NVMe-CLI intel-nvme.c RHEL 6.9 Compile Errors

[Keith Busch]

On Mon, Oct 23, 2017 at 09:48:45AM -0600, Scott Bauer wrote:
> On Mon, Oct 23, 2017 at 03:43:36PM +0000, Jeffrey Lien wrote:
> > Keith, 
> > We noticed these compile errors when trying to build the latest version of the NVME CLI on RHEL 6.9 and SLES 11.4:
> > [ddtest at gfox3-rh69 nvme-cli-kb]$ make
> > NVME_VERSION = 1.4
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c argconfig.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c suffix.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c parser.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c nvme-print.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c nvme-ioctl.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c nvme-lightnvm.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c fabrics.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c json.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c plugin.c
> > cc -D_GNU_SOURCE -D__CHECK_ENDIAN__ -O2 -g -Wall -Werror -std=gnu99 -DNVME_VERSION='"1.4"' -c intel-nvme.c
> > cc1: warnings being treated as errors
> > intel-nvme.c: In function \u2018get_internal_log\u2019:
> > intel-nvme.c:782: error: dereferencing pointer \u2018ehdr\u2019 does break strict-aliasing rules
> > intel-nvme.c:785: error: dereferencing pointer \u2018ehdr\u2019 does break strict-aliasing rules
> > intel-nvme.c:741: note: initialized from here
> > intel-nvme.c:643: error: \u2018ad\u2019 may be used uninitialized in this function
> > intel-nvme.c:645: error: \u2018ehdr\u2019 may be used uninitialized in this function
> > make: *** [intel-nvme.o] Error 1
> > 
> 
> If you need to build it right now while I fix it you can add -fno-strict-aliasing.
> 
> As for the actual error, I think the c-spec allows for a char type to be aliased to
> any other type without invoking undefiened behavior. I'll try and find that in the spec
> to verify. What GCC version does RHEL 6.9 have?

Some kind of weird casting rules. Should be fixed now.