This is the mail archive of the
mailing list for the glibc project.
Re: [PATCH 21/27] [AARCH64] ILP32: introduce syscalls that pass off_t
- From: Joseph Myers <joseph at codesourcery dot com>
- To: Yury Norov <ynorov at caviumnetworks dot com>
- Cc: <libc-alpha at sourceware dot org>, <linux-kernel at vger dot kernel dot org>, <arnd at arndb dot de>, <catalin dot marinas at arm dot com>, <marcus dot shawcroft at arm dot com>, <philb at gnu dot org>, <davem at davemloft dot net>, <szabolcs dot nagy at arm dot com>, <maxim dot kuvyrkov at linaro dot org>, <pinskia at gmail dot com>, Yury Norov <yury dot norov at gmail dot com>
- Date: Thu, 23 Jun 2016 11:57:46 +0000
- Subject: Re: [PATCH 21/27] [AARCH64] ILP32: introduce syscalls that pass off_t
- Authentication-results: sourceware.org; auth=none
- References: <1466485631-3532-1-git-send-email-ynorov at caviumnetworks dot com> <1466485631-3532-23-git-send-email-ynorov at caviumnetworks dot com> <alpine dot DEB dot 2 dot 20 dot 1606211033060 dot 4526 at digraph dot polyomino dot org dot uk> <20160623055756 dot GD13449 at yury-N73SV>
On Thu, 23 Jun 2016, Yury Norov wrote:
> So for now I think it's simpler to have this ABI in sysdeps, and be in
Of course it goes in sysdeps. But not architecture-specific sysdeps.
And "simpler" for initial implementation may not be simpler for future
maintenance; when there are too many implementations of a function, they
have a tendency to get out of sync, and to cause trouble for future global
changes. There are two plausible options that I see:
* sysdeps/unix/sysv/linux/<some-descriptive-name>, where
<some-descriptive-name> is an architecture-independent name for the
relevant ABI feature.
* Have a macro __SOME_ABI_FEATURE in some sysdeps header, and then use
conditionals on that macro in existing implementations in
sysdeps/unix/sysv/linux/generic or sysdeps/unix/sysv/linux.
I think the second one is preferable. If you prefer the first one, you
should be able to justify it by giving some detailed examples of what the
different implementations look like and why there is actually nothing
useful in common between them that would allow shared implementations with
Joseph S. Myers