Re: Using .tar.xz only on ftp.gnome.org (was: install-module / ftp.gnome.org / master.gnome.org)

[Alberto Ruiz <aruiz gnome org>]

Sorry Olav,

I thought this is actually going to happen soon and the decision was
set in stone.

2011/3/21 Olav Vitters <olav vitters nl>:
> On Mon, Mar 21, 2011 at 01:21:38PM +0000, Alberto Ruiz wrote:
>> to me that it will actually be easier to request our sponsors
>> (RedHat/Novell/Oracle...) for more storage than pushing everyone into
>> the pain of having to deal with such unfamiliar format. We are
>
> Could you expand on the pain?

Basically most distros won't ship it by default, at least not for now.
Solaris/RHEL/SLED/CentOS are probably the worst cases here. But also
custom environments (we are talking about things like glib heavily
used in embedded...).
I'm not saying these are showstoppers, I am saying that we should give
them some time to adapt, try to encourage Fedora/OpenSuSE/Ubuntu to
ship lzma by default would be a good starting point. Letting the
enterprise guys at least provide packages/repos for it would be nice
as well.

It is true that people using packages more heavily are the distros and
jhbuild-ish setups, but they automate their infrastructure in a fixed
setup anyway so they are not the ones we should worry the most. I'm
thinking about the casual people trying to build a given tarball from
some platform and not being able to. Maybe we should actually make a
difference between packages from the desktop release set and the
platform/development ones in this regard? (i.e. shipping bzip2/xz for
things like glib/gtk/vala/gjs and just xz for cheese/gnome-shell/...
and personal projects?

Do you think having a .bzip2/.xz transition time would make sense or
make things any better? Say dual during 3.0 and just .xz for 3.2
onwards?

> I won't assume, I'm in no hurry. Nothing will happen before I fully
> understand implications and everyone is aware (d-d-l is just to plan;
> not to check impact).

Yup, sorry, I thought you were going ahead actually. :-)

>> I am not trying to criticise the effort rather than trying to
>> understand why is this such a big win.
>
> Various reasons:
> 1) Way less storage space required
> 2) Everything stored needs to be backed up
>   Seems simple, but a lot of the new machines we have are not backed
>   up.
> 3) Less storage means quicker backups
> 4) Total size of ftp.gnome.org will mean how many people will be able to
>   mirror it
> 5) I want more mirrors in different continents
>   ATM, we have 1. I'd like around 5, but that means 5 times the current
>   bandwidth. Which is free, but slow (USA->.nl)
> 6) Xz makes 'install-module' *much* slower; only one format would
>   alleviate it a bit

6 seems a reason not to :-)

> But mostly:
> 1) I'm still investigating / it is still being planned
> 2) If every problem can be solved, I don't see why not (#4)
> 3) Currently, I think most users of ftp.gnome.org are either packagers,
>   or developers.
> 4) I understand it will have an impact, but from my current
>   understanding of the tarball users, I don't believe it it an
>   inconvience
>   Seems easier switch than CVS->SVN->Git.
> 5) Already in use
>   From what I noticed, GNU already is .xz only
> 6) Open to feedback
>   ATM I haven't noticed anything major; though:
>    *) have to await the more details about Solaris
>    *) didn't ask for feedback
> 7) I'm in *no* hurry.
>
> --
> Regards,
> Olav
>



-- 
Un saludo,
Alberto Ruiz