[OctDev] Misc problems on OS X

[Per P. <per...@ma...>]

> Other problems I've encountered: DarwinPorts builds octave static 
> instead of
> dynamic.  Is this easy to change?

If you mean shared libs, then no. The only known way to build octave is 
static, but with dynamic loading of .oct files.
I spent some time trying to get shared libs working about a year ago, 
and made some progress, but never got it working properly.

>
> SuperLU #include's malloc.h, which doesn't exist.  I'm commenting it 
> out since
> malloc is supposed to be in stdlib.  If it breaks for any system 
> without it, we will
> have to check in configure for #include <malloc.h>.

malloc is in stdlib, malloc.h is in sys/malloc.h

>
> isinf() isn't defined for whatever reason.  I fixed it to use 
> lo_ieee_isinf.

This seems like a previous problem with isnan... (dig, dig...) Yes.
A buggy header file (cmath) causes this problem [1].
Anyhow, _isinf and _isnan exists and
extern "C" int isnan (double);
extern "C" int isinf (double);
may be used as a workaround.

They both live in libSystem:
[per ~] % nm -g /usr/lib/libSystem.B.dylib | grep isinf
          U _isinf
/usr/lib/libSystem.B.dylib(isinf.So):
900171a0 T _isinf
900a03c8 T ___isinfd
900a03ac T ___isinff
          U _isinf

I was pretty sure this had been fixed by now... What version of OSX and 
gcc are we talking about?

>
> gnuplot doesn't compile with either aquaterm or X11 by default on 10.3
> without the X11sdk.pkg.  Suggestions?

What version of gnuplot? Gnuplot 3.8j should be OK. (At least on 
10.2.8, can't speak for 10.3 yet.)
It used to build without X11, but I've been away from much of this the 
past year for several reasons and I'm not up-to-date on anything...

Why not install X11sdk.pkg?

> configure was doing an incorrect test for NArray support.  Now I check 
> for
> dim_vector, which I imagine was only introduced with N-Arrays.

???

/Per

[1] This is an excerpt from a mail from jwe (there were some subsequent 
problems, if you want more detail I can dig it up for you.)

| I've sucessfully compiled octave on OS X 10.2 with the following 4
| changes:
|
| 1) file "lo-ieee.h" added
| extern "C" int isnan (double);
| extern "C" int isinf (double);
| due to a broken <cmath>, the following testcase should detect it:
| AC_TRY_COMPILE([#include <cmath>], [isnan(1.0);],
| [AC_MSG_RESULT("Pass")], [AC_MSG_RESULT("Fail")])
| Just to check, I tried
| AC_TRY_COMPILE([#include <math.h>], [isnan(1.0);],
| [AC_MSG_RESULT("Pass")], [AC_MSG_RESULT("Fail")])
| which does pass as expected.

Since this apparently works if it is in a C source file, I've made
some changes that I think will fix the problem.  Now the isnan calls
that were in lo-ieee.cc are in a C source file instead.  If there are
still problems, then send me a new report.