Re: Performance issues on Windows, suggests a MSVC build

[Philip Nienhuis]

John W. Eaton wrote:
> 
> On 22-Jun-2011, Ben Abbott wrote:
> 
> | I expect that Octave has more to gain from Windows and MacOS users
> | than there are liabilities,
> 
> Can you explain a bit more what you mean?  How do you see more users
> being helpful?  I'd say it depends on the users we attract.  Are they
> supporting us with funds or code or something, or do they just want
> support for free?  If the former, then I say great, but if the latter,
> then I don't see how we gain much.
> 

"Gain" can also come in the form of exposure. 
The more people know about Octave, the more chance that someone will start
contributing.



> | and am troubled by the decreased support for Windows
> 
> We have many Windows users, but not much in the way of Windows
> developers.  How can we change this?
>      <snip>
> Working on Windows systems makes me want to gag, ... <snip>
> 

I'm not sure if expressing this opinion so strongly will help attract more
Windows developers ;-)


But OK, serious mode:

The issues with building Octave under Windows are mostly about the
development prerequisites. 
Compared to Linux, there are several additional and BIG stumbling blocks in
getting a usable Octave-MingW binary (for that matter, some may apply to Mac
OSX as well):

1. Setting up an Octave-compatible MingW build environment. Hard enough, and
currently not very clearly documented to say the least. E.g., the
instructions at octave-forge suggest to put MingW stuff under the MSYS tree,
but the latest MingW installers seem to put MSYS under MingW.

2. Find, where needed patch, and build, all required dependencies on MingW
(blas, lapack, fltk, ...). Much, much work

( 3. Build Octave under MinGW. Maybe easy, maybe not )

4. Once Octave has been built successfully, making some sort of Windows
installable package out of it. Quite a different art.


On Linux, steps 1-2 are easy. Most if not all of the required compilers and
dependencies come with the Linux distro at hand by simply invoking apt-get /
urpmi / ...  One can download & start building Octave right away. 

But on Windows, I haven't even been able to just set up a usable MinGW build
environment (yet).
As I wrote in a previous posting, I tried to use Tatsuro's stuff in an
attempt to skip most of steps 1 & 2, but somewhere along those lines I got
stuck.


So IMO an answer to your question above: ".... not much in the way of
Windows developers.  How can we change this?" would be to:
(1) Facilitate easy setup of a build environment, if possible along the
lines of the current MingW installers.
    Given MingW's dynamics, standardizing on some snapshot and updating
annually or so might be handy
(2) Somehow share prebuilt dependencies (compiled libraries, patched include
stuff, etc). 
    Octave-forge?

Perhaps someone can even zip & upload his/her complete MingW build
environment + dependencies, including some instructions for path settings
etc.

Once there (i.e. having a fairly complete Octave development environment),
MingW would be more or less equivalent to Linux as regards building Octave. 
At that point, rather than getting lost trying to have the build system up
and running, people running Windows would be able to help fix bugs & improve
the Octave code (beyond scripts), IOW become octave contributor, and maybe
even octave developer.

As to ( 4): being able to prepare a Windows installer or so doesn't need to
be common knowledge. One or two developers who can do that would suffice.


Then:

John W. Eaton <jwe <at> octave.org> Wed Jun 22 16:06:28 CDT 2011 writes:
> I remain unconvinced that there are many people out there who are
> competent Windows programmers who are also interested in developing
> free numerical software.  My guess would be that most everyone who is
> interested in developing free software as a volunteer has abandoned
> Windows.  I'm OK with being wrong about this, but we already have a

You are wrong :-)
For example, I know quite a few competent developers who use Python or R,
even FORTRAN (all and happily on Windows), rather than Octave.
In fact, many Python proponents I know are as zealous anti-Octave as some
folks in this thread are anti-Windows.


> Windows binary that should be sufficient to get competent programmers
> started and it has been downloaded a fairly large number of times, yet
> I'm still waiting to see the Windows programmers contributing.


? 
No offense taken, but e.g., most of what I contributed came from Octave
under Windows (but is applicable to any OS). Windows is where I use Octave
the most often.
And I guess you overlooked octave-forge contributors? A number of them do
their development work on Windows as well.
But perhaps I've misunderstood what you wrote.

B.T.W. The 3.2.4 MingW binary is no real Octave building environment. 
Building octave-forge packages: yes. Building Octave itself: no. 


Windows bashing won't help much to attract more Windows users, let alone
contributors.
Many Octave users and potential Octave users are confined to Windows if only
because of choices made by their employer's / faculty's IT department. I'm
in that same situation.


Philip


--
View this message in context: 
http://octave.1599824.n4.nabble.com/Performance-issues-on-Windows-suggests-a-MSVC-build-tp3616648p3622956.html
Sent from the Octave - Maintainers mailing list archive at Nabble.com.