Per-app MSI-based installers

[John Layt]

*Puts on Devil's Advocate cape*

On 2 July 2014 19:02, Kevin Funk <kfunk at kde.org> wrote:
> On Wednesday 02 July 2014 19:35:15 Boudewijn Rempt wrote:

>> But honestly... Is that really large? In this day and age 100mb is not a lot
>> of bytes to download.

For a user in the west who has a large hard drive and a fast internet
connection and really wants that app?  Maybe not.  But for other users
with slower links or old hardware or who just want to have a quick
trial it could be a barrier.  Saying you need to download another
100MB every time you want another KDE app no matter how small will
just leave a reputation for being bloated.  While I'm not one for
clinging on to supporting really limited hardware, we're also not just
here to make life better only for the more well off (or at least I'm
not).

We have the same argument on Mac where most apps are distributed as a
complete bundle, but what about Android or iOS or Blackberry where
users do get jealous of every MB that an app takes up?  Will we build
and support completely different packaging and distribution models on
every platform?

Also, as a distributor, can KDE afford to host such large packages?
Will our mirrors be happy if every app requires a 100MB+ bundle and we
have 100 apps?  I know when it was proposed that Qt ship ICU on
Windows and Android and Mac that the app devs were outraged at the
thought of having to pay for an extra 10MB in every download, as well
as the 30MB extra hard disk space used.

>> And if I've learned one thing from struggling with cross-platform packaging
>> is that it's never worth it to go outside the platform's conventions, at
>> least for applications meant for regular users. That means, no cygwin,
>> macports, fink, homebrew, kde-win installer or emerge. It also means, on
>> Windows, that an installer is complete and includes everything including
>> the right msvc runtime. It means no daemons, no running of apps like
>> kbuildsycoca that are alien to the platform.

Completely agree on the daemons and the like, it's why I keep saying
we should never use dbus or systemd directly in any apps.  All system
integration and IPC needs to be abstracted. (*)  Also mostly agree on
the bits about relying on things like Macports or Cygwin, they are too
heavy and geeky for end user apps and prone to breaking in ways
outside our control.  However, I don't think that prevents us from
doing something small scale ourselves inside our installers to share
libraries between our apps in our own install folder.

OTOH, having shared libs and individual app installers does open us to
the eternal DLL hell of someone having two different apps installed
that have two different version requirements for Frameworks, that's
one scenario having fat packages helps avoid.

(*) One problem of saying this though is that in some cases we're
effectively saying to app devs that either they remove features from
their Windows version (like Marble does) or just don't have them in
the first place on Linux, neither of which is terribly appealing.

> On Windows, users want to have a standalone installer for each application.

Do they, or is that just what they are used to?  Mac users seem to
have very quickly adapted to the idea of a store instead of hunting
down individual downloads, as have every single Android and iOS user.
Will Windows users be that different?  And I still remember the old
days when I had to go looking for DirectX or VC or .NET packages to
install first.

> I think part of the misunderstanding is that the average Windows user (which
> we'd like to target) doesn't want to install the whole KDE experience on his
> desktop.

That I definitely agree with, Windows users do want particular apps
only, and having just 3 or 4 key apps installed is the scenario we
should seek to support.  We may not want to support installing every
possible KDE utility or widget.  However we are a community and need
to look at how all of KDE could benefit from successful apps on
Windows.  Is there really any point in our having one killer app on
Windows that everyone installs, but then ignore everything else we
have on offer?  How do we make it easy for people who try Krita and
are impressed to try other KDE apps?

We also need to think about the modules, how would we package Edu or
Games?  A separate installer for each edu app or game, no matter how
trivial?  Or would a teacher or parent want a single installer they
could choose apps from?  Or perhaps install them all and have a single
launcher interface?

> But seriously, that's cheap nowadays

Not for everyone.

> Windows users are used to big binary blobs. (Just reminds me of HP's printer
> driver installers, which are roughly about 200 MB...)

Which *everyone* complains about and only download because they *have*
to.  We're not in the position of an HP or Adobe, we should try to
make the marginal cost of installing more KDE apps as small as
possible.

Look, there's good arguments either way, and I'm really not in a
position to say what's right, but I'm looking at it from the
perspective of getting as many KDE apps on as many platforms as
possible.  If I wrote an app and faced having to learn 5 different
packaging methods and 5 different distribution methods then I might
not bother.  We need to make it easier.

John.