On 07/23/2018 10:00 PM, Luca Boccassi wrote: > On Mon, 2018-07-23 at 21:51 +0200, Daniel Borkmann wrote: >> On 07/23/2018 06:26 PM, Stephen Hemminger wrote: >>> On Mon, 23 Jul 2018 16:58:39 +0100 >>> Luca Boccassi <bl...@debian.org> wrote: >>> >>>> On Mon, 2018-07-23 at 15:49 +0000, 张 敬强 wrote: >>>>> 在 2018年7月23日星期一 CST 下午6:50:06，Luca Boccassi 写道： >>>>>> Are those headers intended as _public_ API, with all that >>>>>> entails >>>>>> (no >>>>>> breakages, etc etc)? >>>>> >>>>> bpf_elf.h is installed in the Makefile line 92, so I guess yes. >>>>> >>>>> bpf_api.h is not installed, but macro `__section_tail` is >>>>> useful for >>>>> tail calls, which is the only one in that file that may fail >>>>> between >>>>> different >>>>> iproute2 ABI. >>>>> According to the Note section in that file, I guess yes. But >>>>> upstream >>>>> didn't >>>>> install it, it really confuse me. Do you know how to contact >>>>> upstream >>>>> for a >>>>> verification? >>>> >>>> Hi Stephen, >>>> >>>> Is bpf_api.h supposed to be installed by make install? Are >>>> bpf_api.h >>>> and bpf_elf.h public API that should be shipped by distros? >>> >>> Not sure how much of BPF iproute2 should be installing, versus >>> allowing other >>> packages to do it. Daniel? >> >> The bpf_elf.h should be the only one, which is the case today already >> for >> the `make install`. The bpf_api.h can be used by BPF program writers, >> but >> it's not mandatory at all to do this, so I would prefer to leave this >> up >> to program authors instead and not install it. >> >> Thanks, >> Daniel > > Hi Daniel, > > Thanks. So if the bpf_elf.h API can be considered stable (backward > compatibility, etc) then I'll talk with the other Debian maintainers > and consider shipping it in Debian 10.
Great, sounds good! Thanks a lot!
