Installation Problems (GNU Octave (version 7.3.0))

E.4 Installation Problems

This section contains a list of problems (and some apparent problems that don’t really mean anything is wrong) that may show up during installation of Octave.

On some SCO systems, info fails to compile if HAVE_TERMIOS_H is defined in config.h. Simply removing the definition from info/config.h should allow it to compile.
If configure finds dlopen, dlsym, dlclose, and dlerror, but not the header file dlfcn.h, you need to find the source for the header file and install it in the directory usr/include. This is reportedly a problem with Slackware 3.1. For Linux/GNU systems, the source for dlfcn.h is in the ldso package.
Building .oct files doesn’t work.
You should probably have a shared version of libstdc++. A patch is needed to build shared versions of version 2.7.2 of libstdc++ on the HP-PA architecture. You can find the patch at ftp://ftp.cygnus.com/pub/g++/libg++-2.7.2-hppa-gcc-fix.
On some DEC alpha systems there may be a problem with the libdxml library, resulting in floating point errors and/or segmentation faults in the linear algebra routines called by Octave. If you encounter such problems, then you should modify the configure script so that SPECIAL_MATH_LIB is not set to -ldxml.
On FreeBSD systems Octave may hang while initializing some internal constants. The fix appears to be to use
```
options      GPL_MATH_EMULATE
```
rather than
```
options      MATH_EMULATE
```
in the kernel configuration files (typically found in the directory /sys/i386/conf). After making this change, you’ll need to rebuild the kernel, install it, and reboot.
If you encounter errors like
```
passing `void (*)()' as argument 2 of
  `octave_set_signal_handler(int, void (*)(int))'
```
or
```
warning: ANSI C++ prohibits conversion from `(int)'
         to `(…)'
```
while compiling sighandlers.cc, you may need to edit some files in the gcc include subdirectory to add proper prototypes for functions there. For example, Ultrix 4.2 needs proper declarations for the signal function and the SIG_IGN macro in the file signal.h.

On some systems the SIG_IGN macro is defined to be something like this:
```
#define  SIG_IGN  (void (*)())1
```
when it should really be something like:
```
#define  SIG_IGN  (void (*)(int))1
```
to match the prototype declaration for the signal function. This change should also be made for the SIG_DFL and SIG_ERR symbols. It may be necessary to change the definitions in sys/signal.h as well.

The gcc fixincludes and fixproto scripts should probably fix these problems when gcc installs its modified set of header files, but I don’t think that’s been done yet.

You should not change the files in /usr/include. You can find the gcc include directory tree by running the command
```
gcc -print-libgcc-file-name
```
The directory of gcc include files normally begins in the same directory that contains the file libgcc.a.

Some of the Fortran subroutines may fail to compile with older versions of the Sun Fortran compiler. If you get errors like

zgemm.f:
        zgemm:
warning: unexpected parent of complex expression subtree
zgemm.f, line 245: warning: unexpected parent of complex
  expression subtree
warning: unexpected parent of complex expression subtree
zgemm.f, line 304: warning: unexpected parent of complex
  expression subtree
warning: unexpected parent of complex expression subtree
zgemm.f, line 327: warning: unexpected parent of complex
  expression subtree
pcc_binval: missing IR_CONV in complex op
make[2]: *** [zgemm.o] Error 1

when compiling the Fortran subroutines in the liboctave/external subdirectory, you should either upgrade your compiler or try compiling with optimization turned off.

On NeXT systems, if you get errors like this:

/usr/tmp/cc007458.s:unknown:Undefined local
      symbol LBB7656
/usr/tmp/cc007458.s:unknown:Undefined local
      symbol LBE7656

when compiling Array.cc and Matrix.cc, try recompiling these files without -g.

Some people have reported that calls to system() and the pager do not work on SunOS systems. This is apparently due to having G_HAVE_SYS_WAIT defined to be 0 instead of 1 when compiling libg++.
On systems where the reference BLAS library is used the following matrix-by-vector multiplication incorrectly handles NaN values of the form NaN * 0.
```
[NaN, 1; 0, 0] * [0; 1]
⇒
[ 1
  0 ]

correct result ⇒
[ NaN
  0   ]
```
Install a different BLAS library such as OpenBLAS or ATLAS to correct this issue.
On NeXT systems, linking to libsys_s.a may fail to resolve the following functions
```
_tcgetattr
_tcsetattr
_tcflow
```
which are part of libposix.a. Unfortunately, linking Octave with -posix results in the following undefined symbols.
```
.destructors_used
.constructors_used
_objc_msgSend
_NXGetDefaultValue
_NXRegisterDefaults
.objc_class_name_NXStringTable
.objc_class_name_NXBundle
```
One kluge around this problem is to extract termios.o from libposix.a, put it in Octave’s src directory, and add it to the list of files to link together in the makefile. Suggestions for better ways to solve this problem are welcome!
If Octave crashes immediately with a floating point exception, it is likely that it is failing to initialize the IEEE floating point values for infinity and NaN.
If your system actually does support IEEE arithmetic, you should be able to fix this problem by modifying the function octave_ieee_init in the file lo-ieee.cc to correctly initialize Octave’s internal infinity and NaN variables.

If your system does not support IEEE arithmetic but Octave’s configure script incorrectly determined that it does, you can work around the problem by editing the file config.h to not define HAVE_ISINF, HAVE_FINITE, and HAVE_ISNAN.

In any case, please report this as a bug since it might be possible to modify Octave’s configuration script to automatically determine the proper thing to do.
If Octave is unable to find a header file because it is installed in a location that is not normally searched by the compiler, you can add the directory to the include search path by specifying (for example) CPPFLAGS=-I/some/nonstandard/directory as an argument to configure. Other variables that can be specified this way are CFLAGS, CXXFLAGS, FFLAGS, and LDFLAGS. Passing them as options to the configure script also records them in the config.status file. By default, CPPFLAGS and LDFLAGS are empty, CFLAGS and CXXFLAGS are set to "-g -O2" and FFLAGS is set to "-O".