On Fri, Jun 18, 2021 at 4:30 AM Rong Chen [off-list ref] wrote:

On 6/7/21 3:19 PM, Arnd Bergmann wrote:

quoted

On Mon, Jun 7, 2021 at 8:27 AM Rong Chen [off-list ref] wrote:

quoted

On 6/5/21 7:08 PM, Arnd Bergmann wrote:

Thanks for the detailed explanation, and we have asked advice of
internal developers,
we'll enable COMPILE_TEST for randconfig, and check the failures if
COMPILE_TEST=n
in future.

Awesome, thanks a lot.

BTW is there any issue that found by randconfig with COMPILE_TEST=n
should be reported?

It's definitely possible, but I expect that to be rare enough that
investing your
time elsewhere is more helpful. If you do wish to keep testing this
configuration,
this is how I would approach it:

- Since you have a database of known warnings and errors, a good thing to try
  would be analyze it and figure out the types of warning that predominantly
 happen with COMPILE_TEST=n randconfigs, but still happen occasionally
 with COMPILE_TEST=y, as those are more likely real bugs.

- possibly annotate them to find which particular Kconfig option that
gets guarded
  by COMPILE_TEST causes which output. This should be possible because
  there are only a small number of such options.

- Run a small percentage of randconfig builds with COMPILE_TEST=n and the
  cross-check the output against the known classes of diagnostic output
  rather than individual lines.

       Arnd