Skip to content

R version on gifi

21 messages · Thomas Lumley, Jan de Leeuw, A.J. Rossini +4 more

Original
Jan de Leeuw
# 
Pretty soon, hopefully, the RAqua version will make the Darwin/X11
version unnecessary. In the meantime, for convergence, I'll modify the  
Gifi
version in various ways.

-- It's 1.7.1
-- It no longer supports gnome
-- It still uses Tcl/Tk for X11 (using 8.5 from CVS)
!!  It no longer uses anything from fink (readline and dlcompat
     as on Stefano's site, jpeg and png and teTeX from Gerben Wierda's
     i-installer, Tcl/Tk from cvs, X11 from Apple)
-- the package installer will put everything in /usr/local

This means that all references to /sw in configure.ac can go. This
next distribution will appear probably on wednesday. Undoubtedly
some packages will break, because they still use stuff from /sw,
but I'll fix those as we go along.
===
Jan de Leeuw; Professor and Chair, UCLA Department of Statistics;
Editor: Journal of Multivariate Analysis, Journal of Statistical  
Software
US mail: 9432 Boelter Hall, Box 951554, Los Angeles, CA 90095-1554
phone (310)-825-9550;  fax (310)-206-5658;  email: deleeuw@stat.ucla.edu
homepage: http://gifi.stat.ucla.edu
   
------------------------------------------------------------------------ 
-------------------------
           No matter where you go, there you are. --- Buckaroo Banzai
                    http://gifi.stat.ucla.edu/sounds/nomatter.au
Don MacQueen
# 
Couple of questions...

Thanks
-Don
At 10:00 AM -0700 6/16/03, Jan de Leeuw wrote:
Is the user expected to have separately installed any or all of these 
various libraries, or are they included in the R distribution and 
installed by the R installer? Or what?

Specifically for X11, does it assume the user has separately 
installed Apple's X11 and QuartzWM, and if so, is it in any way 
dependent on anything unique to Apple's X11? That is, will it work if 
the user is using XFree86/XDarwin and some (any) other window manager?
Do you mean that at some point in the future you intend that the 
configure.ac in the source distribution will remove all references to 
/sw? I'm not sure this is a good idea; I think I would prefer to have 
the option of building from sources using fink for those other things 
(readline, jpeg, png, tetex, etc) if I want to. Otherwise I have to 
learn how to get them from several other sites, increasing my system 
maintenance load and making it harder to keep them up to date.

Can you give specific and substantive reasons why fink should be avoided?

I get the impression that R for OS X is being moved away from being 
another unix R variant (in the sense that Solaris, various Linuxes, 
SGI, etc. are unix variants), and moved toward being a specialized 
platform-specific version. Assuming my impression is more or less 
correct, I'd like to understand the pros and cons of this move.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      stefano iacus
    

    

      
      #
    

  


  

      
On Luned?, giu 16, 2003, at 20:45 Europe/Rome, Don MacQueen wrote:

        

            

      
      
      
    

        
on this specific point: Darwin R will remain another unix-like build of 
R as it is now but with an additional GUI (the aqua module) and device 
(quartz). You can still build R-devel using fink.
RAqua is just a doubleclicking version of Darwin R.
Nothing will change for the developer I guess, but it will simplify the 
life to MacOS X end users.

I don't think there is the need to remove /sw from the search path at 
configure time.


The motivation of making an Aqua version of R is that I knew about 
several (lots of) people using OS X and Carbon R instead of Darwin/X11 
which is bad as Carbon R is becoming too limited (and for this reason 
no more supported starting with 1.8.0)


stefano

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Thomas Lumley
    

    

      
      #
    

  


  

      
On Mon, 16 Jun 2003, Don MacQueen wrote:

            

      
      
      
    

        
I think the move is to stop *assuming* fink, not to forbid it.

There are at least two goals

1. To produce a binary OS X distribution for people who don't want to
learn Unix, including an (optional) GUI and an (optional) quartz() device
driver.  If you want to stick with a command line and x11(), you're
welcome to. If you want to compile from sources yourself you can competely
ignore the binary distribution. This binary can't install
anything in /sw (because that would clobber fink) and doesn't want to
assume that anything is already there.


2. To produce a simple minimal set of instructions and dependencies for
compiling R on Mac OS X. A set of instructions that includes installing
fink is (I fervently hope) not minimal.  Of course, if you already have
fink installed you can easily use it to get all the necessary software.

You might in future have to specify a configure option to get configure to
look in /sw for system libraries, but this makes it more similar to other
Unices, not less similar.


	-thomas

  
  


  


      

    

    

      
      

        


  

        

      Jan de Leeuw
    

    

      
      #
    

  


  

      
Standard Apple stuff (such as anything in /usr, including X11) will
not be included in the installer package, but everything needed in  
/usr/local
will

I assume it will work with XFree86/XDarwin and with OroborosX, but
I am not going to test it with those, because I don't use them anymore.

I am sure that if more people want /sw, then it will just stay in. I am
just saying that as far as my distribution (and Stefano's) are concerned
it is not needed any more.

Of course you can always use it if you want to by setting LDFLAGS.

I try to avoid fink because it is a maze of nasty dependencies between
packages, which we do not need for R. It is also very inconvenient
and potentially disastrous to have duplicates of binaries and dynamic
libs in /sw and /usr/local and /usr. For instance, when Apple added
ncurses, many things went wrong. Fink is getting more and more
system-foo packages to deal with cases in which there are alternative
installs, and this will obviously become worse over time as more and
more gets added to Darwin. So (a) from the point of view of R
fink is overkill, and (b) fink has been great to have around for three
years but it is reaching the end of its usefulness period. Only my
opinion, of course.

Your impression that R is moved away from "just another Unix
variant" is correct only insofar as RAqua is being added as another
GUI. All the other GUIs (Terminal, Tcl/Tk, XTerm) remain available,
and for those savvy enough to build R fink and gnome can still be
used. There will just be no binaries.

--- Jan

        
On Monday, Jun 16, 2003, at 11:45 US/Pacific, Don MacQueen wrote:

        

            

      
      
      
    

        
===
Jan de Leeuw; Professor and Chair, UCLA Department of Statistics;
Editor: Journal of Multivariate Analysis, Journal of Statistical  
Software
US mail: 9432 Boelter Hall, Box 951554, Los Angeles, CA 90095-1554
phone (310)-825-9550;  fax (310)-206-5658;  email: deleeuw@stat.ucla.edu
homepage: http://gifi.stat.ucla.edu
   
------------------------------------------------------------------------ 
-------------------------
           No matter where you go, there you are. --- Buckaroo Banzai
                    http://gifi.stat.ucla.edu/sounds/nomatter.au

  
  


  


      

    

    

      
      

        


  

        

      Simon Urbanek
    

    

      
      #
    

  


  

      
On Monday, June 16, 2003, at 08:45  PM, Don MacQueen wrote:

        

            

      
      
      
    

        
The really native version doesn't really need to depend on X11 anymore 
since the use of X11 on Mac OS X was meant for applications that are 
not properly ported to OS X yet. Once Quartz and RAqua are complete 
there is no need for X11.

            

      
      
      
    

        
Jan already listed the main technical reasons why it is indeed a very 
good idea. Apart from that, fink is not an official package and was 
only meant as a temporary solution for people who need a (no matter how 
ugly) way to run existing unix programs on OS X. Hardly any real OS X 
user has installed fink (especially since Jaguar is out). Fink was 
great during the first couple of months when native OS X ports were 
hardly existent, but is now obsolete for mainstream OS X use.

            

      
      
      
    

        
It is not a "move" of R. Mac OS X is simply not "another unix variant". 
Darwin is indeed, but Mac OS X is not. You can compile X11 for Darwin 
and use it exactly the way you can use Linux on a PPC hardware. But Mac 
OS X has many very nice (often proprietary) layers that are important 
to the Mac users, but that part of OS X is not "unix". The goal here is 
to release R which fits in the philosophy of the system - ease of use, 
good integration with the existing frameworks, appealing design. These 
are not properties of unix, but of OS X. So what we need is in fact 
Mac-OS-X-like look and feel. The fact that OS X is unix-based helps 
with respect to the R engine itself - we need no special ports of 
packages anymore, but it has a totally different GUI.

Fortunately R makes a distinction between GUI and the engine, therefore 
we can create a real OS X GUI without affecting other platforms - 
including Darwin ;). "Unix" users are used to compile their own 
software, therefore moving fink support to the category 'optional' is 
only logical, since you can still easily enable it with configure 
parameters and/or environment settings. Real Mac OS X users are used to 
nice, binary distributions, therefore we cannot assume fink and we need 
Quartz device and RAqua. It will be a big help for most OS X users. 
(BTW: no Mac users I know (non-developers) have installed X11.)

Therefore the recent changes are IMHO really important from Mac OS X 
user's view - so far most binaries were rather experimental and assumed 
some unix knowledge (note: there was is no official OS X binary!). It 
was ok to use fink for those as a temporary solution, but the official 
binary cannot rely on unsupported non-Apple packages. The only thing 
external part we really need is libdl and I'm sure we can supply it 
simply with R - such as pcre etc., all other libraries are optional.

Cheers,

Simon

---
Simon Urbanek
Department of computer oriented statistics and data analysis
University of Augsburg
Universit?tsstr. 14
86135 Augsburg
Germany

Tel: +49-821-598-2236
Fax: +49-821-598-2200

Simon.Urbanek@Math.Uni-Augsburg.de
http://simon.urbanek.info

  
  


  


      

    

    

      
      

        


  

        

      stefano iacus
    

    

      
      #
    

  


  

      
On Marted?, giu 17, 2003, at 14:33 Europe/Rome, Simon Urbanek wrote:

        

            

      
      
      
    

        
I completely agree with Simon, Jan, Thomas etc.
About libdl: in fact there is no need to link against it and I'll try 
to integrate it in the R sources.
stefano

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      A.J. Rossini
    

    

      
      #
    

  


  

      
Simon Urbanek <Simon.Urbanek@math.uni-augsburg.de> writes:

            

      
      
      
    

        
I think completely agree with Simon here.  MacOSX is great for folks
that like the good things in life, but for chumps who just want a Unix
variant, run a PPC Unix (linux or *bsd or the like).  There are just
enough differences to create user-interface stress, depending on the
particular system configuration...

(of course, I'm one of those who has been using a unix variant for
almost 20 years now, and with the onset of senility, that isn't going
to change).

best,
-tony

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Don MacQueen
    

    

      
      #
    

  


  

      
I appreciate the extra explanations from Simon, Jan, Thomas, and 
Tony. All in all, after having read what you all have to say, the 
things that I found potentially alarming in Jan's message yesterday 
are not so alarming anymore.

-Don

        
At 2:33 PM +0200 6/17/03, Simon Urbanek wrote:

            

      
      
      
    

        
Except for one major flaw in Aqua--the absence of "focus follows 
mouse", as it is sometimes called in an X Windows context. I consider 
the absence a flaw because I find it incredibly useful to be able to 
type R commands into a window that is partially covered by another 
(usually graphics device) window. This is, of course, a (strongly 
held) personal preference, and I'm well aware that many do not 
consider this to be an issue. The Mac OS has always had this 
limitation, and probably always will, so there's nothing to be done 
about it.

More generally, having RAqua, a version of R that is as "Mac-like" as 
possible, is a good thing; I have co-workers who are more likely to 
use such a version of R. As someone who likes OS X a lot, I like 
anything that promotes it, and a good R gui will do that around here.

            

      
      
      
    

        
Jan cited "Gerben Wierda's i-installer" as a source for jpeg, png, 
and teTex. This source is somehow more "official" than fink? But, 
considering what Jan says, i.e. "everything needed in /usr/local 
will" be included with the installer package, it doesn't matter to 
the end user.

            

      
      
      
    

        
I'm glad to hear that, but when Jan says "all references to /sw in 
configure.ac can go" I do tend to wonder whether that is actually the 
case.

            

      
      
      
    

        
I subscribe to Apples "scitech" mailing list, for people interested 
in using the Mac in scientific applications. I'd say X windows is 
pretty common among those folks. Don't know how many of them would 
consider themselves "developers."

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Simon Urbanek
    

    

      
      #
    

  


  

      
On Tuesday, June 17, 2003, at 05:11  PM, Don MacQueen wrote:

        

            

      
      
      
    

        
Yes, this is indeed a very nice feature (I've been using it on unix all 
time), but it can be disastrous at the same time. MS Windows has an 
undocumented registry key which allows you to enable this, but once you 
do that you'll realize that a lot of applications assume 
'topmost-has-focus' state and are almost unusable if the 
'focus-follows-mouse' is enabled (example: if you have a mouse over a 
toolbar your document window is inactive - most applications can't deal 
with that). I'm not sure about this in OS X (since we can't really test 
it ;P), but something similar might happen.

            

      
      
      
    

        
Exactly, that's the point :) We don't want to assume things that are 
non-standard. We should provide them if necessary.

There is still one issue to consider in this context: source packages. 
A really 'plain' Mac OS X can't be used to install source packages 
as-is, basically because there are three missing things: Dev Tools, g77 
and latex. The first one is official, so we could require that (and 
probably have to). G77 is really just a few files, so the installer 
could add it if necessary, but I'm not sure about latex. Is building 
packages w/o latex documentation an option? The direct use of source 
packages seems to me as the greatest benefit of OSX being unix-based, 
therefore i wouldn't like to miss it, even if I was pure Mac user...

Cheers,
Simon

---
Simon Urbanek
Department of computer oriented statistics and data analysis
University of Augsburg
Universit?tsstr. 14
86135 Augsburg
Germany

Tel: +49-821-598-2236
Fax: +49-821-598-2200

Simon.Urbanek@Math.Uni-Augsburg.de
http://simon.urbanek.info

  
  


  


      

    

    

      
      

        


  

        

      stefano iacus
    

    

      
      #
    

  


  

      
On Marted?, giu 17, 2003, at 17:45 Europe/Rome, Simon Urbanek wrote:

        

            

      
      
      
    

        
We (R-core) have planned to make an automated procedure in order to 
provide prebuilt packages on CRAN for Darwin, so that 
install.packages() can eventually behave like on Windows (at least for 
RAqua).
Btw, I guess g77 will be included in the next release of DevTools.

stefano

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Brian Ripley
    

    

      
      #
    

  


  

      
On Tue, 17 Jun 2003, Simon Urbanek wrote:

        

            

      
      
      
    

        
It's a documented part of TweakUI, and I find it very usable.  After all, 
it I have my mouse over something on the toolbar, I have deliberately
moved focus there (just as in CDE), and it is very natural to someone
used to this from a good Unix windows manager.  Only a very few 
applications cause me problems (most notably the Visual Basic IDE).

            

      
      
      
    

        
It could be.  Checking them is not, though.

            

      
      
      
    

        
At DSC we seemed to decide that we would need a binary packages mechanism 
for the GUI MacOS X port.  I suspect you underestimate the difficulties 
(or overestimate the abilities of the users concerned): the Windows
experience is that is hard to overestimate the ability of the users to 
make stupid errors and not realize what.


As for this being `just another unix version': if only!  Simple things on
any other unix-alike like making a Rlapack dynamic library became major
headaches on darwin, only, and that is still not fully resolved.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Don MacQueen
    

    

      
      #
    

  


  

      
At 5:45 PM +0200 6/17/03, Simon Urbanek wrote:

            

      
      
      
    

        
For an OS X user with no unix background, I would think that pdf is 
the preferred format, and sufficient.

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Brian Ripley
    

    

      
      #
    

  


  

      
On Tue, 17 Jun 2003, Don MacQueen wrote:

        

            

      
      
      
    

        
PDF is produced via latex in R.  Would that really be preferred to HTML?
Nothing in R is setup to produce PDF help, and especially not PDF 
hyperlinked between different documents (which is a problem in a dynamic 
framework).

BTW, could this sort of discussion go on in R-SIG-Mac or R-devel but not 
both since R-SIG-Mac is a closed list.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      stefano iacus
    

    

      
      #
    

  


  

      
I think we have to distinguish what is distributed as binary and the 
building process of R.
If we distribute prebuilt packages for OSX we can provide both html and 
pdf files and leave out latex help.
stefano

        
On Marted?, giu 17, 2003, at 19:33 Europe/Rome, Prof Brian Ripley wrote:

        

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Don MacQueen
    

    

      
      #
    

  


  

      
At 6:33 PM +0100 6/17/03, Prof Brian Ripley wrote:

            

      
      
      
    

        
I didn't know that. Does it require a full latex installation, or 
could one get by with just a subset, perhaps just pdflatex? It's 
probably a bad idea to complicate matters that way, though.

            

      
      
      
    

        
Speaking for myself, yes.

Speaking of which, this is another area where OS X is not like other 
versions of unix; the "native OS X" way to start a browser on the 
command line is with the "open" command, i.e., "open filename.html". 
OS X needs, for example,
     browseURL('somefile.html',browser='open')

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Brian Ripley
    

    

      
      #
    

  


  

      
On Tue, 17 Jun 2003, Don MacQueen wrote:

        

            

      
      
      
    

        
pdflatex is just a symbolic link!  You do need a pretty full latex 
installation.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      stefano iacus
    

    

      
      #
    

  


  

      
On Marted?, giu 17, 2003, at 20:13 Europe/Rome, Don MacQueen wrote:

        

            

      
      
      
    

        
again you will be free to use browser="open" in browseURL but there are 
nice APIs to show HTML and PDF pages without the need external browsers.

To be more explicit: the RAqua I have built it is built to run under 
X11 as well, just execute it in an xterm window without specifying 
--gui-aqua. Nothing has changed!

stefano

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Jan de Leeuw
    

    

      
      #
    

  


  

      
Well, what happens if you say library(tcltk) and demo(tkdensity) in
the R running in Xterm ?

        
On Tuesday, Jun 17, 2003, at 11:25 US/Pacific, Stefano Iacus wrote:

        

            

      
      
      
    

        
===
Jan de Leeuw; Professor and Chair, UCLA Department of Statistics;
Editor: Journal of Multivariate Analysis, Journal of Statistical  
Software
US mail: 9432 Boelter Hall, Box 951554, Los Angeles, CA 90095-1554
phone (310)-825-9550;  fax (310)-206-5658;  email: deleeuw@stat.ucla.edu
homepage: http://gifi.stat.ucla.edu
   
------------------------------------------------------------------------ 
-------------------------
           No matter where you go, there you are. --- Buckaroo Banzai
                    http://gifi.stat.ucla.edu/sounds/nomatter.au

  
  


  


      

    

    

      
      

        


  

        

      stefano iacus
    

    

      
      #
    

  


  

      
On Marted?, giu 17, 2003, at 20:31 Europe/Rome, Jan de Leeuw wrote:

        

            

      
      
      
    

        
well, you already know the answer. This version of RAqua is built  
against AquaTclTk, so it won't work. But the x11() device will work  
fine. In this case two different versions of tcltk are competing X11  
and AquaTclTk.

But what I meant is that the X11 (itself) part works fine. One can  
still configure Darwin R to use X11/TclTk.

I need to stress the fact that we are talking about two different  
aspects: distributed binaries and building options.

1. If the user downloads and installs RAqua is not because he wants to  
use X11 stuff inside RAqua! He probably doesn't know what X11 standa  
for!
2. If you, me, Don etc, want to use all the X11 stuff, using fink  
instead of other sources of prebuilt dlls, this is still possible...and  
surely none of us will try to mix things (aqua/x11)

What I don't understand from this thread is: is Don asking binaries for  
darwin/x11 or just asking for the possibility of continuing building  
Darwin R on his own and configure R in his preferred way? I think he is  
asking for the second alternative: in this case, this is still  
possible. Just don't configure with --with-aqua (who oblige us?) and  
specify X11/TclTk headers at config time, run it in xterm. Moreover,  
RAqua is built using a script inside src/modules/aqua which is not part  
of the make R procedure. So if you execute it is because you do want  
Aqua and not X11.

We can consider the possibility of distribute two versions of binaries:  
"a' la" "RAqua" and "a' la" Jan :)

stefano

            

      
      
      
    

        

      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Don MacQueen
    

    

      
      #
    

  


  

      
At 8:59 PM +0200 6/17/03, Stefano Iacus wrote:

            

      
      
      
    

        
It's not so much that I'm asking for a certain version; rather, I 
want to understand what direction things are going.

I do want to be able to build from source code and work in an X 
windows environment.

I do think having a binary distribution for the Aqua environment is a 
very valuable contribution. Even if I don't use it myself, I can use 
it to promote the use of R among my co-workers.

I'm not asking for binaries for darwin/x11.

When I build from source code, I don't mind reducing, or even 
eliminating, dependencies on fink, provided that the alternatives are 
not much more difficult to use. Despite the problems with fink that 
Jan describes, it is still an easy-to-use *single* source for all the 
additional libraries that R needs.

Does that help?

As always, I am very, very, happy that R is available for the Mac, 
and appreciate your work, and Jan's work, and that of everyone else 
who has contributed to bringing R to the Mac.

            

      
      
      
    

        
-Don