| 1 |
On Thu, Sep 29, 2011 at 12:19 PM, Volker Armin Hemmann |
| 2 |
<volkerarmin@××××××××××.com> wrote: |
| 3 |
> Am Mittwoch 28 September 2011, 17:15:34 schrieb Grant Edwards: |
| 4 |
> |
| 5 |
>> |
| 6 |
>> Regardless, my point was that Linus's statement that it's unacceptable |
| 7 |
>> to break things seemed rather disingenuous given the API churn that |
| 8 |
>> Linux has compared with the BSD kernels. |
| 9 |
> |
| 10 |
> Linux has zero userland visible API 'churn'. |
| 11 |
> |
| 12 |
> You can't have less than zero. |
| 13 |
|
| 14 |
Uh, that can't be right. Largely, libc masks things. |
| 15 |
|
| 16 |
Several kernel options explicitly state in their description that they |
| 17 |
require new-enough versions of this or that userland tool to function |
| 18 |
properly. Randomizing module base addresses is one of those, IIRC. |
| 19 |
Some things related to sysfs. sysfs itself. I think some network |
| 20 |
filesystems. modutils. |
| 21 |
|
| 22 |
If there's no API churn, it should be pretty trivial to run a current |
| 23 |
userland on top of, e.g. 2.6.0-pre1, or even 2.6.0. I also STR 2.6.9 |
| 24 |
being a common pin point where a bunch of userland tools required |
| 25 |
that-or-newer. |
| 26 |
|
| 27 |
And that's ignoring dropping things like A.OUT support. |
| 28 |
|
| 29 |
I'm not arguing whether or not it's reasonable (it almost certainly |
| 30 |
is), but there certainly is churn. |
| 31 |
|
| 32 |
-- |
| 33 |
:wq |
Archives