[R] Trouble with HTML search engine

[Prof Brian Ripley]

On Thu, 22 Apr 2004, Damon Wischik wrote:

> 
> There have been a number of posts to this list by people having trouble
> with the HTML search engine. Often these troubles are caused by incorrect
> setups (user hasn't installed Java properly, or Java is disabled, or
> Javascript is disabled). Sometimes the trouble persists even when Java is
> installed properly.
> 
> I have written an alternative HTML search engine, which is based on
> Javascript rather than Java. (Hopefully, this means that there is less
> that can go wrong.) I haven't wrapped it up in a package yet, because I
> don't know how well it works or even if there is any interest. If you
> would like to try it, you can:
> 
> source("http://www.statslab.cam.ac.uk/~djw1005/Stats/Interests/search.R")
> helpHTML()
> 
> (Or you can download the code and run it locally. You will also need to
> download searchtemplate.html from the same location. You can then run
> helpHTML(searchTemplate=localfilename) to tell it to use your local copy
> of searchtemplate.html.)
> 
> ISSUES WITH MY CODE
> -------------------
> I've tested it on Windows XP cross {R1.8.0,R1.9.0} cross {IE,Firebird}.
> I've also tested it on Debian 3.0 with R1.8.0 cross {Mozilla,Firebird}.
> I'd be grateful to learn whether it works elsewhere.
> 
> Searching is a bit slow. On my newish computer it takes three seconds or
> so. On an older departmental machine it takes ten seconds. Such is
> (Javascript) life.
> 
> My searching algorithm is not the same as the current searching algorithm.
> I wrote this for my own use, and so I've used a scoring mechanism which
> reflects the way I like to search. The text at the top of the search page
> explains some of the options.
> 
> There are obviously things I don't understand about the current help
> setup. If anyone is sufficiently interested in this to explain them to me,
> I would be grateful. (1) On my Windows XP setup, R writes an index in the
> directory it was installed. What if it doesn't have write permission?  

That *is* described in the rw-FAQ.

> (2) On my Debian setup, R copies all of the HTML help into a temporary
> directory. Why not just refer to the files where they are, rather than
> copying them all across? Because I don't understand these two points,
> I've

It doesn't, it copies some files and links others.  The reason it copies
some is that fairly recent security checks mean the code fails if those
are links.  The reason that it copies/links is that the HTML files have
links of the form ../../pkgname/html/funcname.html which will not work
across library trees.

> written my indexing routines to (a) create a search index in a temporary
> directory, and (b) refer to the files in their install directories. My
> indexing routines run the same under both Windows and Linux.
> 
> 
> ISSUES WITH R ------------- 
> The R "Installation and administration"  document tells us that "Sun's
> Java Run-time Environment j2re 1.4.2_02 does not work under Linux". Prof
> Ripley said on 1 April 2004 that "if Linux/Unix, Sun JRE 1.4.2_02/3/4 are
> broken"  This is news to those like myself who run Linux with Sun's JRE
> 1.4.2_03 and find that all their other applets work fine. (Though I'm sure
> there are bugs in the JRE, as in most complex projects.)

Note that a patch-level release stopped existing code working, on some but 
not all platforms.  If we did that in R without any mention in the relase 
notes, I am sure you would complain.

> On my computer, the trouble boiled down to this: the Javascript which
> displays search results was unable to interface with the Java applet which
> performed the search. As far as I am aware, there are no published
> standards which govern this interface. Therefore it is necessary to rely
> on vendor documentation (insofar as we can say that organizations which
> distribute free software are vendors). In the case of Mozilla, this
> interface is called LiveConnect; some documentation is available at
> mozilla.org. Generally speaking, an object on a web page may export
> certain methods, making them available to Javascript. For example, an
> object which contains a Java applet typically makes available the static
> methods of the classes in that applet. Again, I am not aware of any
> published standards on which methods are exported, so again we have to
> rely on vendor documentation. In the case of Sun's Java, the documentation
> explains how to use these exported methods
> http://java.sun.com/j2se/1.4.2/docs/guide/plugin/developer_guide/js_java.html
> However, the R HTML search page does not follow this documentation. I
> found that if I alter the R HTML search page to conform to this
> documentation, it works.

Can you tell us exactly what changes you made, please?

> It is always going to be difficult to write portable code when there are
> no published standards, only vendor-specific documentation. I have
> therefore attempted, in my Javascript search, to stick to pure ECMAscript
> (though undoubtedly I have failed in places).

-- 
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