R devel - Feb 2010 - grid unit bug? (PR#14220)

The following seems to me to be at least a perverse trap, if not an outright
bug:
> is.numeric(unit(1,"npc"))
[1] TRUE
> is.numeric(1*unit(1,"npc"))
[1] FALSE
> is.numeric(unit(0,"npc") +unit(1,"npc"))
[1] FALSE

...etc.
i.e. is.numeric() appears to be TRUE for class "unit" but false for
class
("unit.arithmetic" "unit" ). Seems to me it ought to b the
same for both.


Bert Gunter
Genentech Nonclinical Biostatistics

(FWIW, I think grid graphics is brilliant!)

This was R version 2.11.0dev for Windows btw (not that it makes a
difference):

sessionInfo()

R version 2.11.0 Under development (unstable) (2010-02-15 r51142)=20
i386-pc-mingw32=20

locale:
[1] LC_COLLATE=3DEnglish_United States.1252=20
[2] LC_CTYPE=3DEnglish_United States.1252  =20
[3] LC_MONETARY=3DEnglish_United States.1252
[4] LC_NUMERIC=3DC                         =20
[5] LC_TIME=3DEnglish_United States.1252   =20

attached base packages:
 [1] datasets  splines   grid      tcltk     stats     graphics  grDevices
 [8] utils     methods   base    =20

other attached packages:
[1] TinnR_1.0.3     R2HTML_1.59-1   Hmisc_3.7-0     survival_2.35-8
[5] svSocket_0.9-48 lattice_0.18-3  MASS_7.3-5    =20

loaded via a namespace (and not attached):
[1] cluster_1.12.1 svMisc_0.9-56



=A0
=A0

Paul:

I figured that would be the problem.

I encountered the issue when I tries to check arguments in a validDetails
method for a grob.

Could one substitute the following function in the grid namespace?

is.numeric <- function(x)if(is.unit(x))TRUE else is.numeric(x)

(or make the first clause FALSE)

Obviously, messing around like this might be dangerous -- or at least would
compromise execution speed.

Cheers,
Bert

Bert Gunter
Genentech Nonclinical Biostatistics
 
 

-----Original Message-----
From: paul murrell [mailto:R-bugs at r-project.org] 
Sent: Wednesday, February 24, 2010 4:22 PM
To: gunter.berton at gene.com
Subject: Re: grid unit bug? (PR#14220)
> The following seems to me to be at least a perverse trap, if not an >
outright
> bug:
> 
>> is.numeric(unit(1,"npc"))
> [1] TRUE
>> is.numeric(1*unit(1,"npc"))
> [1] FALSE
>> is.numeric(unit(0,"npc") +unit(1,"npc"))
> [1] FALSE
> 
> ...etc.
> i.e. is.numeric() appears to be TRUE for class "unit" but false
for > class
> ("unit.arithmetic" "unit" ). Seems to me it ought to b
the same for > both.

These results simply reflect the underlying data structures (simple
"unit"s
are
just numeric vectors, but more complex "unit.arithmetic"s are lists). 
I
don't
see how I can "hide" the numeric-ness of simple units (just like
there's no
way
to stop a "ts" object like 'Nile' from satisfying
is.numeric()).  I could
re-implement simple units as lists, but (apart from the work involved) that
would be (even) less efficient.

1. Is there a situation where this causes a problem?

2. Do you have a possible "fix" in mind?

Paul

> 
> Bert Gunter
> Genentech Nonclinical Biostatistics
> 
> (FWIW, I think grid graphics is brilliant!)
> 
> This was R version 2.11.0dev for Windows btw (not that it makes a
> difference):
> 
> sessionInfo()
> 
> R version 2.11.0 Under development (unstable) (2010-02-15 r51142)=20
> i386-pc-mingw32=20
> 
> locale:
> [1] LC_COLLATE=3DEnglish_United States.1252=20
> [2] LC_CTYPE=3DEnglish_United States.1252  =20
> [3] LC_MONETARY=3DEnglish_United States.1252
> [4] LC_NUMERIC=3DC                         =20
> [5] LC_TIME=3DEnglish_United States.1252   =20
> 
> attached base packages:
>  [1] datasets  splines   grid      tcltk     stats     graphics  >
grDevices
>  [8] utils     methods   base    =20
> 
> other attached packages:
> [1] TinnR_1.0.3     R2HTML_1.59-1   Hmisc_3.7-0     survival_2.35-8
> [5] svSocket_0.9-48 lattice_0.18-3  MASS_7.3-5    =20
> 
> loaded via a namespace (and not attached):
> [1] cluster_1.12.1 svMisc_0.9-56
> 
> 
> 
> =A0
> =A0
> 
>