[R] Can't there be a cd command?

[Duncan Murdoch]

On 5/10/2006 11:10 AM, Gabor Grothendieck wrote:
> On 5/10/06, Duncan Murdoch <murdoch at stats.uwo.ca> wrote:
>> On 5/10/2006 10:45 AM, Gabor Grothendieck wrote:
>> > On 5/10/06, Duncan Murdoch <murdoch at stats.uwo.ca> wrote:
>> >> On 5/10/2006 9:29 AM, Paul Johnson wrote:
>> >> > It is a FAQ in our Linux lab.  People start emacs and fire up R via
>> >> > ess, and then they have no idea 'where they are".  For computer
>> >> > experts, it is not a problem, but for people who don't know much about
>> >> > computers, it is a pretty big problem.  They have data in some
>> >> > subdirectory, but almost invariably they don't get emacs & R started
>> >> > from that same place.
>> >> >
>> >> > Unfortunately, for our users, it does not help to simply re-label
>> >> > setwd as cd.  Both commands imply a deeper understanding of the OS
>> >> > than they have.  Also, unfortunately, these are the same people who
>> >> > don't understand that FAQs exist and should be consulted. These people
>> >> > are so new/timid that asking in r-help would be the last thing to
>> >> > cross their mind.
>> >> >
>> >> > I've wondered if it would not help to have the R prompt include the
>> >> > directory name, as in an x terminal.
>> >>
>> >> I think file system directories aren't as central in R as they are in a
>> >> shell, so it would just be distracting.  Most of the time I work in the
>> >> R workspace, not in the file system.
>> >>
>> >> To me the solution is to allow interactive file selection by default,
>> >> i.e. the default on read.table and similar functions should be
>> >> file.choose(), rather than having no default and throwing an error.
>> >> This won't help you in the short run (because file.choose() on Linux
>> >> isn't all that friendly to beginners), but perhaps it would encourage
>> >> someone to make it better.  file.choose() is quite nice in Windows (and
>> >> I think on the Mac), so beginners there could be told
>> >>
>> >> mydf <- read.table()
>> >>
>> >> and they'd get something useful.
>> >>
>> >> Martin Maechler has disagreed with me about this in the past, but hasn't
>> >> convinced me that he's right, he's just convinced me that doing nothing
>> >> is easier than arguing about it.
>> >
>> > I agree with Martin regarding read.table; however, the underlying idea is
>> > good and could be achieved via simple wrappers which are the same
>> > as the corresponding underlying functions except for the default argument
>> > to file:
>> >
>> >    read.table.choose <- function(file = file.choose(), ...)
>> > read.table(file, ...)
>> >    read.csv.choose <- function(file = file.choose(), ...) read.csv(file, ...)
>> >    read.delim.choose <- function(file = file.choose(), ...)
>> > read.delim(file, ...)
>> >
>> >   # test
>> >   mydata <- read.table.choose()
>> >
>> > in a package available to the users or possibly even in R core.
>>
>> No, I don't think this is a good idea.  It would be just as easy to tell
>> people to type
>>
>> read.table(file=file.choose())
>>
>> with no new package or function necessary.  I want the existing basic
>> function to work when used by a beginner in a simple way.
> 
> I don't think that an idiom of multiple function calls is as simple as
> issuing a single function call with no args.
> 
> One possibility is to have
> a keyword "choose" on the file function in the same way that file
> accepts the keyword "clipboard".  Then one could write:
> 
>   mydata <- read.table("choose")
> 
> I think the wrapper approach is probably preferable though and seems
> consistent with the way R deals with this in other places such as the
> ISOdate wrapper.
> 
> 
>>
>> What is it that you find objectionable about having a default for the
>> file argument in read.table?  I think Martin has said that he doesn't
>> want non-UI functions to be involved with UI functions, but I don't see
>> that:  if your code works now, it will be completely unaffected by
>> setting a default for the argument.  (Sorry if I summarized the argument
>> incorrectly, Martin, I didn't look it up.)
> 
> That would be my objection too.  UI should not be tied to the non-UI core.
> Its basically a loose coupling argument.

I don't accept that argument, because in R everything* is interactive. 
There isn't a non-UI core.  The function arguments are part of the user 
interface.

Duncan Murdoch

* Well, maybe this should be restricted to functions visible to the 
user, but everything we've been discussing so far is visible and is part 
of the user interface.