RFC: Making the xfs font server optional in Fedora Core and its derivatives.

[Mike A. Harris]

Nicolas Mailhot wrote:
> Le mardi 23 mai 2006 à 10:22 -0400, Adam Jackson a écrit :
> 
>> I'm of the opinion that leaving xfs enabled isn't really a big deal 
> 
> Unfortunately, as every single FCxT1 release shows, leaving xfs enabled
> means various users will have their X server fail mysteriously because
> of problems locating "fixed".

Actually, thanks to our fairly well beaten into shape xfs initscript,
the number of people who have seen the "can't find font fixed" error
during X server startup in the last several years has declined
significantly to a negligible few, and that's usually due to
experimental test changes that have occured during development.
The alternative to seeing the above error, would be having the
X server start up with no error, and having applications fail
individually, or start up with no visible fonts.  That's not
really a better situation than seeing the above error.

> Notwithstanding the fact that they probably do not have a single
 > main app needing "fixed" anymore.

That's rather speculative.  Define "main app".  That definition
will vary from person to person, and if there are 2000000 users
out there, there will probably be 1000000 definitions.

Wether a "main app" fails or not doesn't matter.  What matters is
"Will an app that someone expects to work, actually work?"


> Core font system is not "just working" and "low impact". Every once in a
> while it manages to remind everyone it's still lurking under the X
> server. As long as the current system is unchanged it will stay one of
> the X server main points of failure.

Sure, I can agree with that.  But as I mentioned above, if we solve
the problem that causes that error to be displayed and prevents the
X server from starting, by making sure fixed is always available, then
it just masks the fact that one's font configuration may be incomplete
or broken.

When you see the "can't find font fixed" error, it is a clear sign right
now that either xfs isn't running, or xfs is misconfigured, or one or
more core font directories is misconfigured or improperly prepped.

IMHO, replacing that problem, with an X server that starts up ok,
but where the core fonts system may still be misconfigured or otherwise
unuseable by one or more apps, simply changes the problem to be a
random app failure the user can't easily diagnose - causing them to
flood bugzilla with various "my app starts with boxes instead of
letters" problems, or worse, they click on an icon on their working
desktop, and the app never starts at all - so they file a bug report
against that application.

At least with the current failure mode, the person has to fix the
font subsystem configuration problem before they can even start the
server, and once they do fix that - they are unlikely to see individual
applications failing miserably for no visible reason.  They're then
less likely to file useless bug reports in bugzilla.

It's certainly arguable which scenario is the better to choose, as
it has came up many times on many mailing lists.  The above is my
opinion only, and I must admit that I am biased a bit against seeing
bogus bug reports landing in bugzilla with my name on them as bug
owner.  ;o)

The solution Adam suggested in the previous mail, of using
fontconfig/Xft to provide core fonts is IMHO the best solution period.
The main problem with that solution however, is that people have
proposed that as a solution for 3 years, and nobody has shown any
real interest in stepping forward to implement that solution to the
best of my knowledge.


> +1 to reduce its perimeter in FC to something minimalistic, lightweight
> and robust. Even if it means reducing its functionnalities to the bare
> minimum required by the spec. Let the apps which didn't bother with
> fontconfig migration bear the weight of core font management, if they
> really want to keep it alive forever.

While I certainly agree with you 2000% in principle there, and I have
indeed done that approach with some other changes to things in the
past years, this is one that is a bit deeper IMHO.  When we have
applications like "emacs", "mplayer" and other apps people would
probably consider "killer app, I must have or I'll go use an OS that
works", we have to consider our actions much more carefully.  Are
we really that big of a distribution that we can say "we don't care
about your core font using apps anymore"?

Maybe I overestimate the number of people who would get upset over
such a decision?  Is it worth taking the risk to find out?  ;o)
And more importantly, if such risk is indeed taken, _who_ foots the
criticism for the decision when people come looking for heads to
roll?  If I am to make any potentially drastic changes of this
nature, I definitely want to have a scapegoat to pile things on.
;o)


-- 
Mike A. Harris  *  Open Source Advocate  *  http://mharris.ca
                       Proud Canadian.