Re: Windows binaries

[Philip Nienhuis]

"John W. Eaton" <address@hidden> wrote:
>
> What is the current state of Octave binary distributions for Octave?
>  
>  There is a link on www.octave.org/download.html to 
>  http://sourceforge.net/projects/matlinks, but I think the binary that
>  is available there is not current.
>  
>  The FAQ for Octave on Windows systems mentions several other
>  possibilities depending on whether you already have Cygwin installed.

Speaking of it, the current state of the Octave for Windows FAQ is also
a bit outdated.
I am busy updating it, amongst other things waiting for some input from
other developers.
Besides, the FAQ needs a bit of reorganizing too, its present state (one
big pile of text) is too long, it should be split into pieces cf. the
average Linux-HOWTO. I lack time to do that, I currently only have time
for updating parts of it. I'd happily transfer it -partly or as a whole-
to other volunteers.

Back to the subject:
I do agree with Andy Adler that the "complete binaries" on Sourceforge
are currently the best options; IMO the ones from Matlinks should be
deprecated, let alone because they are older versions (2.1.3x rather
than 2.1.42 on SF).

>  I would like to see all of these efforts merged in some way so that we
>  can point to one place and have a simple set of instructions (i.e.,
>  get this file, run it, click next, next, next, finish, possibly
>  selecting some options along the way, and then you are done).

Right. But a problem is that the Windows platform is by no means
"monolithic" - there are considerable differences between all successive
versions (and sub-versions) from 95 to XP. Language differences
("localizations") only add to the confusion. All in all, the "just a few
clicks" solution might only help the average low-level Windows/Octave
user in the sense of providing just a basic Octave installation; I think
more advanced stuff like .oct files etc. had better go into add-on
packages and may even need to be restricted to the true
multithreaded/-tasking/-user versions of Windows, i.e. NT or higher.

>  It seems that the most useful way to install Octave on a Windows
>  system is with a set of the necessary Cygwin tools (perhaps the
>  compiler and other bits could be optional if you don't care about
>  building .oct files).  It would be nice to be able to automatically
>  detect whether Cygwin needs to be installed or upgraded, without
>  wiping out an existing Cygwin install.  I've received several
>  complaints about that, and although I realize that the people who
>  complained probably did not follow directions, but they are usually
>  unhappy and point fingers in our direction.

Andy Adler's 2.1.42 packages on SourceForge cover many if not all of
these aspects.
  
>  If you have been making a binary distribution but no longer have time
>  or any interest in doing it, can you please share a fairly precise set
>  of directions for building the binary distribution so someone else
>  might be able to take over where you left off?
>  
>  Should the Octave binary distribution include everything needed, or
>  should there be a set of packages (octave, gnuplot, etc.)?

A good point. For example, Andy included just gnuplot binaries + on-line
help in his packages, but IMO gnuplot is by itself a useful program.
Dedicated gnuplot Windows packages also include demos etc. which may
help people to explore gnuplot which in turn may help them in getting
more out of the combination of gnuplot & Octave. Bot OTOH providing an
Octave installer including a basic working gnuplot graphical output
solution in /bin _is_ very useful.
I think this illuminates that some discussion on what to include or not,
and in what form, may be needed.

>  How should Cygwin be handled?  Should we include the necessary parts
>  with the Octave distribution, or should it be a separate package?
>  Would telling people to download Cygwin from the Cygwin site be
>  enough?  If so, then would it be worth building a Cygwin package so
>  that we could just point people to the Cygwin site and tell them to
>  download and install Cygwin, then install this package on top of that?
>  I'm not sure whether this is a simple enough solution.  We will still
>  need to point them to gnuplot binaries in that case, unless our Cygwin
>  package includes it (which seems like a bad plan).  Would most Windows
>  users actually do all of this, or would they think it is too many
>  complicated steps and give up?

Comparing such an installation scenario with almost all other Windows
packages, I think the answer is clear: the average (and even the "a
little above-average") Windows user expects a "just a few clicks"
solution, optionally followed by installation of "just some more clicks"
add-ons.
I do not intend to provoke, but just compare such a "first Cygwin
installation, then Octave installation" with Matlab's installation
procedure.
Again, I would prefer the "complete binaries" on Sourceforge.

But the points raised by JWE above imply yet other problems to be faced:
how to enable Windows users to safely *update* their Octave installation
under Windows w/o forcing them to learn too much Linux-like concepts
like tar, gunzip, etc., and at the same time leaving their Cygwin base
stuff intact. Or shouldn't Cygwin be updated regularly to allow for more
'modern' Octave options which demand more recent Cygwin binaries? I fear
something like the RedHat/Mandrake-like .rpm hell I was once caught in,
i.e. where a naive intention of updating just one package may lead to
obligatory updates of one, a few, or even a whole chain of other related
packages.
On the way, care should be taken to not silently break other Cygwin
programs depending on specific Cygwin versions.
Furthermore, I guess that "complete Windows binaries" should at least
include tar and g(un)zip/b(un)zip2 to save unwary Windows users
installation of additional Cygwin parts using quite non-Windows like
procedures.

In the new Octave-Windows FAQ (to be uploaded soon I hope) I already
included some info on updating Octave/Windows installations, but it is
still based on Linux (Cygwin) procedures.

Again and again, the underlying question boils down to this: how far do
Windows users have to be shielded from Linux/*nix concepts?

>  I have recently updated the directions for installing Octave from
>  source on Windows systems and you can find that in the file
>  README.Windows in the latest source distribution (or by browsing the
>  CVS archive at www.octave.org if you just want the one file).  This
>  may be of some help, but I suspect most people who are building the
>  binary distributions for Windows are already familiar with all those
>  details.

Does that replace the identically-named file on
ftp://ftp.octave.org/pub/octave/ (there lives an older README.Windows
version dated Aug. 01)?
I'll include a link to it on a new version of the Octave-Windows FAQ,
hopefully to be uploaded this month. I'd prefer to link to
ftp://ftp.octave.org/pub/octave/ rather than the CVS.

Philip Nienhuis