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: First draft of the Y2038 design document


On Tue, Oct 27, 2015 at 12:32:50AM +0000, Joseph Myers wrote:
> On Mon, 26 Oct 2015, Rich Felker wrote:
> 
> > While _FILE_OFFSET_BITS or _TIME_BITS avoids an ABI transition in
> > libc, they force an ABI transition for all third-party libraries that
> > use the types. A library that uses off_t in its API and that's built
> > with _FILE_OFFSET_BITS=64 is ABI-incompatible with a version of
> > itself, or an application, built with _FILE_OFFSET_BITS=32. Thankfully
> 
> And Paul Eggert's analysis suggested that this ABI transition has largely 
> happened - that affected libraries on GNU/Linux systems generally are 
> built with _FILE_OFFSET_BITS=64 now.

Yes. It's just frustrating that applications have to explicitly
request it an many new developers aren't aware of this.

> > Another goal that might make this less distasteful than
> > _FILE_OFFSET_BITS would be a planned timeline for switching the
> > default to 64.
> 
> We need to get the fts patch reviewed.  At some point I might look at 
> fixing bug 14106 (with new __*64 exports for all affected functions in any 
> supported ISO C / POSIX standard) if noone else gets there first.  Then, 
> given Paul Eggert's analysis, we could seriously look at changing the 
> _FILE_OFFSET_BITS default (cf 
> <https://sourceware.org/ml/libc-alpha/2014-03/msg00290.html> as a patch 
> for that purpose - using _FILE_OFFSET_BITS=64 while building glibc itself 
> would be hard, and no doubt the same would apply for _TIME_BITS=64).

Yes, I agree these are practical concerns for changing the default
that should be fixed first. I just wish this had been planned and
dealt with 10+ years ago.

> I think for _TIME_BITS we'd similarly need to wait for widespread use 
> before changing the default.

Agreed. It would be helpful to have an aim though, possibly
coordinated with major distributions. Otherwise I fear by default it
might end up being 2037. ;-)

Rich


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