This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: PATCH: Add x32 dummy sysctl


On Mon, May 21, 2012 at 5:37 PM, Roland McGrath <roland@hack.frob.com> wrote:
>> I disagree.
>>
>> I think that glibc, of all places, must carry the burden.
>
> There's just no good reason for that.
>
>> There is a certain notion of consistency that must be kept across all
>> ports, even new ports, to make it easy to build a large collection of
>> userspace applications (see Chris' comments).
>
> I understand that notion. ?But it competes with another important
> goal: get rid of cruft when we can. ?Since sysctl is already
> deprecated by the kernel, all code should stop using it ASAP. ?For
> existing machines, we have no choice but to keep the ABI we already
> have. ?(We could become more aggressive than what I've proposed so
> far, and remove sysctl from the API while leaving it in the ABI.)
> For new machines, we aren't forced to carry it and so there is just
> no justification adequate for ABI bloat.

Certainly, I agree 100%. Yet because we have no policy we have not set
any expectations about *how* this will be done. Expectation is king
here. We need a clear policy, we tell the distros, and then we act on
it. Anything less is chaos.

For example:
Step 1. During X.Y-1 announce to distro maintainers the list of API
functions we are deprecating in X.Y, and ask for feedback.
Step 2. Mark the API deprecated in X.Y.
Step 3. Remove the API in X.Y+?, but leave it in the ABI.
Step 4. Remove the ABI in X.Y+?

Comments?

Cheers,
Carlos.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]