Troubles with Hewlett-Packard

[uunet!MathWorks.Com!moler]

I'm very annoyed right now with Hewlett-Packard.  So I thought I would
vent my anger here, and maybe even get a little sympathy or help.

Let me say, as preamble, that I am generally a fan of Hewlett-Packard.
They make terrific hardware.  They make terrific calculators.
I love the "Dashboard" utility for Windows.  But I am not happy
with their math library and operating system for HP700 workstations.

Here at the MathWorks, we're trying to get a new release of MATLAB
out on several different computers at the same time.  We're working
now with Suns, PCs, Macs, HPs, IBM RS, and a couple of different
Vaxes.  We have a lot of experience with developing multiplatform
technical software.

The first thing that set me off today was the Fortran compiler and
math library. (We can link Fortran and C subprograms into MATLAB; we
call them "Mex files".)  Here is a test program that illustrates just
two of my complaints.

      real*8 y
      complex*16 z
c
c     The following should produce y = IEEE floating point infinity.
c     But this message is produced at compile time:
c
c     Warning on line 9 of xxx.f: divide by zero detected; zero result inserted
c                                                          ^^^^
      y = 1.d0/0.d0
      write(6,*) y
c
c     The following should NOT produce y = 0.
c     It appears that zabs(z) is being computed as
c        sqrt(real(z)**2 + imag(z)**2)
c     without any scaling to protect against under/overflow.
c
      z = dcmplx(3.0d-200,4.0d-200)
      y = zabs(z)
      write(6,*) y
c
c     The output is y = 0.0d0
c
      end

It's been several years since I taught a numerical analysis or
mathematical software class, but stunts like this would have been
worth D's in my classes.

The second thing that set me off today concerns versions of HP-UX.
We have two HP700s at MathWorks; one is running 9.01, the other is
running 9.03.  We just discovered that the MATLAB "save" command is
working on one, but not on the other.  We don't know why.  We'll find
out and fix it, but we sure wish we didn't have to worry about this
sort of think so often.

There are more items on the list, but that's enough for today.

  -- Cleve Moler
     Chairman and Chief Scientist
     The MathWorks
     moleramathworks.com