On Thu, Mar 02, 2006 at 02:54:57AM +0100, Lennart Poettering wrote: > On Thu, 02.03.06 01:16, Joe Orton ([EMAIL PROTECTED]) wrote: > > If you compile everything, including neon, with _FILE_OFFSET_BITS=64, > > then everything works fine (at least from neon's perspective). If you > > mix and match apps and libraries built with _F_O_B=64, then yes, you'll > > have lots of strange and horrible problems. So... don't do that! > > > > Having neon export _F_O_B=64 by default would cause similar problems for > > apps which *don't* build using that flag. Really, just avoiding _F_O_B > > definitions completely is the only safe bet. > > I believe that *all* apps and libs should define _F_O_B=64 > (if supported by the libc). > > The thing is that it is not my decision whether to compile libneon > with 64bit support or without. If Debian (Laszlo Boszormenyi) chose to > compile the lib with 64bit off_t enabled it would become > ABI-incompatible with all other distributions that do not enable > it. Therefore I am asking you, as the maintainer of NEON, to make that > switch upstream. In my opinion all new libraries/apps should use 64bit > off_t, support for 32bit off_t should be retained only for legacy > sources.
As I wrote above: exactly because _F_O_B=64 is "all-or-nothing" the only place that can be applied with reliable results is at distribution level, since that is the only place where you can apply it everywhere. Having individual packages randomly flip the switch is how you get into this mess, not how you get out of it. No, neon will not do that by default just to fix the problem for you: it would *create* the problem for the next guy. The viable solutions: 1) solve it consistently at distro level 2) avoid off_t in the API completely 3) add LFS-style interfaces and stick to them I agree that (1) sucks because it changes ABI. (2) is a bit more work, probably reasonable for neon if you've got a few round tuits, (3) is easy so I've done that for 0.26.x in the mean time. joe _______________________________________________ neon mailing list [email protected] http://mailman.webdav.org/mailman/listinfo/neon
