[R] R on Tru64 OSF 5.1

[Ray Brownrigg]

> From: =?iso-8859-1?q?J=20Swinton?= <js229 at yahoo.com>
> 
> Has anyone achieved a current successful build of R
> 1.8 or 1.9 or earlier for HPUX Tru 64 OSF 5.1?
> 
> There is no binary version of R for HPUX Tru 64 OSF
> 5.1. The R admin manual mentions that the native make
> fails on "Alpha/OSF (aka Tru 64)" and gnu make must be
> used instead. There are problems reported with
> building R versions around 1.4 from 2002 on R-devel,
> some of which were fixed with using gmake instead of
> make, some of which were fixed by changes to R
> sources, but it is unclear if current R compiles on
> the current OSF 5.1, which is what I need to know
> before trying to get our local IS admins to try again
> after their first three failures.
> 
> Jonathan
> 
It depends what you mean by "successful build".  I succeeded in getting
R-1.8.1 to "./configure" and "make" on a system:
Compaq Tru64 UNIX V5.1A (Rev. 1885); Wed Nov 20 00:37:24 NZDT 2002

However it does not pass  "make check", with the following message:
496306:/vol/R/src/osf1/R-1.8.1/bin/R.bin: /sbin/loader: Fatal Error: call to unresolved symbol from /vol/R/src/osf1/R-1.8.1/modules/lapack.so (pc=0x3ffbfde3308)

during reg-tests-1.

This is possibly a consequence of the following (Info) warnings issued
at make time:

f77 -fpe3   -g -c dlapack0.f -o dlapack0.lo
f90: Info: dlapack0.f, line 1721: This directive is not supported in this platform.
CDEC$          NOVECTOR
---------------^
f77 -fpe3   -g -c dlapack1.f -o dlapack1.lo
f77 -fpe3   -g -c dlapack2.f -o dlapack2.lo
f77 -fpe3   -g -c dlapack3.f -o dlapack3.lo
f90: Info: dlapack3.f, line 21481: This directive is not supported in this platform.
CDEC$          NOVECTOR
---------------^
f90: Info: dlapack3.f, line 21495: This directive is not supported in this platform.
CDEC$             NOVECTOR
------------------^
f90: Info: dlapack3.f, line 21522: This directive is not supported in this platform.
CDEC$          NOVECTOR
---------------^

The weird thing is that the problem occurs during the following code
fragment:

## predict.arima0 needed a matrix newxreg: Roger Koenker, 2001-09-27
u <- rnorm(120)
s <- 1:120
y <- 0.3*s + 5*filter(u, c(.95,-.1), "recursive", init=rnorm(2))
fit0 <- arima0(y,order=c(2,0,0), xreg=s)

which runs perfectly OK if pasted into an interactive session.

I did spend some time trying to isolate/reproduce the problem without
success.  Unfortunately, this is the only Alpha/OSF1 system we have and
it is not used much (particularly not for R), so the installation has
not been well tested for 'real use'.

Let me know if you want the necessary configuration changes.

Hope this helps,
Ray Brownrigg <ray at mcs.vuw.ac.nz>	http://www.mcs.vuw.ac.nz/~ray