ubuntu bulletproof x

[Adam Jackson]

On Mon, 2007-09-03 at 14:53 -0500, Douglas McClendon wrote:
> Adam Jackson wrote: 
> > Yeah, okay, force me to clarify.  Grumble.
> > 
> > There are cases where we can't tell what monitor the user has.  They're
> > almost completely described by "either the card can't do DDC, or the
> > cable is broken".  The former is a vanishingly small class of hardware,
> > voodoo1 basically.  The latter happens depressingly often particularly
> > with projector setups.

> So, to save you the trouble of rereading all of my posts.  Can you 
> explicitly confirm this (which it sounds like you did, but not in a way 
> that clearly addressed the point I tried to make half a dozen times last 
> night).
> 
> Repeat after me-
> 
> "There is *NEVER* a situation, when the monitor fails to provide correct 
> information, due to a broken or absent edid implementation, and which at 
> the same time, sufficient information could be parsed from the .inf that 
> came on the CD with the monitor, to provide the user, a reasonable 
> experience requiring no user interaction beyond putting the cd in the 
> drive".  (and at which time, the X driver could not have accomplished 
> the same thing automatically without the .inf)

Absent EDID in the sink device never happens anymore.  It's a
requirement for Vista certification.  I'm fairly sure it was required
for XP cert.  It's a requirement for shipping any DVI sink device.  It
is _mandatory_.

We can fail to get EDID, either because the cable broke the DDC pins, or
timing bugs in the I2C code, or BIOS bugs if we're using VBE DDC, or
it's a really old monitor, or there's a crap KVM switch in the middle,
or phase of the moon, or whatever.

I have not found ISOs for every OEM CD for every monitor that ever
shipped.  I doubt I ever could.  Therefore the following claim is merely
statistical.  However, on no OEM CD that I've ever found does the
included INF file - or any other resource intended to be parsed by the
machine - provide the same set of information as the EDID block for the
monitor.  It may provide a subset.  The only subset I've ever seen is
sync ranges.

I'm not saying I'm happy about that.  I would love to see a
counterexample.  But it's all the empirical evidence I have.

- ajax