[OSGeo-Discuss] What would you want from an OSGeo Git Service ?

[Sandro Santilli]

On Fri, Apr 15, 2016 at 02:21:22PM +0200, Mateusz Loskot wrote:
> On 15 April 2016 at 09:52, Sandro Santilli <strk at keybit.net> wrote:
> > What would you want from an OSGeo Git Service ?
> 
> My 10 features are:
> 
> 1. LDAP (log in with OSGeo User ID or OpenID)
> 2. Private repositories
> 3. Organizations (e.g./GDAL)
> 4. User space (fork /GDAL/gdal into /mloskot/gdal, also private fork)
> 5. Issue tracker (internal, as external would never be as well integrated)
> -- Milestones, labels/categories, commit keywords (e.g. Fixes #123)
> -- Comment issues via mail is not critical, but nice.
> 6. Wiki (internal, see above)
> 7. Code review (comments on diff lines is a minimum)
> 8. Pull requests
> 9. CI (integration with Travis CI and AppVeyor is a minimum)
> 10. Webhooks and any other mean to integrate with IRC, Gitter, Slack,
> whatever teams like to use.
> 
> The comparison table [1] so far, I think, makes it clear GitLab is the only
> self-hosted solution which is close to what we've got now: Subversion + Trac.
> It also matches my 10 points.
> 
> Gogs wins due to low*** maintenance requirements,
> but it will require custom development what, I think,
> is a deal breaker - we have NO resources for this.
> 
> GitLab wins feature-wise, but its maintenance might turn
> very demanding***. If bigger hassle than SVN+Trac this
> also might be deal breaker - we have VERY limited resources.
> 
> ***We need to allocate budget for admins!

I would add that IFF all projects migrate to a new infrastructure
(like Gogs or GitLab) the overall maintainance cost might _reduce_
as enabling a feature (like "tickets update via commit log")
currently requires changing the configuration of each project
repository, rather than a single configuration valid for
everybody (or an autonomous setting for it).

> Finally, GitHub, wins: feature-wise, marketing-wise, with 'zero' maintenance
> - most, if not all, of our projects already prefer GitHub.

1. marketing wasn't in your 10 points ?
2. 'zero' maintainance is true for any hosted solution (including
   gitlab.com, for example).

> The only reason we haven't done it already is the cost.

Done what ? Most projects don't need private repositories so could
move at no cost. Or you mean the cost of having LDAP authentication ?

> Let's allocate budget for paid account.

Are you willing to collect quotes for hosted git service plans
meeting all of your 10 features above ? Do we have an idea about
the actual needs in term of number of repositories or other similar
parameters required ?

Personally I wouldn't like to see an "open source" foundation pay for
services based on "closed source" software, but I guess there are open
source based companies offering hosted services too.

> However, the very first question is still open:
> Do we want or need to switch at all?
> Are all teams happy with the OSGeo SVN+Trac setup?

I guess this needs to be an individual question, not a per-team one.
Personally I'd like to stop using SVN, in general, and being able
to easily publish my development branches for consideration by
projects (and discussion over them). Current trac setup allows to
switch to git but management of forks and pull requests are not
available out-of-the-box (didn't do any research about plugins
to do that). Add to that the problem of needing to enable each feature
on each of the projects and find that it migth be just easier
to switch all to Gogs (28 repositories are already registered there,
and it is only an experimental deploy).

Looking forward for others to answer this fundamental question:
Are you happy with OSGeo SVN+Trac setup ?

--strk;