Skip to content

R-1.4.1 scheduled for Jan 30

5 messages · Peter Dalgaard, Brian Ripley, Dirk Eddelbuettel

Original
Peter Dalgaard
# 
We plan to put out R-1.4.1 on Wednesday. This is mainly to get the
fixes that we have already made into an official version, and to clean
up the problem with the Windows build. 

Several on the core team are under fairly heavy load at this time, so
we'll probably be reluctant to fix further bugs unless they are
clearly serious and/or trivial to fix.

Those of you that have unusual platforms could be doing us a favour if
you try building snapshot versions and report back if there are
problems.

        -p
Dirk Eddelbuettel
#
On Thu, Jan 24, 2002 at 01:32:31PM +0100, Peter Dalgaard BSA wrote:
I could make a snapshot Debian release which would then be thrown at the
twelve distinct architectures Debian supports. On the other hand, 1.4.0
already builds on 9 of these as shown on

   http://buildd.debian.org/build.php?pkg=r-base
   
As I presume no fixes were made to correct the mips and mipsel failures,
would such a snapshot release really provide incremental information?

Dirk
Brian Ripley
#
On Thu, 24 Jan 2002, Dirk Eddelbuettel wrote:

            

      
      
      
    

        
Some, yes.  Sometimes fixes for one platform break another.  As a real
example, Rinternals.h was causing two C++-based compilers to object.
I changed a couple of declarations to a form that Metrowerks was known to
accept, and it passed on four others.  Then Luke found HP-UX's C compiler
objected, and had to change it again.

I really don't know how much independent evidence Debian's different
architectures provides, but my guess is that no else will be able to check
a 64-bit Linux.

Does that inform a cost-benefit analysis for you?

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Dirk Eddelbuettel
    

    

      
      #
    

  


  

      
On Thu, Jan 24, 2002 at 03:35:39PM +0000, Prof Brian D Ripley wrote:

            

      
      
      
    

        
[...]

Good point. On the other hand, all this will show is gcc et al on different
Linux arches.

            

      
      
      
    

        
Sure. But one other problem I forgot earlier is that the timeframe might
simply be too tight -- it could be Sunday before most of the systems catch
up, and that would leave little time for R-core to try to mend things.

I could try to do it this eve if there is a view at your end that it is
useful. I guess it wouldn't hurt, overall. 

Dirk

  
    

      
      
    

  


  


      

    

    
2 days later

    

      
      

        


  

        

      Dirk Eddelbuettel
    

    

      
      #
    

  


  

      
"Brian" == Brian D Ripley <Prof> writes:
  Brian> I really don't know how much independent evidence Debian's different
  Brian> architectures provides, but my guess is that no else will be able to
  Brian> check a 64-bit Linux.

As an interim result, R-patched-2002-01-24 compiled fine on the i386, hppa,
ia64, powerpc. sparc, arm. It failed (just as previously) on mipsel (see
below). The m68k box was out of diskspace and postponed, others haven't tried
yet.

Full logs at http://buildd.debian.org/build.php?&pkg=r-base

The mipsel failure is in:

running code in `arith-true.R' ... OK
comparing `arith-true.Rout' to `./arith-true.Rout.save' ...161c161
< [1] FALSE
---

            

      
      
      
    

        
190,193d189
< 
< n= 1006 : quantile(rErr, c(.95,1)) = 0.002086 0.0495 
< 
< n= 1039 : quantile(rErr, c(.95,1)) = 7.645e-14 3.195e-07 
198c194
< [1] FALSE
---

            

      
      
      
    

        
make[4]: *** [arith-true.Rout] Error 1
make[4]: Leaving directory `/build/buildd/r-base-1.4.0.cvs20020124/tests'

If anybody here is motivated to look into this in detail, let me know.

Dirk