[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [patches] please add timepps.h



В Fri, 26 Feb 2010 14:20:29 +0000 (UTC)
"Joseph S. Myers" <joseph@xxxxxxxxxxxxxxxx> пишет:

> On Fri, 26 Feb 2010, Alexander Gordeev wrote:
> 
> > So, what's the decision?
> 
> EGLIBC only accepts patches that meet the legal requirements for
> inclusion in FSF GLIBC.  Thus, we cannot accept a new header without
> a copyright assignment to the FSF.  (It should also be formatted
> according to the GNU Coding Standards and use the same license as
> GLIBC - LGPL, not GPL.)
> 
> I would also suggest submitting such a new API for FSF GLIBC -
> although as there are no functions exported from the shared
> libraries, just inline functions, binary compatibility issues do not
> arise and so it is certainly possible in principle to add such APIs
> without them going in FSF GLIBC; there would be no need to disable
> them by default as described in
> <http://www.eglibc.org/archives/patches/msg00642.html>.
> 
> New interfaces certainly need documentation in the libc manual to be 
> accepted by EGLIBC.  (Unfortunately, FSF GLIBC is laxer there and has 
> added many functions without documentation.)
> 
> That said, I would question whether this is the right place for this 
> interface - there are an enormous number of ioctls in the Linux
> kernel in support of different drivers, filesystems, etc., and I do
> not believe it is the place of libc to provide wrappers round all of
> them rather than just the basic ioctl function that other libraries
> can use to provide friendlier interfaces.  These functions do not
> seem to be closely entangled with the implementations of existing
> functions, or related to such functions.
> 
> And C library headers should provide a namespace-clean interface;
> they should not gratuitously include other headers that are only
> needed for the implementation, not the interface, and should only use
> symbols in the user namespace where strictly necessary as part of the
> interface.  (The user should be able to define arbitrary
> user-namespace identifiers - e.g. "ret" 
> - as macros before including a header, except for those identifiers 
> specifically documented as part of the interface to the header.)
> This would tend to indicate that the right thing to do is to put
> these functions in the library, not inline in the header - so meaning
> that the ABI compatibility issues *do* arise.  Even if you provide a
> separate library, as I'd suggest, making the headers namespace-clean
> is still a good idea.

OK, thank you all for the advices and active participation! I'll come
back if there are any results.

-- 
  Alexander

Attachment: signature.asc
Description: PGP signature