Re: [LTP] [PATCH] [RFC] lib/tst_test.c: Fix tst_brk() handling

[Cyril Hrubis <chrubis@suse.cz>]

Hi!
> > This makes the tst_brk() handling cleaner and saner as instead of
> > propagating the tst_brk() result in a return value an abort flag is
> > introduced into the shared memory.
> >
> > Now:
> >
> > - All the processes but the library one that reports the results exit
> >   with 0
> > - tst_brk(TBROK, ...) increments result conters, sets the abort flag.
> >   and exit current process
> > - all other tst_brk() variants will just increments the countes and
> >   exits the current process
> 
> It removes the easy way for parent to check that child hasn't run into
> any issues,
> but I can't recall a specific test we have today that depends on it.

I suppose that we can make the tst_brk flag part of a public API if
anyone needs that but I guess that in the case of tst_brk(TBROK, ...)
all we want is to make the test processes exit as soon as possible.

> > This makes the tst_brk() behavior well defined so we can now even call
> > tst_brk() with TFAIL and TPASS as well.
> 
> What's the use-case for it? Wouldn't it be more clear to just report
> TPASS + exit?

I think this makes actually the API more consistent. I.e. tst_res()
reports result and tst_brk() reports result and exits the current
process.

I think that we all carry a mental baggage that associates the tst_brk()
call with an error, but that is something we forced upon ourselves.
AFAIK it's short for tst_break, which itself only suggests that it does
exit the current process, similar to break being used in switch()
statement.

> > Open question (may be done in a separatep patch):
> >
> > Should tst_brk(TBROK, ...) apart from setting the flag also send sigkill
> > signal to the test process group to kill any leftover test processes?
> 
> Or heartbeat checking the abort flag and doing it from the library?

The heartbeat handler may be a good place to put this check into and I
was also thinking of adding checks to all SAFE_MACROS() and make them
something as cancelation points since they are supposed to exit the test
on a failure anyways.

But that could be only done once we have the flag in place and finally
have persistent way how to check that something went wrong.

-- 
Cyril Hrubis
chrubis@suse.cz

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp