[dpdk-dev] Build with coverage fail on rte_flow_dynf_metadata_offs is not flagged as experimental

[David Marchand]

On Sat, Nov 23, 2019 at 11:24 AM Andrew Rybchenko
<arybchenko at solarflare.com> wrote:
>
> Hi,
>
> I've failed to quickly make a patch to fix it.
>
> make config T=x86_64-native-linux-gcc O=build_coverage
> make -j8 EXTRA_CFLAGS='-ggdb --coverage' EXTRA_LDFLAGS=-lgcov
>
> results in
>
> ...
>    CC ethdev_private.o
>    CC rte_ethdev.o
>    CC rte_class_eth.o
>    CC rte_flow.o
>    CC rte_tm.o
>    CC rte_mtr.o
>    CC ethdev_profile.o
> rte_flow_dynf_metadata_offs is not flagged as experimental
> but is listed in version map
> Please add __rte_experimental to the definition of
> rte_flow_dynf_metadata_offs
> /home/arybchik/src/dpdk/mk/internal/rte.compile-pre.mk:114: recipe for
> target 'rte_flow.o' failed
> make[4]: *** [rte_flow.o] Error 1
>
> I've manually reproduced it using gcc (Ubuntu 7.4.0-1ubuntu1~18.04.1) 7.4.0
> but seen it on many different distros.

Oh, good catch.

We have a problem in the check script and maybe in the experimental tag.

- This rte_flow_dynf_metadata_offs symbol is in .data but gets an
associated symbol (seen with rhel7 gcc):
$ objdump -t master/build/lib/librte_ethdev/rte_flow.o |grep
rte_flow_dynf_metadata_offs
0000000000000000 l     F .text.startup    000000000000000a
_GLOBAL__sub_I_65535_0_rte_flow_dynf_metadata_offs
0000000000000620 g     O .data    0000000000000004 rte_flow_dynf_metadata_offs

The script noticed the symbol in text and is complaining about not
seeing it in text.experimental.
But well, this is a bug as we did not protect for expansion in:
if grep -q "\.text.*$SYM$" $DUMPFILE &&
        ! grep -q "\.text\.experimental.*$SYM$" $DUMPFILE

I suppose this would do the trick:
if grep -q "\.text.*[[:space:]]$SYM$" $DUMPFILE &&
        ! grep -q "\.text\.experimental.*[[:space:]]$SYM$" $DUMPFILE


- I don't think variables were considered in the rte_experimental
framework: seeing the __rte_experimental macro, I doubt it would work
for variables to be put in .text.
Do we need a different marker for those? or did I miss something else?
Neil?


Think we will have to live with this in 19.11...


-- 
David Marchand