On 9/5/19 7:25 AM, Jeff Law wrote:
On 9/5/19 1:34 AM, Florian Weimer wrote:
(Martin, can you give this a test using GCC 10? I don't expect any
issues because the zero-length array is completely gone. The patch
passed testing on x86_64-linux-gnu without regressions, but with GCC 9.)
The union wrapper is unnecessary because C allows to read any object
as a sequence of chars.
2019-09-05 Florian Weimer <fweimer@redhat.com>
[BZ #24962]
* locale/localeinfo.h (_nl_category_names): Remove union wrapper.
(_nl_category_names_get): New function.
* intl/dcigettext.c (category_to_name): Call it.
* locale/findlocale.c (_nl_find_locale): Likewise.
* intl/loadlocale.c (_nl_load_locale): Likewise.
* locale/newlocale.c (__newlocale): Likewise.
* locale/setlocale.c (_nl_category_names): Adjust definition.
(_nl_category_name_idxs): Likewise.
(new_composite_name): Call _nl_category_names_get.
(setlocale): Likewise.
Might be easier for me than Martin. I just threw it into my github
repo, so my tester will pick it up as of now. Will report back in a few
hours.
Thanks Jeff. I'm sure your coverage will be more comprehensive
than mine. (The warning Florian is fixing only shows up with
the patch for PR 91631 I posted yesterday so you need both in
your tested to verify them.)
It's pretty easy for me to test it on x86_64-linux: with your
patch, Florian, and my enhanced GCC 10 I don't see any warnings.