[U-Boot] [PATCH] config.mk: Add -Wundef to CFLAGS

[Albert ARIBAUD]

Hi Simon,

On Thu, 3 Oct 2013 17:49:23 -0600, Simon Glass <sjg at chromium.org> wrote:

> Hi,
> 
> 
> On Wed, Oct 2, 2013 at 1:27 PM, Albert ARIBAUD <albert.u.boot at aribaud.net>wrote:
> 
> > Hi Masahiro,
> >
> > On Wed, 21 Aug 2013 13:33:19 +0900, Masahiro Yamada
> > <yamada.m at jp.panasonic.com> wrote:
> >
> > > Hello, Albert and U-Boot developers.
> > >
> > >
> > > The current status of this patch is Changes Requested.
> > >
> > > I love -Wundef option to be in, but it looks like
> > > difficult for me to post the version 2.
> > >
> > >
> > > The first choice to meet Albert's requirement is
> > >
> > > > Therefore I ask:
> > > >
> > > > - that this patch be submitted along fixes to build failures it
> > > >   causes, as a proper patch series, by a single individual,
> > >
> > > Sorry, I cannot do this because:
> > >
> > > I am not familiar with architectures other than ARM.
> > > I understand only a few devices.
> > > To fix warnings in a correct way, a close look is often needed,
> > > but I cannot cover the whole code in the U-Boot tree.
> > >
> > > If possible, could anyone take over this task?
> > >
> > >
> > >
> > > The other option is
> > >
> > > >   collected by someone in an officially created git repo or branch;
> > >
> > > OK, I can do this.
> > > But I am not sure this will go well.
> > >
> > > Even if I create a new repo u-boot-wundef,
> > > how many people will pay attention to this repository?
> > >
> > > Most of users/developers track upstream repos
> > > where -Wundef warnings are never displayed.
> > >
> > > This means no one will have the motivation to fix the warnings.
> > >
> > >
> > >
> > > If this patch is desired, in which way should we continue?
> > > Comments are welcome.
> >
> > Sorry not to have followed up earlier.
> >
> > As I said, I want fixes for trivial cases -- cases where, for instance,
> > a macro is used in an #if which has absolutely *no* definition in the
> > whole codebase. I do not want fixes for all cases.
> >
> > OTOH, to find out which failures would be trivial to fix and which ones
> > would not, you'd have to go through all of them, which could be
> > time-consuming, depending on the number of targets.
> >
> > I thus suggest we use the typical U-Boot strategy: right at the
> > beginning of the merge period, we apply the -Wundef patch onto all
> > repos (or on the main repo and then pulled into others) and then wait
> > for screams.
> >
> > Screams should come fst from custodians, who routinely build for all
> > targets of their assigned architecture. They will see which boards fail
> > due to undefined macros and will report those failures on the list,
> > copying the board maintainers (or subsystem owners if the issue is not
> > board-specific.
> >
> > All boards not fixed before merge window closure release will be
> > declared dying; all those not fixed before 2014.01 will be declared
> > dead, and orphaned and put in the scrapyard (and then, people who want
> > them back can always resurrect *and fix* them).
> >
> > Subsystems... will have to be fixed some way or other.
> >
> > Of course, anyone with an interest can spontaneously provide fixes for
> > boards or subsystems affected.
> >
> > Adding Tom and Wolfgang for advice, but of course comments are welcome
> > from everyone.
> >
> 
> Breaking a board by introducing a warning is bad, I don't think we can do
> that.

The idea *is* to break the board so that users or maintainers realize it
builds on an undefined macro preprocessor symbol -- or so that we
custodians realize no one is interested enough in that board to fix
it, which is a sign that it should be removed from the code base. We've
done such a thing before, actually, when a change required attention
from a lot of maintainers.

> But fixing common.h would be easy enough at least.

That can be done for cases where the missing symbol should indeed exist.
When it should not exist any more, nothing can fix it but removing the
code that uses the zombie symbol. And in both cases, breaking the build
with a warning is the best way to find and resolve the issue. 

> Can you run buildman on all boards and see how many filesgenerate warnings?

I assume this request was directed at Masahiro Yamada.

I second the request: if only a few boards are affected, then instead
of a system-wide change plus breaks, we could apply my first proposal,
that is, Masahiro asking maintainers for changes and coordinating
them in a branch, then a series.

> Regards,
> Simon

Amicalement,
-- 
Albert.