Build fails for R 1.6.2: [base-Ex.Rout] Error 1

I am trying to build 1.6.2 from source RPM's on RedHat 9, but encounter
a failure when running the base tests. An inspection of
base-Ex.Rout.fail reveals the following:

 >
 >  stopifnot(  is.na(var(1)),

+            !is.nan(var(1)))
Error: !is.nan(var(1)) is not TRUE
Execution halted

On Tue, 8 Apr 2003, Christopher Fonnesbeck wrote:

I am trying to build 1.6.2 from source RPM's on RedHat 9, but encounter
a failure when running the base tests. An inspection of
base-Ex.Rout.fail reveals the following:

Yes. The version of gcc in RedHat 9 will not distinguish NA and NaN at
least sometimes, when optimization is used.  The bug report for RedHat is
at
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=88174

A fix has been added to the development version of R.

Thomas Lumley <tlumley at u.washington.edu> writes:

On Tue, 8 Apr 2003, Christopher Fonnesbeck wrote:

I am trying to build 1.6.2 from source RPM's on RedHat 9, but encounter
a failure when running the base tests. An inspection of
base-Ex.Rout.fail reveals the following:

...

Yes. The version of gcc in RedHat 9 will not distinguish NA and NaN at
least sometimes, when optimization is used.  The bug report for RedHat is
at
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=88174

A fix has been added to the development version of R.

Well, let's call it a workaround. The fix is gcc 3.3.

...but it's a one-liner: Inside the 

static double R_ValueOfNA(void)

in src/main/arithmetic.c

declare x volatile, i.e.

static double R_ValueOfNA(void)
{
    /* The gcc shipping with RedHat 9 gets this wrong without
     * the volatile declaration. Thanks to Marc Schwartz. */
    volatile ieee_double x;
    x.word[hw] = 0x7ff00000;
    x.word[lw] = 1954;
    return x.value;
}

Peter Dalgaard BSA wrote:

Thomas Lumley <tlumley at u.washington.edu> writes:

On Tue, 8 Apr 2003, Christopher Fonnesbeck wrote:

I am trying to build 1.6.2 from source RPM's on RedHat 9, but encounter
a failure when running the base tests. An inspection of
base-Ex.Rout.fail reveals the following:

...

Yes. The version of gcc in RedHat 9 will not distinguish NA and NaN at
least sometimes, when optimization is used.  The bug report for 
RedHat is
at
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=88174

A fix has been added to the development version of R.

Well, let's call it a workaround. The fix is gcc 3.3.

...but it's a one-liner: Inside the
static double R_ValueOfNA(void)

in src/main/arithmetic.c

declare x volatile, i.e.

static double R_ValueOfNA(void)
{
    /* The gcc shipping with RedHat 9 gets this wrong without
     * the volatile declaration. Thanks to Marc Schwartz. */
    volatile ieee_double x;
    x.word[hw] = 0x7ff00000;
    x.word[lw] = 1954;
    return x.value;
}

I can verify that the above change does work for R 1.6.2 and passes make 
check under RH 9.

Chris, download the source tar file from CRAN and as Peter has pointed 
out, make the change in line 119 in arithmetic.c.  Then:

./configure
make
make check
make install

HTH,

Marc Schwartz