Re: Octave port to OpenVMS on a Dec Alpha / VAX

[John W. Eaton]

On 10/26/2017 11:06 AM, Przemek Klosowski wrote:
> On 10/25/2017 10:09 AM, Dietz, Guido wrote:
> > Are you aware of any working Octave port to OpenVMS on a Dec Alpha or 
> > a VAX? Where may I find that?
> I am not aware of one, but I worked on OpenVMS in the past and I am 
> pretty sure that I had Octave running, by bringing up GCC/G++ and 
> recompiling Octave. I don't remember any details because it was 20 years 
> ago, but I think it wasn't difficult, just tedious.

I discussed this a bit with Guido off the list.  It would be an 
interesting thing to do as a portability test if nothing else.  I do 
have some experience with VMS, but it's been at least a decade.

To summarize:

I think it would be best to start with the current sources.  We aim for 
portable C++11.  For any non-standard POSIX features, we use a 
well-defined set of wrappers around functions from gnulib.

If you had asked me about this six months ago, I probably would have 
thought it would be next to impossible, but recently I saw some posts on 
the gnulib mailing list about improving support for VMS systems:

http://lists.gnu.org/archive/cgi-bin/namazu.cgi?query=VMS&submit=Search%21&idxname=bug-gnulib&max=20&result=normal&sort=date%3Alate

So maybe this idea is not too crazy.

I'd begin by disabling all optional features.  This should happen 
automatically if the configure script can't find the necessary libraries 
to support optional features.  Then you can add optional dependencies as 
need them.

The build system requires some basic Unix tools to run, /bin/sh and 
utilities, plus a reasonably current version of GNU Make, Perl and 
possibly Python.  Is that all possible with VMS?  I'm sure there are 
other things I'm forgetting.

But then...

What compilers are available for VMS?  Is there a GCC port?  Or is there 
only VMS C, C++, and Fortran?

The only info I can find for VMS C++ compilers is this:

  http://h41379.www4.hpe.com/commercial/cplus/alpha_doc/index.html

and the document linked there has a date of 2007.  Since Octave requires 
C++11 features now, I am sure the compiler documented there will not be 
sufficient.

The most recent version of GCC for VMS that I could find was either 3.x 
or 4.0.x and those are both too old for building Octave.

Unless there is another more modern C++ compiler available, I don't 
think it will be possible to build the current version of Octave for VMS.

Building some older version of Octave from a dozen years ago might be 
possible, but will surely bring up its own set of porting issues.  Given 
how much Octave has changed in the last ten years or so, solving those 
problems will probably not be of much help to the current version of 
Octave.  That makes this idea much less interesting to me.

jwe