svn commit: r467193 - in head/www/gitlab: . files

[Sunpoet Po-Chuan Hsieh]

On Fri, Apr 13, 2018 at 9:35 PM, Joseph Mingrone <jrm at freebsd.org> wrote:

> Matthias Fechner <idefix at fechner.net> writes:
> > Dear sunpoet,
>
> > Am 12.04.2018 um 20:33 schrieb Sunpoet Po-Chuan Hsieh:
> >> Log:
> >>   Fix Gemfile for rubygem-default_value_for 3.1.0 update
>
> >>   - Bump PORTREVISION for package change
>
> >> Modified:
> >>   head/www/gitlab/Makefile
> >>   head/www/gitlab/files/patch-Gemfile
> > I really enjoy your work and it is totally fine for me, if you upgrade a
> > gem that you copy the old one and make sure that gitlab uses the old.
> > (like you did with r467106, thanks a lot for this!)
>
> > But please do not start to patch the Gemfile!
> > This will break gitlab. Have you tested gitlab with this modification?
> > (for sure not, I reported exactly that version of default_value_for
> > upstream and they rolled back the change as it broke gitlab, they broke
> > 3.0.4 and rolled back the modification with 3.0.5, now you changed
> > gitlab to use the version that for sure will not work!)
>
> > Please revert this modification and ask me (the maintainer) before you
> > start to modify the port I maintain.
> > Thanks a lot that you understand this.
>
> > Gruß
> > Matthias
>
> I relate.  Sunpoet, I also appreciate your work, but this type of
> breakage can be frustrating.  We need to figure out a solution.
>

Hi,

As I replied before, I'll keep the Gemfile unchanged.
The rubygem ports will follow the version requirement in Gemfile.

In fact, I stored the Gemfiles of gitlab, mastodon and vagrant in a
specific directory.
When I update a rubygem port, I check the dependent ports to see if they
accepts the newer version.

Regards,
sunpoet


> > On 13.04.2018 12:07, Alexey Dokuchaev wrote:
> >> I don't quite understand how/why would anyone in their right mind write
> >> software than breaks when one of its dependencies have *minor* version
> >> bump?  Is this specific to Gitlab, or the nature of Gemfiles is really
> >> that broken?
>
> This practice seems to be more common and in the extreme case, exact
> versions of all components are specified.  A few examples that come to
> mind other than Ruby Gems (with Bundler) are node and Haskell's stack.
>
> To match what upstream is requesting, I have wondered if the ability to
> specify maximum versions would be useful.
>
> RUN_DEPENDS=
> rubygem-default_value_for>=3.0.1<3.1:devel/rubygem-default_value_for
>
> The added complexity is probably not worth it for the ability to run
> this sort of test.
>
> devel/rubygem-default_value_for % check_update_breaks_something
> *** DANGER, DANGER *** www/gitlab requires devel/rubygem-default_value_for
> < 3.1 !
>
> A less drastic approach would be to add PORTSCOUT entries to prevent
> premature Gem port updates.  In this case
>
> PORTSCOUT=  limit:^3\.0


> in devel/rubygem-default_value_for/Makefile


> Could this work for everyone?


> J.
>