Re: [PATCH linux-kselftest/test v6] lib/list-test: add a test for the 'list' doubly linked list

[Brendan Higgins]

On Wed, Oct 30, 2019 at 9:27 AM shuah <shuah@xxxxxxxxxx> wrote:
>
> On 10/30/19 4:42 AM, Dan Carpenter wrote:
> > On Wed, Oct 30, 2019 at 01:02:11AM -0700, David Gow wrote:
> >>> ERROR: that open brace { should be on the previous line
> >>> #869: FILE: lib/list-test.c:680:
> >>> +static void list_test_list_for_each_entry_reverse(struct kunit *test)
> >>> +{
> >>>
> >>>
> >>> I am seeing these error and warns. As per our hallway conversation, the
> >>> "for_each*" in the test naming is tripping up checkpatch.pl
> >>>
> >>> For now you can change the name a bit to not trip checkpatch and maybe
> >>> explore fixing checkpatch to differentiate between function names
> >>> with "for_each" in them vs. the actual for_each usages in the code.
> >>
> >> Thanks, Shuah.
> >>
> >> Yes, the problem here is that checkpatch.pl believes that anything
> >> with "for_each" in its name must be a loop, so expects that the open
> >> brace is placed on the same line as for a for loop.
> >>
> >> Longer term, I think it'd be nicer, naming-wise, to fix or work around
> >> this issue in checkpatch.pl itself, as that'd allow the tests to
> >> continue to follow a naming pattern of "list_test_[x]", where [x] is
> >> the name of the function/macro being tested. Of course, short of
> >> trying to fit a whole C parser in checkpatch.pl, that's going to
> >> involve some compromises as well.
> >
> > Just make it a black list of the 5 most common for_each macros.
> >
>
> How does black listing work in the context of checkpatch.pl?
>
> >>
> >> In the meantime, I'm sending out v7 which replaces "for_each" with
> >> "for__each" (adding the extra underscore), so that checkpatch is
> >> happy.
>
> This change is required just to quiet checkpatch and I am not happy
> about asking for this change. At the same time, I am concerned about
> git hooks failing on this patch.
>
> >
> > It's better to ignore checkpatch and other scripts when they are wrong.
> > (unless the warning message inspires you to make the code more readable
> > for humans).
> >
>
> It gets confusing when to ignore and when not to. It takes work to
> figure out and it is subjective.
>
> It would be great if we can consistently rely on a tool that is used as
> a criteria for patches to accept patches.

Agreed. I can see the point of not wanting to write an exception into
checkpatch for every exception of it's general rules; however, it
would be nice if there was a way to maybe have a special comment or
something that could turn off a checkpatch error. That way, a
checkpatch error/warning always means some action should be taken, and
if a rule is being ignored, there is always documentation as to why.

Otherwise, I don't feel strongly about this.

Cheers