AW: Re: Linux in a binary world... a doomsday scenario

AW: Re: Linux in a binary world... a doomsday scenario

[dirk]

>
>El Thu, 8 Dec 2005 13:23:11 +0000 (UTC),
>Dirk Steuwer <dirk@steuwer.de> escribió:
>
>
>> For a hardwaredatabase i like to see a structure. Kernel developers are 
>> required to enter the support into the database, when submitting the 
>driver.
>> Ongoing status will be logged there as well. Status and devices can only be 
>
>> entered by kernel developers.
>
>[Please don't remove the CC list]
>
>This sounds like the typical nightmare that never is 100% accurate and
>needs lots of mainteinance (developers not updating the wiki, etc) as
>Lee Revell pointed out.
>
>IMO the one way of creating such database is automating it. If you
>could get a list of the device IDs supported by drivers you 
>could (?) use the pciid/usbid/whatever list to build a user-readable
>database of the devices supported by the linux tree. Maybe it won't
>100% perfect but...

Yes, i can see the problem.
How about interconnecting it with the bugtracker?
If there is a bug, and if it is related to some hardware, it is logged in the database as broken for that kernel version. If the bug is fixed, support status is ok again.
All that needs to be done is entering the device once into the database, status is broken by default, and take it from there?
Then it gets some goals (similar to bugs) assigned if it is a complex device. i.e. for a graphic device:
* 2d graphic support
* 3d graphic support
* framebuffer
* vesa

one can close goals similarly to bugs, and if a second tester find something broken, it gets filed as a bug.
As i see it the system has to be simple to use and if used, provide an advantage to kernel developers. This makes sure it will get used and could provide an instant status about linux hardware support.

regards,
Dirk

AW: Re: Linux in a binary world... a doomsday scenario

[dirk]

>
>El Thu, 08 Dec 2005 16:49:46 +0100,
>dirk@steuwer.de escribió:
>
>> How about interconnecting it with the bugtracker?
>
>bugzilla is probably the best example of why human-managed "databases"
>are never 100% accurate and need lots of mainteinance 8) (take a look
>at mozilla's or kernel's bugzilla...). I'm tracking manually some 
>of the new devices supported in http://wiki.kernelnewbies.org/LinuxChanges
>but there're so many changes under drivers/* that god knows how many
>things I am missing. Expecting that people will maintain a wiki or a
>buzgilla or anything similar properly is like expecting that people
>will document or compile-test their patches before submitting them :P
>
>I think that the infrastructure for building such database automatically
>is already there: In the same way MODULE_DEVICE_TABLE is used by hotplug
>& friends to load the right module you can use MODULE_DEVICE_TABLE to
>build a database of the devices supported by a kernel compiled with
>"make allmodconfig", parse it and put it in a web page.


OK this would get all device ids into the database, but it would not say anything about how well the device is supported. It could be anything from the moment you add it / partial support /  pick this - excellent hardware choice random for newbie

So how about the kernel testing discussion months ago. Should there be a progam that people/developers could run, if they wanted to, that collects information from the current kernel and before sending it to kernel.org it asks the user about how happy he is with the hardware currently connected/installed to the machine.

Alternatively/Additionally you have an automated database and people keep adding things in a wikimanner to it, with some apropriate structure to it.

But still, if even bugzilla is not the perfect tool, can it be improved?

regards,
Dirk

Re: AW: Re: Linux in a binary world... a doomsday scenario

[Lee Revell]

On Thu, 2005-12-08 at 16:49 +0100, dirk@steuwer.de wrote:
> Yes, i can see the problem.
> How about interconnecting it with the bugtracker?
> If there is a bug, and if it is related to some hardware, it is logged
> in the database as broken for that kernel version. If the bug is
> fixed, support status is ok again.
> All that needs to be done is entering the device once into the
> database, status is broken by default, and take it from there?
> Then it gets some goals (similar to bugs) assigned if it is a complex
> device. i.e. for a graphic device:
> * 2d graphic support
> * 3d graphic support
> * framebuffer
> * vesa 

This is a grave oversimplification of how it would work.  Look at sound
hardware, vendors come out with new devices so fast that it would be a
full time job to keep that ALSA wiki 100% accurate.  At least half the
bugs that users report aren't real bugs.  There are gazillions of small
variations of the same device.  Vendors are deceptive at best and lie
through their teeth at worst.  A "Sound Blaster Live! SB0410" could use
a COMPLETELY different and VASTLY inferior chipset to a "Sound Blaster
Live! SB0350".  They release two devices with identical PCI ids but
different hardware.  And they don't tell us word one about their plans.
Often the users know a lot better than the developers do what devices
are supported to what degree.  Etc.

If we followed your scheme 95% of supported hardware would be listed as
broken.

Lee 

Re: AW: Re: Linux in a binary world... a doomsday scenario

[Lee Revell]

On Thu, 2005-12-08 at 15:13 -0500, Dave Neuer wrote:
> On 12/8/05, Lee Revell <rlrevell@joe-job.com> wrote:
> > On Thu, 2005-12-08 at 16:49 +0100, dirk@steuwer.de wrote:
> > > Yes, i can see the problem.
> > > How about interconnecting it with the bugtracker?
> > > If there is a bug, and if it is related to some hardware, it is logged
> > > in the database as broken for that kernel version. If the bug is
> > > fixed, support status is ok again.
> > > All that needs to be done is entering the device once into the
> > > database, status is broken by default, and take it from there?
> > > Then it gets some goals (similar to bugs) assigned if it is a complex
> > > device. i.e. for a graphic device:
> > > * 2d graphic support
> > > * 3d graphic support
> > > * framebuffer
> > > * vesa
> >
> > If we followed your scheme 95% of supported hardware would be listed as
> > broken.
> >
> > Lee
> 
> Well, let's start proposing solutions that will work then.
> 

I like the idea of a centralized database, split up by subsystem, that's
maintained by the developers in a similar way to the ALSA soundcard
matrix.  If a user finds an inaccuracy in the soundcard matrix or
discovers a new hardware model that works they submit a bug report
against the soundcard matrix, ideally containing a patch against the
XML.

It HAS to live alongside the code in the same version control system as
the code itself so it won't drift.

Lee