[geos-devel] Ready to switch from SVN to GIT ?

[Mateusz Loskot]

On 6 April 2017 at 20:08, Regina Obe <lr at pcorp.us> wrote:
> On Thu, Apr 06, 2017 at 06:41:27PM +0200, Mateusz Loskot wrote:
>>>
>>> Drone is rather an interesting addition, not something to rely on, from my POV.
>>> It does not support builds on Windows. I'm not even sure it supports
>>>  builds on OS X.
>>>  (I mean 'guest' OS, not host).
>
> Drone is just one option of many and we don't need to choose just one.

>From POV of someone who wants to fix a bug, submit a pull request, get
it built and tested within 5-10 mins, make it ready for a project
maintainers to review and merge - IOW, growing community of
small-scale/ad-hoc contributors, testers and troubleshooters, Drone
has zero value.

> From my POV, I'm a windows packager and I like to have the packages (test builds) built automatically with each run of PostGIS / pgRouting GEOS / eventually GDAL (though this I just build at the end except on debbie which tests each change).
>
> Github won't do this for me.

GitHub+AppVeyor will do it for you
https://www.appveyor.com/docs/packaging-artifacts/

> This is getting to why I need more control than what github provides and support the Jenkins buildbots.

https://wiki.jenkins-ci.org/display/JENKINS/GitHub+Plugin

>>> Tiny projects like GEOS, non-profits like OSGeo simply can not afford
>>> investment in developing its infrastructure at low-level.
>
>> I asked you in particular, and I'll keep asking each developer partecipating in a free software project community. It's something some people enjoy doing, because maybe they think they could use the
>>  investiment they make in the future (effectively using the software or just the skills built improving it).
>
> I think the problem to focus on here, is why we don't have more investment in OSGeo to support this and other things if so many companies are relying on our work.

Shortly, because OSGeo projects do not need it: GitHub+CIs work for them well.
The projects even don't bother to analyse their needs and consider
simple question: "What are pros/cons of OSGeo infrastructure, shall we
stay?".
We've been trying for, correct me if I'm wrong, years now, asking
question: "What you need, how can we make you happ(y|ier)?" and nobody
comes back with answers!
They simply move their processes outside, quietly.

OSGeo sponsors are needed to scale up on GitHub, Travis CI and
AppVeyor, negotiate discounts and pay for the deals.
If GitHub+CIs dies, then shift money to NKOTB. If none available, then
consider self-hosting. Before that happens, we are YAGNI'ing, wasting
efforts and money.

> Rather than accepting the status quo and saying we're too poor let's just use GitHub. Let's ask how can we be better at marketing and asking for money or infrastructure?
> We are too focused on doing software development on a shoe-string budget and not asking enough of WHY DO WE HAVE TO WORK THIS WAY?

It's not "poor? go to GitHub". I'm certain, OSGeo would have to start
paying, but I'm fairly sure the monies, even if bigger would deliver
better experience.
"If OSGeo offered you access to paid {GitHub|BitBucket}+{Travis
CI|BitBucket Pipelines}+AppVeyor , would it make your project
happier?" - Shall we poll OSGeo folks :-)

Another problem w/ maintaining OSGeo-hosted infrastructure is lack of
admins even remotely interested in getting a contract.
Year ago Sandro, if I'm correct, pioneered sysadmins contracting for
OSGeo (https://lists.osgeo.org/pipermail/sac/2016-May/007032.html).
Year later... I have expected there would be more (actually, a lot)
people within OSGeo community interested in hugging 5000 EUR for 50
hours job.
Sandro has done great job, he's also shown the value of sysadmins,
capacity of OSGeo budget in action, but we need Sandro x 5 at least.

Nobody has come forward really. Why? I think we know :-)

> As I already mentioned GitHub doesn't satisfy all my needs.  I'd like to for example test on other OS like FreeBSD, CentOS

Travis CI and Docker can do it for you.

> And I even have dreams of having pre-compiled builds for these people to increase the number of people testing things out for us.

Again, no problem (eg. https://docs.travis-ci.com/user/deployment/packagecloud/)
Those services are usually referred as CI, but they cover CD (cont.
depoyment/delivery) too.

>>> I'll repeat myself, GitHub-based ecosystem is the only cost-effective
>>> solution available for not-so-rich non-profits.
>
>> We don't need to pick one, as you can see.
>> We're using all of them: github.com, gitlab.com, git.osgeo.org/gitlab and git.osgeo.org/gogs. Decide where to focus and invest your time !
>
>
>>  --strk;
>
> I concur with strk.  For my purposes making sure things are ready at time I need to distribute is most important to me.


OK.
As I mentioned, I also found my way through.

> The more environments we have for testing the better.

As a side project, sure, great.
As a large non-profit org, that is unnecessary distraction.

The only reason I keep arguing about external infrastructure is because,
from POV of development workflow and practice, no arguments have ever been
presented, here or OSGeo lists, that prove self-hosting is an
advantage for OSGeo.
I'd love to see one, so I can finally shut my mouth up :-)

Best regards,
-- 
Mateusz Loskot, http://mateusz.loskot.net