[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Tue, 12 Dec 2000 22:52:35 -0600 (CST)
On Tue, 12 Dec 2000, Andy Adler wrote:
> I was thinking about this recently, and I wondered if it
> would be possible to use the approach used in the Perl
> module Win32::API
> This allows you to load and call arbirary functions
> in windows dlls from perl. I looked at the source, and it doesn't
> seem to be that hairy, although it has some assember in it.
It does do the job if you have a DLL on Windows and know what
function to call, and it's a real neat way doing that. However,
you still need a "well-formed" DLL so that you can call the
functions, and moreover it has no way of dealing with global
variables that are exported from a DLL. I do remember looking at
this way back when it first surfaced, and thought about using it
for another purpose.
The problem with dynamically linked Octave functions is a bit
different, and IMO the simplest approach is to add the proper
macros to the code to export/import the global variables and do
it right just once. We definitely want uniformity in how you
build .oct files on various platforms.
It's really not a hard problem, just an annoying one, and one
that will take someone who really needs it make it go. I needed
a few .oct files, and I decided that it was simply easier to
deal with *huge* oct files than finishing what I had started
(which is now lost unfortunately). I only used Octave/win32 when
my previous laptop didn't do Linux very well, but that problem
went away a long time ago, and so did my motivation for spending
much time on this issue. Many issues, not much time as usual.
If someone is interested, I can provide some insight on what
needs to be done. Here's the basic scenario:
Two possible ways of building DLL version of Octave runtime:
1. One massive DLL -- this is much easier for reasons that
becomes apparent once your undertake this effort.
2. Keep Octave's current library structure, and then build
individual DLLs. This is tricky, since you now have both
multiple dependent DLLs.
If you choose , then the following steps at the very minimum:
1. Code inspection to figure out what, if any, Octave global
variables need to be exported. If none, life gets much easier.
If any, then those need to be encased in various declspec
macros to export/import properly.
2. Write an export definition file. Easy.
3. Create the final DLL. Easy.
If you build one large DLL, then there are some issues you need
to deal with in Octave-like code-bases, and the primary one is
that you can't have multiple definitions (that's ok if you split
things in multiple archives, and the linker takes care of it). I
don't recall if I ran into this case or not.
Hmmm ... now it seems rather simple, and I wonder what kind of
problems I ran into when I had tried it.
I remember bringing up the issue of libtool'izing Octave, and I
don't believe JWE had the time for it. The advantage there is
that newer releases of libtool can create the DLLs and import
libraries without adding lots of extra stuff to Octave's config
I go on an extended trip in 3 days, and I'll have plenty of quality
"air time" (with my luck, most of that will be in airports), and
perhaps I'll take Octave with me and see where I had left of. No
promises, and certainly no guarantees. I may simply forget all
about it by the time I leave for the airport.
To be quite honest, if I had the time to work on Octave, I'd probably
be doing it where I really think Octave needs it -- cross platform
visualization. When I'm careful, most, if not all, of my code is
portable between matlab and octave, but I still have to boot back
into Windows when on the road to use Matlab to look at my results.
Stinks. I see that a KDE project (Kmatplot???) works with Octave,
and it does look quite nice. Too bad it's using Qt, which is non-free
on Windows. Enough rant on a topic that is irrelevant in this thread.
Octave is freely available under the terms of the GNU GPL.
Octave's home on the web: http://www.octave.org
How to fund new projects: http://www.octave.org/funding.html
Subscription information: http://www.octave.org/archive.html
- mkoctfile, Giancarlo Mascetti Ph.D., 2000/12/12