Message-ID: <1428783814.17822.150.camel@x220>
Date:	Sat, 11 Apr 2015 22:23:34 +0200
From:	Paul Bolle <pebolle@...cali.nl>
To:	Stefan Hengelein <stefan.hengelein@....de>
Cc:	Gregory Fong <gregory.0xf0@...il.com>,
	Michal Marek <mmarek@...e.cz>,
	Valentin Rothberg <valentinrothberg@...il.com>,
	Andreas Ruprecht <rupran@...server.de>,
	Martin Walch <walch.martin@....de>,
	linux-kbuild@...r.kernel.org,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 1/2] kconfig: Print full defined and depends for
 multiply-defined symbols

On Sat, 2015-04-11 at 21:58 +0200, Stefan Hengelein wrote:
> 2015-04-11 20:56 GMT+02:00 Paul Bolle <pebolle@...cali.nl>:
> > On Sat, 2015-04-11 at 18:36 +0200, Stefan Hengelein wrote:
> What i meant to say, you won't get a prompt (or for mconf, won't see
> it in the menu) if THUMB2_KERNEL is disabled, FRAME_POINTER will
> simply be enabled when the default condition in the definition without
> the prompt is satisfied.
> 
> Therefore it might be misleading to add it to the conditions.

That's a NAK to this patch, isn't it?

> >> I personally would prefer to
> >> additionally find the second definition that doesn't have a prompt and
> >> other dependencies instead of adding them to the first entry, but
> >> that's just my personal preference.
> >
> > I notice myself getting rather grumpy. (That usually translates to:
> > "Drop it, and revisit in a few days".) Let me explain.
> >
> > This is the arm64 entry:
> >     config FRAME_POINTER
> >             bool
> >             default y
> >
> > This is the hexagon entry
> >     config FRAME_POINTER
> >             def_bool y
> >
> > This is the m32r entry:
> >     config FRAME_POINTER
> >             bool "Compile the kernel with frame pointers"
> >             help
> >               If you say Y here [...]
> >
> > And this is the sparc entry:
> >     config FRAME_POINTER
> >             bool
> >             depends on MCOUNT
> >             default y
> >
> > You'd expect these entries to yield really simple results when doing a
> > search in menuconfig. But the results show unparseable crap[1]. (And I'm
> > afraid Gregory's patch would make that even worse. Gregory: please prove
> > me wrong.)
> 
> would you please define unparseable crap?

This is what I see for m32r:
    Depends on: DEBUG_KERNEL [=y] && (CRIS || M68K || FRV || UML || AVR32 || SUPERH || BLACKFIN || MN10300 || METAG) || [...] 
    Selected by: FAULT_INJECTION_STACKTRACE_FILTER [=n] && FAULT_INJECTION_DEBUG_FS [=n] && STACKTRACE_SUPPORT && !X86_6[...]

No one is going to understand what that means. (Did I say I was grumpy?)
Sure, it might be actually correct for most architectures. But it
resembles in no way what one expects to see after reading just the m32r
entry.

>  the only odd thing i notice
> when i call menuconfig on hexagon is a really long "Selected by: "
> list

Yes. That list makes no sense whatsoever. (Did I say I was grumpy?)

> > So to the grumpy me it looks like either:
> > - menuconfig handles these redefinitions incorrectly in its UI;
> > - these redefinitions are actually complicated (as in: somehow they
> > concatenate the dependencies, etc.) and we should probably disallow
> > them. Because otherwise looking at a Kconfig entry tells you very little
> > about what is actually going on for the architecture you're interested
> > in.
> >
> > What is the grumpy me missing here?
> 
> Redefinitions are more of an "overwrite" than a "add conditions to the entry".

That's again a NAK to this patch, isn't it?

> It's perfectly reasonable for architecture A to say: if these
> conditions hold, i want to enable option B, not matter what the
> Kconfigfile in lib/ says (like arm64 does with FRAME_POINTER, it is
> always on, (depending on if there are other dependencies around it)).
> 
> Redefinitions are a little more complicated...
> If you have two options with the same symbol and both have a prompt,
> you will see it two times in conf. Meaning, Kconfig doesn't merge both
> declarations but they are separate two different instructions,
> affecting the same symbol.

You lost me there.

> With menuconfig it's the same, it will show both definitions in the
> menu, they might however be in another submenu, depending on the
> dependencies both definitions have.
> 
> The kconfig rules state only one definition should have a prompt, but
> as you can see, m32r does violate this "rule" and it doesn't break
> kconfig ;)
> 
> That's why i said i'd prefer to have both declarations printed than
> adding the conditions from the second definition to the printed entry
> of the first....
> If the order is different, you might only see the definition without
> the prompt (what happens for hexagon) and miss the second possibility
> to enable the feature.

I'm beyond confused now. (But happy to have dragged you into this
discussion. I think we're making progress.)

I'd really prefer things to be simpler: how is anyone reading the
Kconfig entries I quoted going to realize all that?


Paul Bolle

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists