On 01/02/2018 07:09 AM, John Ferlan wrote:
On 01/02/2018 04:28 AM, Michal Privoznik wrote:
Unfortunately, since gnulib's commit of 2c5d558745 there's an unused parameter to stat_time_normalize() function which gnulib developers don't want to fix [1]. Therefore, we have to work around it by temporarily suspending -Wunused-parameter.
1: http://lists.gnu.org/archive/html/bug-gnulib/2018-01/msg00000.html
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> ---
While we have 'gnulib update free' push rule, this one is not trivial at all and thus I have not pushed it. It's ugly and I don't like it. So any ideas are welcome.
.gnulib | 2 +- bootstrap | 4 ++-- src/storage/storage_util.c | 3 +++ 3 files changed, 6 insertions(+), 3 deletions(-)
No bright ideas on this other than perhaps only including changes just prior to the particular one that breaks things or somehow revert just that one in our local copy.
We can provide a downstream-only patch against the gnulib file that adds the attribute unused marker in our builds, since upstream didn't like the patch. I'll work up something like that in a moment...
Other thoughts require additional local changes just to "work around" the broken definition. Such as grabbing the "old" stat-time.h, rename it, save it in libvirt sources, then use that new name instead of the new .gnulib file. That's probably worse than what you've done though.
At this point, I think your solution to minimize the impact to one include file is perhaps the easiest/best solution albeit not perfect.
Yes, the #pragma usage is more concise than carrying a downstream gnulib patch, but the two approaches are not that much different in maintenance, so it will be a matter of taste which variant of the patch to use.
It's interesting that there is no desire to fix this problem in .gnulib especially since there are already 2 patches proposed and in reality the change fundamentally breaks on every platform other than __sun when STAT_TIMESPEC is defined just because it's possible (or more desirable in the reviewers feedback) to compile with ignore unused-parameter.
I'm also going to reply in the upstream gnulib thread. When the warning is in a .c file, they are justified in not caring. But when it is in a .h file, it really does seem like something worth cleaning up in gnulib.
Wonder what would happen if someone posted a patch to .gnulib to revert the change for the reason that it breaks other platforms that don't desire to configure in that manner. Perhaps Eric Blake would have some thoughts or additional muscle with the .gnulib maintainers.
-- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3266 Virtualization: qemu.org | libvirt.org