On Sat, Feb 23, 2019 at 09:06:06 +0000, CROZIER Richard wrote:
> As someone else has pointed out though, actually for me the problem is
> that even the latest OS, e.g. Ubuntu 18.04 has an out of date Octave
> package in the repositories. For me, I routinely build Octave from
> source,
Is there anything we here can do as a community to help counter this
belief that Octave from Ubuntu LTS is "out of date"? It's a relative
term. Compared to Octave 3.2, Octave 4.2 is an amazing improvement.
This whole conversation is, perhaps for the first time in this community, giving me the feeling that I have an advantage by being a windows user. The standard windows model is ‘go get that thing and install it. With few exceptions, everything the program needs to run is self contained.’ Yes, i recognize that that puts more burden on you developers. But installing an updated binary (think Firefox, libreoffice, etc) is just a necessary process and sometimes automated.
If I have a system with Ubuntu 16.04, which is old but still supported,
I will get Octave 4.0. If I just want Octave, I probably don't care that
it's 4.0, so it's not "out of date" to me, it works perfectly fine.
Yet when users on help list ask anything, a frequent response is ‘you’re using an out of date version can you try a more recent one, that was fixed in 4.x’ With windows I can just tell them to go download the more recent binary. On Linux does that mean ‘go build from source’? How easy is it for someone to use a different package repo to get a newer version?
I
will also get GNOME 3.18 and LibreOffice 5, which also work perfectly
fine even though they are not the bleeding edge latest release.
But we’re a comparison/replacement product. Libreoffice 5 stands alone as a stable basic product. For some people ‘basic’ usage of octave requires some 2010 matlab compatibility that just isn’t implemented yet. So ‘out of date’ means very different things to different users. A lot was implemented between octave 4 and 5 that could render the product useless for some users.
I had mentioned a while back that it might be worth generating a comparison function list with past matlab versions and encouraging old missing function implementation so we could verify ‘99% function compatible with matlab 2006’, or similar. We just got a host of newbie users posting homework style questions in help, and at least one ranted about ‘omg this sux none of the functions I want are implemented’. If it’s not obvious how a user can check/expect some level of compatibility, and it’s not trivial to upgrade to the newest version to get a recently implemented function, and the only other way is to compile from source, new user retention will be the worse for it.
> but if I want people to use stuff which uses the latest(ish)
> features, they have to build Octave too. Remember this is *hard* for
> ordinary users who don't even know about make for example.
I agree, ordinary users should absolutely not be building Octave from
source, unless they want to.
In windows, I can at least always be sure I’m getting the latest release, since binary release is the only distro model. If I really really needed some dev function, I’d have to manually implement or crosscompile. But that delta (in time) is typically small. Thanks to you fine folks I never have “we’re up to 5.1 but we haven’t built a win binary since 4.0” problems. When 5.1 comes out, 5.1 is quickly there to download.
Users should absolutely not be downloading some binary tarball from a
web site to install Octave (or any other application).
Again, that’s the lovely (virus ridden) windows model, and lots of people are familiar with that model even if they usually work on linux. I don’t think people would object that heavily even if maybe they “shouldn’t” do it that way.
I’m still thinking the OS agnostic virtual appliance model could be worthwhile.