[Rd] make check fails 1.7.1 with R-release patch in regression tests (PR#3868)

[Peter Dalgaard BSA]

khendricks at ivey.uwo.ca writes:

> Built from R-1.7.1 source with R-release.diff patch attached on a 
> RedHat 9 system for Linux x86-64 (dual Opteron system) runing in 64 bit mode.
> 
> Build completes fine.  But "make check" fails (R is killed because
> Out-Of-Memory)
> during reg-tests-1.R.   I have verified that the R-Release.diff patch has been
> applied 
> and the changes to the stopping critieria in that file have been made. 
> 
> My x86_64 machine has over 2 gig of main memory. 
> 
> Is the stopping cirtieria a function of the size of the machine registers in
> use? 
> 
> All other tests seem to pass with no problems.  I have reproduced this error
> condition
> with and without optimized "atlas" libs in the build and even with -O0 (to
> remove possiblility of compiler optimization error).
> 
> I am stumped on how to proceed.  Obviously the stopping criteria for some method
> is never being
> reached resulting in the out of memory condition.
> 
> Any hints on how to track this down further appreciated.  All altas tests run
> and pass with flying colors (if that helps).

Hmm. 

(1) Does it happen *without* the patches?

(2) The canonical way to investigate is to
 
        (a) figure out which file is being processed and rerun R with
        output to the terminal. I.e.

        export SRCDIR=../R/tests # must be set, value depends on build setup 
        bin/R --vanilla < $SRCDIR/reg-tests-1.R

        This avoids buffering and so the last few lines should tell
        you approximately where things go wrong.

        You might want to cut the file down a bit, but in this case it
        is probably not necessary.

        (b) Run R under the debugger 
        bin/R -d gdb
        run --vanilla  < $SRCDIR/reg-tests-1.R

        when it hangs, try hitting control-C and do a "bt" (backtrace)
        command in gdb. You can then poke around in the stack and see
        if you can spot the problem. Or get back to us with some new
        clues. 
-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard at biostat.ku.dk)             FAX: (+45) 35327907