[R] help.start in R-1.7.0 with Netscape 7.0.

[Prof Brian Ripley]

On Mon, 21 Apr 2003, Rolf Turner wrote:

> No avuncular advice has been forthcoming from the list about
> my problem with getting html-help, but I've been fiddling around
> with it myself and have had a certain amount of insight.  (???)
> 
> The source of the problem seems to be the way that the variable
> isLocal is defined in browseURL():
> 
> 	isLocal <- length(grep("^(localhost|):",
>                           Sys.getenv("DISPLAY"))) > 0
> 
> This seems to make isLocal FALSE in any context in which I work.
> 
> For instance, I am currently sitting at an NCD Xwindows terminal,
> logged into host ``erdos''.  If I execute
> 
> 	> Sys.getenv("DISPLAY")
> 
> I get
>               DISPLAY 
> "131.202.169.153:0.0"
> 
> the string of digits being the IP address of the terminal.  Not
> a "^(localhost|):" in sight.

That's deliberate, and necessary to work around loopholes in Netscape's
remote access protocols.

> I thought to rectify this --- in a kludgy and perhaps dangerous
> way --- by creating for myself a local version of browseURL()
> in which isLocal is always TRUE.  But a funny thing happened:
> 
> I ***still*** got the annoying phenomenon of Netscape's trying
> to open a new browser window (and wanting a new ``profile'')
> every time I asked for help on another function.

Yes, that's intentional, see below.

> To investigate a bit, I created a local version of help() in
> which I could stick some browser() calls to find out the values
> of certain variables.  Blow me down, but didn't things now
> work just fine!  Took out the browser() calls; things still
> worked.  Removed the local help() --- back to not working.
> 
> Even more mysterious to me is the fact that if I just put in
> a local help(), by doing
> 
> 	> help <- help
> 
> then things don't work.  But (ah-ha!) the way I created my local copy
> of help() was to make a local ascii file of the code of help() (with
> a ``help <- '' added at the beginning) and then source this file.
> When this procedure is followed, things ***do*** work!
> 
> This applies even though
> 
> 	> all.equal(help,get("help",pos=grep("base",search()))
> 
> yields TRUE, so that nothing has actually changed in my local copy.
> 
> So help() doesn't find the local copy of browseURL() unless there is
> also a local copy of help() --- and that local copy needs to have
> been created by sourcing a text file!!!  This doesn't make any sense
> to me.  It verges on the mystical.  What on earth is going on?

Namespaces.

> More to the point is there any way that I can persuade help()
> that ``isLocal'' should be taken to be TRUE?
> 
> I don't follow the rationale of ``isLocal'' anyhow.  If I'm sitting
> at machine melvin, but am typing away into a window that is connected
> to host clyde, and I start R in that window, then help.start() will
> invoke the Netscape that lives on clyde.  (Steps must of course have
> been taken to permit clyde to open a Netscape window on melvin's
> display, but that's all done in the normal course of events.)  Hence
> the Netscape that gets fired up will look for the *.html help files
> on clyde and everything will work just as if I were sitting at
> clyde's console.  So why would ``isLocal'' ***ever*** need to be
> FALSE?

That's not how it works.  You would be using *any* netscape window on
melvin, not just one started from clyde.

You are shouting at the wrong audience here: the problem is the Netscape
protocol and how it is implemented (or not) in various versions of
Netscape.  Under Linux/Unix it does not even start a window if not exists 
in 7.0, but did before, yet it does not allow multiple windows which it 
did before.  That `verges on the mystical' indeed.

-- 
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