Re: [PATCH] can: rockchip_canfd: Drop obsolete dependency on COMPILE_TEST

[Jean Delvare <jdelvare@suse.de>]

Hi Geert,

On Fri, 22 Nov 2024 15:33:21 +0100, Geert Uytterhoeven wrote:
> On Thu, Nov 21, 2024 at 2:50 PM Jean Delvare <jdelvare@suse.de> wrote:
> > Your reasoning would hold only if building a limited, maybe
> > not-working driver, was a purpose in itself. I personally can't see
> > any value in doing this.  
> 
> It may help detecting more configuration issues using randconfig.

Maybe I would agree if build time, energy and engineering time were
free and unlimited resources.

In our world though, the time spent by a test build farm to build a
known-crippled driver would be better used to start the next test build
earlier. This makes for better quality coverage. Or the energy used to
do that could also be saved altogether.

And this suboptimal use of build time and energy is only the best case.
The worst case is if the randconfig case causes a "false positive" build
error or warning. This will generate a report, which will be sent to
mailing lists, read by human beings and investigated, wasting
engineering time.

I'm not making this up, this is in fact exactly what happened and led
me to investigate a first "depends on OF || COMPILE_TEST" case a few
years ago. I found out that the build warning would only possibly
trigger if COMPILE_TEST was selected and OF was not. It would never
trigger with any configuration options combination not including
COMPILE_TEST. Fixing it would require adding #ifdefs and using macros.
But why should we make the code more complex, less readable, to prevent
a warning which can never happen in a non-test scenario? And more
importantly, why should we allocate engineering time to look into such
issues, when there are so many more useful tasks to work on?

-- 
Jean Delvare
SUSE L3 Support