RSiteSearch for words ``as one entity''.

If firefox is involved on a Mac, that is one difference.  My Mac is using 
'open' (the default, I believe) and that is calling Safari (the default, I 
believe).

Yes, it would be the job of browseURL to encode URLs where required.  But at 
least on Windows and using the -remoteURL mechanism (as on Linux) it is not. 
R itself does have URLencode.

On Thu, 11 Sep 2008, Rolf Turner wrote:

On 11/09/2008, at 10:09 AM, Marc Schwartz wrote:

	<snip>

As Prof. Ripley noted, try using debug(RSiteSearch) and then step
through the code to see where $HOME is being added to the URL variable.
That can help pin it down further. If the URL is not corrupted by the
time browseURL() is called, then use debug(browseURL) to trace within
that function to see if something is happening there.

	I have done some digging.  When I do:

	RSiteSearch("logistic regression")

	the RSiteSearch() function calls browseURL(qstring) where qstring is 
qs1 equal to

	"http://search.r-project.org/cgi-bin/namazu.cgi?query=logistic+regression&max=20&result=normal&sort=score&idxname=Rhelp02a&idxname=functions&idxname=docs"


	Similarly when I do RSiteSearch("{logistic regression}") qstring is 
qs2 equal to

	"http://search.r-project.org/cgi-bin/namazu.cgi?query={logistic+regression}&max=20&result=normal&sort=score&idxname=Rhelp02a&idxname=functions&idxname=docs"

	i.e. the same except for the braces.

	If I do browseURL(qs1) outside of RSiteSearch() it works as expected.
	If I do browseURL(qs2) I get the error.

	Inside browseURL I find that it calls

	system(paste(browser, remoteCmd, "> /dev/null 2>&1 ||", browser, 
quotedUrl, "&"))

	and browser is "/usr/bin/open", remoteCmd is the qstring with the 
opening and closing
	quote marks attached (escaped with a ``\'') amd quotedUrl is the 
same.  This is true
	irrespective of which qstring (qs1 or qs2 is used).

	Note that the double backslash shows up in remoteCmd == quotedUrl at 
this stage.

	Finally I put the system command into a file ``junk'' and from the 
command window
	(outside R) executed ``sh junk''.

	With the braces around ``logistic regression'' I can the same error 
message that
	I got in R.  When I simply edited out the braces, the Firefox window 
opened at
	the appropriate URL.  So the problem is in those braces and the 
single/double
	backslash at the start was a red herring.

	Finally, while I was doing all this I got Henrik Bengtsson's message, 
and tried
	his suggestion:

	library(R.utils)
	browseURL(toUrl(qs2))

	and BINGO!!! that worked.  (Thanks very much, Henrik.)

	It's still mysterious why the braces have their deleterious impact on 
my Mac
	but do not have that impact on Professor Ripley's machine.  (But 
that's the
	story of my life:  THEY always pick on ***me***! :-) )

	So my problem is solved --- via a workaround --- but there's still 
some vestigial
	strangeness.

	Thanks to all who helped me, especially (again) to Henrik Bengtsson

		cheers,

			Rolf

######################################################################
Attention:\ This e-mail message is privileged and confid...{{dropped:9}}

-- 
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