics        +61 (8) 6488 3383 (self)      
The University of Western Australia   FAX : +61 (8) 6488 1028
35 Stirling Highway                   
Crawley WA 6009                e-mail: berwin at maths.uwa.edu.au
Australia                        http://www.maths.uwa.edu.au/~berwin

Berwin A Turlach wrote:
> G'day Duncan,
> 
> 
>>>>>>"DM" == murdoch  <murdoch at
stats.uwo.ca> writes:
> 
> 
>     DM> I have found a strange scoping problem in Sweave.  [...]
> 
>     DM> The strange thing is that while the value in partytotal is
>     DM> output correctly as [...]
> 
>     DM> but the dotchart contains the wrong values: it shows sorted
>     DM> values, but not sorted names, [...]
> No bug, but a feature. :)
> (But then, a well known software producer seems to use `feature' as a
> euphemism for `bug', so perhaps I shouldn't call it a feature.)
> 
> There was recently a discussion on r-help (?) about Sweave producing
> different output in the text and plot when random numbers were
> generated, and it seems as if you have run into the same trap:
> 
>       Code in chunks that produce pictures is executed several times.
>       First, to produce the output in the text.  And then once more
>       for *each* format in which the figure has to be produced.  I.e.,
>       if you want a PDF and a PostScript version of the figure, the
>       code is executed a total of three times.
Thanks, that's what caught me.  Is that a quote from the discussion, or 
from the docs somewhere?  It makes sense in hindsight, but it's not 
obvious ahead of time, so it should be stated fairly prominently in the 
docs.

Duncan Murdoch
> 
> All instances of this feature reported so far involved commands that
> produced random numbers and the poster was surprised that the output
> in the text and the figures differed (and that the two figures were
> different).
> 
> In your case the first execution of the code assigns the names to
> partytotal, sorts partytotal and produces the data.  On the next
> execution, when the picture is produced, partytotal is already sorted
> but you reassign the names.  Then the partytotal is sorted again and
> the plot is produced.  But that re-assigning of names lead to the
> disconnect between values and names.
> 
> Hope this helps.
> 
> Cheers,
> 
>         Berwin
> 
> ========================== Full address ===========================>
Berwin A Turlach                      Tel.: +61 (8) 6488 3338 (secr)
> School of Mathematics and Statistics        +61 (8) 6488 3383 (self)      
> The University of Western Australia   FAX : +61 (8) 6488 1028
> 35 Stirling Highway                   
> Crawley WA 6009                e-mail: berwin at maths.uwa.edu.au
> Australia                        http://www.maths.uwa.edu.au/~berwin

I will add it to the FAQ.

fritz
>>>>> On Sun, 19 Feb 2006 08:42:09 -0500,
>>>>> Duncan Murdoch (DM) wrote:
[...]
>       Code in chunks that produce pictures is executed several times.
>       First, to produce the output in the text.  And then once more
>       for *each* format in which the figure has to be produced.  I.e.,
>       if you want a PDF and a PostScript version of the figure, the
>       code is executed a total of three times.
  > Thanks, that's what caught me.  Is that a quote from the discussion,
or
  > from the docs somewhere?  It makes sense in hindsight, but it's not 
  > obvious ahead of time, so it should be stated fairly prominently in the 
  > docs.


  > Duncan Murdoch
  >> 
  >> All instances of this feature reported so far involved commands that
  >> produced random numbers and the poster was surprised that the output
  >> in the text and the figures differed (and that the two figures were
  >> different).
  >> 
  >> In your case the first execution of the code assigns the names to
  >> partytotal, sorts partytotal and produces the data.  On the next
  >> execution, when the picture is produced, partytotal is already sorted
  >> but you reassign the names.  Then the partytotal is sorted again and
  >> the plot is produced.  But that re-assigning of names lead to the
  >> disconnect between values and names.
  >> 
  >> Hope this helps.
  >> 
  >> Cheers,
  >> 
  >> Berwin
  >> 
  >> ========================== Full address =========================== 
>> Berwin A Turlach                      Tel.: +61 (8) 6488 3338 (secr)
  >> School of Mathematics and Statistics        +61 (8) 6488 3383 (self)
  >> The University of Western Australia   FAX : +61 (8) 6488 1028
  >> 35 Stirling Highway                   
  >> Crawley WA 6009                e-mail: berwin at maths.uwa.edu.au
  >> Australia                        http://www.maths.uwa.edu.au/~berwin