Re: Binary distributions (was: Re: Release goals for 3.6)

[Lukas Reichlin]

On 02.08.2011, at 19:08, John W. Eaton wrote:

> On  2-Aug-2011, Lukas Reichlin wrote:
> 
> | Although I am an Octave-Forge developer and not an Octave one, I'd
> | like to share a few thoughts on the future development of
> | Octave. IMHO fully functional and up-to-date binaries for Windows
> | and Macintosh are of utmost importance:
> 
> I think building binary packages is nice, but I don't know that it
> should be the responsibility of the Octave project to build and
> distribute them.
> 
> But if you want to take on this project, then I will do what I can to
> help.  First, I think you will have to find people who are interested
> in and capable of building binaries for Windows and OS X, and then
> coordinate their efforts so that the binaries can be distributed.

Unfortunately, I'm not (yet) capable of building sorted-out binaries for the 
Mac but I'm trying to help with MacPorts portfiles as good as I can. At least 
Julien Salort is able to build one for Mac OS X.

> Also, if you do build binaries and distribute them, what level of
> support are you willing to provide for free?  For example, when bugs
> are reported, will they be fixed?  I don't think the oct2mat problem
> was ever fixed with the 3.2.x binary distribution for Windows that is
> on Source Forge.  That kind of problem makes Octave look bad, and it
> seems like it would be easy to fix simply by dropping a single package
> from the binary installer.

I think that even binaries without support would be better than no binaries at 
all. Providing no binaries because there might be a problem doesn't make Octave 
look any better ;-) And such problems would arise too if the user compiles 
Octave from source on the affected platform.

> 
> | One may ask why it is important to attract new users: Some of them
> | may become developers one day! Think of the trophic pyramid. The
> | same is true for professional sports which is not possible without a
> | broad field of amateurs which encourages young talent.
> 
> We've discussed this a lot in the past and I remain unconvinced.
> 
> As far as the type of developers we currently need for Octave in order
> to distribute binaries, your argument seems to point to a circular
> dependency.  We need more people who are capable of building binaries
> on non-free systems, so to attract those potential developers, we have
> to build binaries for the non-free systems?  Wouldn't that mean that
> we will have already done the work that we are trying to find others
> to do?  We have had binaries available for Windows for several years
> now, along with fairly comprehensive directions and scripts (in Octave
> Forge SVN) and so far we don't seem to have even one volunteer who
> really appears to want to take on the job of building binaries for
> Windows.

I was thinking of new developers for Octave and/or Forge packages who implement 
new features, fix bugs or write documentation rather than building binaries. 
Suddenly someone pops up and creates a profiler, a GUI or a JIT compiler and 
this is IMHO more likely to happen if we have a broader base of users.


>>>> IMHO fully functional and up-to-date binaries for Windows and
>>>> Macintosh are of utmost importance:
>>> 
>>> Yes, of course, absolutely no debate, and we're having a lot of
>>> trouble with 3.4 to produce those. How can we avoid this for 3.6? It's
>>> a very important problem.
>>> 
>>> - Jordi G. H.
>> 
>> A few ideas:
>> - For psychological reasons, binaries are offered on the Octave page, not 
>> the Octave-Forge page.
> Windows and OS X are treated specially because there is no other
> reasonable channel for users to get binaries?  Or do we also
> distribute .deb and .rpm files?

Well, Windows and OS X are IMHO important because they don't have a package 
management system (by default) like the Linux flavours you mentioned above.

>> - A concerted effort (kind of binary sprint :-) to address all building 
>> problems.
>> - Once the build problems are solved, build binaries for each development 
>> snapshot such that new problems are recognized at an early stage.
> Who is building the binaries?

The binary managers? ;-)

>> - A release is not a release without binaries for all supported platforms. 
>> If Octave developers are not able to build binaries, who will be anyway?
> Is a release of GCC not a release because the GCC developers do not
> build binaries for all supported platforms?

This statement was intended as a proposal for a new way of thinking to avoid 
troubles for 3.6 that Jordi mentioned above. But yes, a fresh GCC release is 
pretty useless to me if I don't have already a working compiler to build it :-) 
Unless I want to do all the bootstrapping, I need a binary. This could be 
provided by some GCC guys, Apple, ...

> 
> What exactly is the list of all supported platforms?

That's up to the Octave community to decide. IMHO Windows, Mac and Linux should 
be on that list for sure.

>> - Some sort of designated binary manager, similar to the job of a release 
>> manager.
> Are you volunteering for this job?  :-)

Nope :-( But I wish I was capable of.

Lukas