[LTP] [PATCH v2 1/1] lib: Multiply slow config only for a real tests

[Li Wang]

On Fri, Jan 17, 2025 at 5:16 PM Petr Vorel <pvorel@suse.cz> wrote:

> > On Fri, Jan 17, 2025 at 4:41 PM Petr Vorel <pvorel@suse.cz> wrote:
>
> TL;DR: wrong patch, please ignore it.
>
> ...
> > > +static inline int foo(unsigned int timeout)
>
>
> > What is the foo() used for?
>
> I'm sorry, I noticed this after sending as well. But got slow down due
> TST_NO_DEFAULT_MAIN in tst_test.c as well.
>
> > > +static inline int tst_multiply_on_slow_config(unsigned int timeout)
> > > +{
> > > +#ifndef TST_NO_DEFAULT_MAIN
> > > +       if (tst_has_slow_kconfig())
> > > +               timeout *= 4;
> > > +#endif /* TST_NO_DEFAULT_MAIN */
> > > +       return timeout;
> > > +}
>
>
> > But the tst_test.c has defined the TST_NO_DEFAULT_MAIN macro
> > so it will go complie with the second branch always.
>
> > IOW, the tst_has_slow_kconfig() will never be performed.
>
> Yes, you're right this would not work (it took me a while to find it as
> well).
>
> I hoped we would have some smart evaluation which would allow not having
> to add
> anything to files in testcases/lib/, but this will not work.
>
> We can either combine your approach with this (have a new definition +
> static
>

That won't works as well, define a new macro like TST_IGNORE_SLOW_KCONFIG
in testcase/lib/* is useless. Because the libltp.a is independent of other
things under
testcase/ dir. It has been decided when we link the libltp.a.

So eventually we have to go the way Andrea suggests.



> function in tst_test.h) or use struct tst_test flag as Andrea suggested.
> I'm not
> sure what is better, I slightly preferred the definition, because one day
> we
> might want to get rid of struct tst_test workarounds in testcases/lib
> therefore
> I would prefer not to add yet another.
>
> FYI tst_test struct use is forced by code in safe_clone() lib/tst_test.c:
>
> pid_t safe_clone(const char *file, const int lineno,
>                  const struct tst_clone_args *args)
> {
>         pid_t pid;
>
>         if (!tst_test->forks_child)
>                 tst_brk(TBROK, "test.forks_child must be set!");
>
> This could be also guarded by new definition. Then it should have probably
> a
> different name than TST_NO_SLOW_KCONFIG_CHECK. Sure, we postpone this
> cleanup
> after the release.
>
> BTW we have TCONF on starvation.c. This test would work if we don't
> prolong it
> even longer with tst_has_slow_kconfig(), thus might want to remove TCONF
> and
> disable tst_has_slow_kconfig() there as well. We can do it with both ways.
>
> Kind regards,
> Petr
>
>

-- 
Regards,
Li Wang