Re: Building snapshot versions for Windows

[PhilipNienhuis]

Clemens Buchacher wrote
> Hi,
> 
> (Apologies for duplicates, I was not subscribed.)
> 
> I would like to use Octave on my Windows machine at work. Getting there
> turns out to be quite some effort, as described below. Please advise if
> I have taken a wrong turn somewhere.
> 
> Currently, the main thing keeping me from using Octave is this bug:
> 
>  https://savannah.gnu.org/bugs/?func=detailitem&item_id=37122
> 
> As John pointed out to me, even though Rik fixed the bug within a month
> of the original report, 9 months later there is no Windows binary
> available which actually contains the fix. Maybe that fix should have
> been applied to the stable branch?
> 
> However, even if it had made it into stable, there is quite a bit of
> delay until a new version is released and built for Windows. For my
> purposes, I think it would be better to use snapshot versions from the
> default branch, at least for testing.

A while ago a test snapshot based on MSVC was made by Michael Goffioul. See
the maintainers ML. AFAIK it it is still available. With a little luck it
was made after the fix you mention.

I regularly make Windows builds using MXE (every other day or so). It see no
problems putting up a binary snapshot somewhere, were it not that they are
indeed for testing only (that's what I do with them). Especially now that
developments go relatively rapidly.

My own feeling is that we'd better be careful with binary development
snapshots, especially for Windows, because I'm afraid that unstable versions
would proliferate rapidly and leave a wrong impression of Octave's intended
capabilities.
Formal release candidates would be better suited for wider distribution in
binary form.

Perhaps we need a watermark or so in the terminal stating "development
snapshot, for testing purposes only" in big capitals and sprinkled with
exclamation marks, before binary snapshots can be put up.


> I will have to test the classdef
> branch as well. Currently I am not allowing the use of classdef because
> Octave does not yet support it. But classdef is strongly preferred and
> it is getting difficult to argue for Octave compatibility while we
> cannot even use it due to the above...
> 
> So I am trying to do that. But there seem to be many different ways to
> install and build Octave for Windows. Some are described here [1],
> others I found somewhere in the source tree, but the instructions are
> confusing and do not inspire confidence in their accuracy. I see also
> the recent discussion of MXE builds, which seems promising. Is this the
> one I should be aiming for?
> 
> I managed to complete the build by applying the attached patch to
> mxe-octave. I had to upgrade to gcc 4.8.0, because gcc 4.7.2 does not
> build with texinfo 5.0, which is the current version installed by Arch
> Linux. Backporting the fix needed to 4.7.2 is a much larger patch than
> the one needed to get gcc 4.8.0 to build with mxe-octave. But gcc 4.8.0
> currently breaks C++ mex files [2], so you may want to wait until that
> bug is fixed.

Sorry I got lost here. Didn't you mix up MinGW and MXE?
MXE is for cross-compiling a Windows build in Linux; AFAIK it still invokes
gcc 4.7.2.
Once you have the required dependencies for MXE itself in place it downloads
and builds gcc 4.7.2 and all other Octave dependencies.


> Unfortunately, I can only test the Windows build at work, since I do not
> have any Windows machines at home. How do you guys test your Windows
> builds?

VirtualBox?

Philip



--
View this message in context: 
http://octave.1599824.n4.nabble.com/Building-snapshot-versions-for-Windows-tp4652646p4652649.html
Sent from the Octave - Maintainers mailing list archive at Nabble.com.