[LTP] [RFC] shell wrappers for tst_checkpoint

[Cyril Hrubis <chrubis@suse.cz>]

Hi!
> There is a need to use the tst_checkpoint interface from shell, but
> we don't have wrappers for it (yet).
> 
> Patch 1 of the series contains one possible implementation for that,
> patch 2 - an example of usage in the context of the memcg_functional test case.
> 
> I'd like to get some feedback from LTP users.
> 
> My primary concern is about two issues:
> 
> 1. The new test API for C is cool and takes the responsibility on
>    maintaining the infrastructure for tst_checkpoint. However, I couldn't find
>    a way to implement something similar in shell, so I switched back to using
>    two separate functions for that - TST_CHECKPOINT_SETUP, TST_CHECKPOINT_CLEANUP.
> 
>    There may be a better way...

Well we can always put the cleanup into the tst_exit() which is executed
both on clean and unclean exit (tst_brkm). And the setup could be done
right after the checks for TCID and TST_TOTAL. Both supposedly if
something as TST_NEEDS_CHECKPOINTS=1 has been set before sourcing
test.sh.

Also nothing stops us from bringing the shell API closer to what we have
for C :).

We would have to source the test library script after we had defined all
the setup() cleanup() and test() functions and after setting the
corresponding TST_ID=foo, TST_NEEDS_CHECKPOINTS=1, etc. But as far as I
can tell it seems doable.

And it would be really cool if we managed to use the shared memory for
test results as well so that result from child processes (subshells)
would be propagated automatically as well.

> 2. What is the best location for the new supplemental binaries (tst_checkpoint_wait,
> tst_checkpoint_wake)? tools/apicmds/ltpapicmd.c or separate source files in
> testcases/lib/ (similar to tst_sleep)?

I would go for separate binary in testcases/lib/ rather than adding to
the ltpapicmd.c which should really be removed at some point in future,
once there are no users for the binary tst_resm.

-- 
Cyril Hrubis
chrubis@suse.cz