R devel - Jul 2012 - Questions about imports to a namespace

Hi,

I am a bit unsure about using imports in packages and have a few
questions. I'm sure the answer to some of these is 'it depends', but
I'm interested in what others think and 'best practices' when this
is
the case.



1) If I use an import or importFrom declaration in a NAMESPACE, should
I also qualify the package in calls to the imported function (e.g.
maptools::readShapePoly)?

2) If the 'best practice' is to qualify the calls, when does one stop;
e.g. is it good/bad to say base::sapply?

3) If one is calling a (S3) generic, should one qualify the namespace?
Can this cause problems?

4) Is it better to check for arguments that are missing, or use a
default value of, say, NA; i.e. is missing(arg) preferable to
is.na(arg)?

5) I would like to declare a (S3) generic 'cleanup(x)' but
'cleanup(...)' already exists in R.utils, which my package depends on.
Should I do it?

6) I am used to writing 'requires(package)' for each package I use
within a function. If I have used import or importFrom in my
NAMESPACE, should I still do this?



Looking forward to your advice,
Simon Knapp

On 12-07-02 10:53 PM, Simon Knapp wrote:
> Hi,
>
> I am a bit unsure about using imports in packages and have a few
> questions. I'm sure the answer to some of these is 'it
depends', but
> I'm interested in what others think and 'best practices' when
this is
> the case.
>
>
>
> 1) If I use an import or importFrom declaration in a NAMESPACE, should
> I also qualify the package in calls to the imported function (e.g.
> maptools::readShapePoly)?
No, you won't need that.  I would expect it to slow things down a bit, 
since it needs two searches instead of one:  look up "maptools", look
up
"readShapePoly".
>
> 2) If the 'best practice' is to qualify the calls, when does one
stop;
> e.g. is it good/bad to say base::sapply?
>
> 3) If one is calling a (S3) generic, should one qualify the namespace?
> Can this cause problems?
I think it wouldn't hurt other than the slight speed hit.
> 4) Is it better to check for arguments that are missing, or use a
> default value of, say, NA; i.e. is missing(arg) preferable to
> is.na(arg)?
That depends.  It's easier to set an argument to a special value in a 
call from another function than it is to conditionally set it to be 
missing, but there are only a couple of general purpose choices for 
"missing" values:  NA and NULL, and you might want a user to be able
to
specify those.  (You can also use negative values for counts, etc., in 
particular contexts.)
> 5) I would like to declare a (S3) generic 'cleanup(x)' but
> 'cleanup(...)' already exists in R.utils, which my package depends
on.
> Should I do it?
No.  Use a different name.  If you are using R.utils, your users are 
also using it, and their code may want to call R.utils::cleanup without 
the R.utils:: prefix.
> 6) I am used to writing 'requires(package)' for each package I use
> within a function. If I have used import or importFrom in my
> NAMESPACE, should I still do this?
No, it just slows things down, and can cause other problems for users 
because it changes the search list.

Duncan Murdoch