MI: reinstate nvme_mi_init_ep for endpoint initialisation in test

[jk-ozlabs]

Ideally, we would be using the actual implementation of
nvme_mi_init_ep() for our tests, rather than open-coding it in the test
init.

This change exports the nvme_mi_init_ep from libnvme-mi.so, but it
remains excluded from the headers, as it's only intended for use in the
transport API. This means we can call it from the tests, but keep is
somewhat-internal.

We put this into a specific _TEST section of the version script, to keep
separate from the public symbols.

This prevents us from diverging the endpoint init process in our
testcases.

Signed-off-by: Jeremy Kerr jk@codeconstruct.com.au

[codecov-commenter]

Codecov Report

Merging #400 (76a7472) into master (b7fe911) will increase coverage by 0.06%.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##           master    #400      +/-   ##
=========================================
+ Coverage    7.61%   7.68%   +0.06%     
=========================================
  Files          29      29              
  Lines        4414    4413       -1     
  Branches      828     828              
=========================================
+ Hits          336     339       +3     
+ Misses       3937    3933       -4     
  Partials      141     141              

Impacted Files	Coverage Δ
test/mi.c	67.10% <100.00%> (-0.22%)	⬇️
src/nvme/mi.c	35.29% <0.00%> (+1.56%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b7fe911...76a7472. Read the comment docs.

[igaw]

Do you want to add the crc32 method as well? That would make my coverage hack go away :)

[igaw]

And please add a fat comment in the map file to warn anyone depending on this symbol!

[jk-ozlabs]

OK, will add crc32 too, good plan. Means we'll be testing more of the actual code.

[igaw]

Just got another idea: What about just have a separate library linked for testing only?

libnvme_mi_test = library(
   [...]
   link_args: ['-Wl,--version-script=' + mi_version_script_arg],
   [...]
)

and only have the internal symbols exposed in libnvme_mi_test (updating the mi_version_script_arg)

[jk-ozlabs]

Yep, could do - we could even drop --version-script for the libnvme-mi-test.so build, making all symbols global, perhaps.

This is all conditional on the coverage getting detected across the two builds though, I guess. Any reliable way to test this?

[igaw]

I think you run

rm -rf obj && meson setup obj -Db_coverage=true --werror && meson test -C obj && ninja -C obj coverage --verbose

locally you should see the coverage report. I assume it will work as the source files match?

[jk-ozlabs]

Yep, that seems to work well. Proposed in #403

[igaw]

Cool thanks!