R-beta: R with gnuwin32

guido@sirio.stat.unipd.it guido@sirio.stat.unipd.it
Mon, 16 Mar 1998 11:30:20 +0100 (GMT+0100)

Peter Dalgaard BSA writes:
> .........
> 2) But it really belongs on R-devel, not R-help. R-help is for helping
> users with installing and using R, not debating its future or
> discussing technical issues. Could I ask you to do_subscribe and then we
> move the discussion there? (Don't worry, R-devel is not a very high
> volume list.)
  Ok. I am here.
> .......
> I tried essentially the same stunt, but couldn't figure out how to get
> the Windows console and graphics going, "Multiple Document Interface"
> and all that. In particular, there's trouble with the gnuwin header
> files versus the MS Platform SDK used in the sources. Does this mean
> that you have a version running just like the "official" one? And
> could it run under Win3 too? 
  (a) The gnuwin header miss some definition used by R. I hand copied
  them (they are long but not too much) from the header of
  a old Borland compiler (16 bit; so I also crossed my finger).
  I do not know if hand copy is correct but I mean "a copy made
  by my finger not using the cut and paste". Since now
  the header are freely available but non redistribuible I 
  suppose that this is legal. 
  (b) So I have a version of R that "works".
      "works": - I could run all the demos but the dynload one (I didn't
	       - I played with it for an   hour preparing the
	         slides for my lesson of this afternoon (poisson regression
	         + graph)
	       - I make use of the functions in the eda library (dynloaded).
	         and I tried some silly homemade dll. 
  (c) Running as the "official" one? No. The basic problem are
      with the R sources that load data/library...
      The system.win file, the one that I used from R-0.61.1, has
      a not working "library" (also but perhaps I make some mistake a 
      broken "data")
      For example to run the demos I used a silly ad hoc data function
      of the type
      data<-function(a) source(paste(basedata,a),sep="")
      On the contrary, some of the command under unix are implemented
      by shell commands. So the src/library/base/system.unix can't be
      Hence this part must be fixed. This is why, yesterday, I asked
      for documentation of what must go in  src/library/base/system.
      Perhaps, it was an inappropriate request since I can
      look to the unix one. But, you know. After, some hours
      I had the system running but I could not do "data(iris)"
      and with "no iris data there is no statistics".
  (d) About Win30. I don't know. 

> > It seems
> > also a little faster than the october binary version 
> > (perhaps since I compiled the fortran sources directly
> > using g77).
> I wouldn't be surprised, although Watcon does some register-passing
> tricks, which one would think made it faster, (as well as incompatible
> with other C compilers...!)
I just compare the time of
  for (i in 1:1000) a<-solve(matrix(rnorm(100),10,10))
matrix and the gnuwin32 version was 20% faster. 
I suppose that this is not a  fair test since
is biased towards the comparison of the fortran code (and I used g77).

> > (c) Is someone interested in this port? 
> Definitely!
  If you want, I can make the patch available in a couple of days.
  I just want to recompile from the scratch so that you have not
  to bother about the fact that I have forgotten a file o something
  like this. But I have now time today and tomorrow (teaching,
> How about the cross-compilers?
  I suppose that if you have gcc running somewhere and configured
  to support win32 as target you can cross-compile. 
  But, I have not tried.
> > I believe that is not appropriate to make a binary distribution available
> > before (i) the base/R/system.win file has been fixed (or at least
> > before I, or better, someone else understand, what is going);
> > (ii) more testing.
> Agreed, but binaries (of R *and* all the packages) will in a
> not-so-long perspective be needed for practical people. 
  Ok. But with a stable distribution.

> > (d) I tried to recompile the regex library in the src/regex
> Hmm. Does this line in src/regex/WHATSNEW ring a bell? (Not that I
> know what I'm talking about, I just use grep...)
> you have to build regex.h explicitly by "make h".  The two known bugs
  Thanks. So regex.h must be built. The line before the one
  that you quoted say that regex.h is generated by mkh.
  What kind of monster mkh is?

guido masarotto
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch