[netCDF #AFH-985814]: netcdf configure script fails to allow -Bstatic

> > Howdy,
> >
> > During a recent build attempt of software that uses the netcdf
> > static libraries, we found that we were unable to build the software
> > using the -Bstatic flag.  Further investigation seems to point to the
> > configure script of the netcdf software finding false positive
> > detections of errors when using the -Bstatic flag in the compile step.
> > It appears that standard license manager output (like that seen below)
> > is being dumped into the test file std err files (conftest.err) and is
> > disallowing the use of the -Bstatic flag in the building/linking of the
> > netcdf software.
> >
> > ------------- License Manager Std out -----------------------
> > pgf90 6.2-5 64-bit target on x86-64 Linux
> > Copyright 1989-2000, The Portland Group, Inc.  All Rights Reserved.
> > Copyright 2000-2006, STMicroelectronics, Inc.  All Rights Reserved.
> > yes
> > checking if Fortran "doubleprecision" is C "double"...
> > pgcc 6.2-5 64-bit target on x86-64 Linux
> > Copyright 1989-2000, The Portland Group, Inc.  All Rights Reserved.
> > Copyright 2000-2006, STMicroelectronics, Inc.  All Rights Reserved.
> > --------------------------------------------------------------
> >
> > We bypassed this false positive occurrence by forcibly setting the
> > static flag variable in the configure file for each of the compilers:
> >
>
> Howdy Carter.
>
> I don't think the compiler output with the copyright is causing any problems.
Lots of compilers do that (but it is annoying to look at!)
>
> Have you tried building without the static flag set?
>
> Have you tried setting FCFLAGS and F77FLAGS to -Bstatic?
>
> Unfortunately our test machine with the portland group compiler has suffered
a tragic fate, and is no longer around. I have asked our sysadmin to restore
the portland group compilers on one of my test machines, and then I can try the
-Bstatic option myself.
>
> Ed

Howdy again Carter,

I have looked into the -Bstatic flag for portland group compilers. I don't have
a full understanding of it, but it turns off the use of shared libraries by the
program, and allows larger fortran arrays, as I understand it.

I have gotten the netCDF 3.6.x snapshot to build with portland group, using the
-Bstatic flag. The problem (on my machine) was that using that flag seems to
put the fortran compiler into 32-bit mode on this 32/64 machine. So I had to
use the -m32 option with gcc (i.e. by setting the CFLAGS).

The clue was in the config.log file:

configure:22961: gcc -c  -static conftest.c
conftest.c: In function 'sub_':
conftest.c:4: warning: incompatible implicit declaration of built-in function
'exit'
configure:22964: $? = 0
configure:22967: /opt/pgi/linux86/7.0-2/bin/pgf95 -Bstatic -c conftestf.f
conftestf.f:
configure:22970: $? = 0
configure:22973: /opt/pgi/linux86/7.0-2/bin/pgf95 -o conftest -Bstatic
conftestf.o conftest.o
/usr/bin/ld: warning: i386:x86-64 architecture of input file `conftest.o' is
incompatible with i386 output

As this shows, the configure script is compiling a C (conftest.c) and a fortran
file (conftestf.f) and linking them together. But it is finding that the
compiled C file is 64-bit and the compiled fortran is 32-bits.

I got it to work like this:
make distclean && CFLAGS=-m32 FC=/opt/pgi/linux86/7.0-2/bin/pgf95
FFLAGS=-Bstatic F90FLAGS=-Bstatic ./configure --disable-cxx && make check

The -m32 option tells the GNU C compiler to produce 32 bit code.

I have added this configuration to our nightly snapshot testing, to ensure that
we don't break it!

Thanks,

Ed


Ticket Details
===================
Ticket ID: AFH-985814
Department: Support netCDF
Priority: Normal
Status: Closed