CVS Update: xc (branch: trunk)

[Felix Kühling]

Am Dienstag, den 01.02.2005, 06:49 -0500 schrieb Mike A. Harris:
> On Tue, 1 Feb 2005, Felix Kühling wrote:
> 
> >Am Dienstag, den 01.02.2005, 00:33 -0500 schrieb Mike A. Harris:
> >> On Sun, 30 Jan 2005, Felix Kuehling wrote:
> >> 
> >> Why not use "BusType" instead of "ForcePCIMode", as the former 
> >> seems to be more standardized, and other drivers have been 
> >> changed to use BusType in the past, which is also more generic.
> >
> >Troll's answer: There will never be PCI-Express Savages. ;-)
> >
> >Seriously, the option has nothing to do with the real bus type. It is
> >really intended for debugging and for testing PCI stuff on AGP hardware.
> >I think this option name makes this more clear than BusType.
> 
> There may or may not ever be PCI-E Savages, but that's kindof 
> besides the point.

Sorry, I just couldn't resist.

>   The Radeon (and other?) drivers used to have 
> a "ForcePCIMode" option used for the exact same purpose that 
> you're indicating you're using it for, and it was renamed to be 
> BusType in newer radeon driver code to be consistent with other 
> drivers and to generically handle different bustypes.

True. But do I need such a generic option in a driver that is never
going to need it. Maybe a generic option looks nice to you and me, but
makes things more confusing for users. For example it wouldn't make
sense to set BusType to AGP on PCI hardware. But a user may be tempted
to think that this option enables him to magically upgrade his PCI card
to an AGP card. My point is that an option name and semantics should be
chosen such that it clearly indicates what you can and can't do with it.
ForcePCIMode does that.

> 
> When adding new options to drivers, it is a good idea to see if 
> there is already a common option used in others, and aim for 
> consistency. 

I did. I was fully aware that the Radeon driver's ForcePCIMode option
was once changed to the more generic BusType option. But I also saw that
the r128 driver still has a ForcePCIMode option. I thought r128 was a
better model for the Savage driver.

>  It really sucks if someone adds a config option and 
> then later renames it to something more sensible and/or generic, 
> because it breaks user's deployed X server configuration when 
> they upgrade to a new OS release.

I agree. But I'm pretty sure there will never be a need to change this
option to a more generic one in this driver.

> 
> The result of this is potentially tonnes of end user bug reports 
> unless the older option is retained for backward compatibility 
> with what it was called before.
> 
> Just some thoughts...
> 

-- 
| Felix Kühling <fxkuehl at gmx.de>                     http://fxk.de.vu |
| PGP Fingerprint: 6A3C 9566 5B30 DDED 73C3  B152 151C 5CC1 D888 E595 |