Skip to content

Similar error as (PR#6958)

4 messages · b-h@mevik.net, Peter Dalgaard, Brian Ripley

Original
b-h@mevik.net
# 
When compiling R 1.9.1 on a Linux system with the environment variable
LANG=3Dno_NO, I get a similar error as reported in PR#6958:

## FACTS:

$ locale
LANG=3Dno_NO
LC_CTYPE=3D"no_NO"
LC_NUMERIC=3D"no_NO"
LC_TIME=3D"no_NO"
LC_COLLATE=3D"no_NO"
LC_MONETARY=3D"no_NO"
LC_MESSAGES=3D"no_NO"
LC_PAPER=3D"no_NO"
LC_NAME=3D"no_NO"
LC_ADDRESS=3D"no_NO"
LC_TELEPHONE=3D"no_NO"
LC_MEASUREMENT=3D"no_NO"
LC_IDENTIFICATION=3D"no_NO"
LC_ALL=3D
$ cd /usr/src/R/R-1.9.1
$ ./configure
[...]
$ make
[...]
make[3]: Entering directory `/usr/src/R/R-1.9.1/src/library/stats4'
building package 'stats4'
mkdir -p -- ../../../library/stats4/R
mkdir -p -- ../../../library/stats4/man
make[4]: Entering directory `/usr/src/R/R-1.9.1/src/library/stats4'
dumping R code in package 'stats4'
Error in eval(expr, envir, enclos) : couldn't find function "setGeneric"
Execution halted

If unset LANG before running make, the error doesn't appear, and the
build passes make check.

I've tried this on a Debian 3.0 stable system and a Red Hat 9 system,
with identical results.

## HYPOTHESIS:
the problem is that LC_COLLATE (in effect) is no_NO.  This matters for
the line

          cat `ls $(srcdir)/R/*.R` >> $${f}; \

in ${R_HOME}/src/library/stats4/Makefile.

With LC_COLLATE=3Dno_NO, ls will list the files in the order BIC.R
mle.R  AAA.R, making the require(methods) in AAA.R come last instead
of first in $${f}.  (In Norwegian, Aa is an old (but still often used in
names) way of writing the letter =C5, and lately, ls(1) has become
"intelligent" enough to sort them together, at the end of the
alphabet.)

## IDEA/SUGGESTION/RAMBLING:

Instead of relying on getting LC_COLLATE correct all places where that
is needed, perhaps one could use something like 111.R instead of AAA.R
(what to use instead of zzz.R I don't know).  Or perhaps use a
mechanism not based on sorting to ensure a file is read first or last?


--=20
Bj=F8rn-Helge Mevik
Peter Dalgaard
# 
b-h@mevik.net writes:
Oh great... Same thing in da_DK actually, and for the same reason
(witness my surname).
The cynic will say that some locale will probably have the idea of
sorting numerals after letters. Anyways, the right way of fixing it is
to prefix all those ls-constructs with LC_COLLATE=C, which really
isn't harder to enforce than any naming convention.

BTW, we just had two weeks of alpha and beta releases in which to
find this sort of stuff, you know....

BTW2, what happened to your mail? Been a while since I saw
quoted-unreadable damage like that.
Brian Ripley
#
On 24 Jun 2004, Peter Dalgaard wrote:

            

      
      
      
    

        
Right, but hard to find.  This works on Windows: unfortunately Unix has a
separate Makefile for every standard package and they are not even in the
same style so my pattern-matching failed.  It seems I missed just one 
(some others are in stub packages which only have one file).

            

      
      
      
    

        
Ye...s!

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Peter Dalgaard
    

    

      
      #
    

  


  

      
Prof Brian Ripley <ripley@stats.ox.ac.uk> writes:

            

      
      
      
    

        
[I even might have found it myself, but I set LANG=C, LC_CTYPE=da_DK
in my shells because I otherwise go nuts from the semi-translated
system messages ("hiding the randomness frog", etc..)]