Messages in this thread |  | | | Date | Mon, 10 Dec 2007 00:39:51 +0900 | | From | Tejun Heo <> | | Subject | Re: 2.6.24-rc4-git5: Reported regressions from 2.6.23 |
Alan Cox wrote:
>> Newly broken ones will be regressions. How many do we fix by the
>> change? On SATA, setting the correct transfer chunk size doesn't seem
>> to fix many.
>
> Regressions are not some kind of grand evil. Better to regress the odd
> device than continue to break entire controllers.
We need to put more weight on regressions as it at least makes releases
predictable to users. Anyways, I wasn't saying it was some absolute
maxim. I was literally asking how many so that we can evaluate the
trade off.
>>> Tejun - instead of backing out important updates for 2.6.24 we should
>>> just blacklist that specific drive for now and sort it nicely in 2.6.25,
>>> not revert stuff and break everyone elses ATAPI devices.
>> We'll need to blacklist setting transfer chunk size, eek, and let's
>> leave that as the last resort and hope that we find the solution soon.
>> Blacklist takes time to develop and temporary blacklist for just one
>> release doesn't sound like a good idea.
>
> It seems to be sensible to me *if* it is just this one device we are
> somehow confusing and that one device is holding up fixing everything
> else.
Yeah, if it's this one device, I fully agree. Let's see how debugging
turns out.
--
tejun
|  |