Message-ID: <20130129051321.GB3283@leaf>
Date:	Tue, 29 Jan 2013 16:13:21 +1100
From:	Josh Triplett <josh@...htriplett.org>
To:	Dmitry Torokhov <dmitry.torokhov@...il.com>
Cc:	Joe Millenbach <jmillenbach@...il.com>, Greg KH <greg@...ah.com>,
	Stephen Rothwell <sfr@...b.auug.org.au>,
	linux-next@...r.kernel.org, linux-kernel@...r.kernel.org,
	Brian Swetland <swetland@...gle.com>,
	"Mike A. Chan" <mikechan@...gle.com>,
	Sheng Yang <sheng@...ux.intel.com>,
	Yunhong Jiang <yunhong.jiang@...el.com>,
	Xiaohui Xin <xiaohui.xin@...el.com>,
	Jun Nakajima <jun.nakajima@...el.com>,
	Bruce Beare <bruce.j.beare@...el.com>,
	Tom Keel <thomas.keel@...el.com>,
	Alan Cox <alan@...ux.intel.com>,
	Jamey Sharp <jamey@...ilop.net>
Subject: Re: linux-next: manual merge of the tty tree with the input tree

On Mon, Jan 28, 2013 at 04:23:57PM -0800, Dmitry Torokhov wrote:
> On Tue, Jan 29, 2013 at 10:59:17AM +1100, Josh Triplett wrote:
> > On Mon, Jan 28, 2013 at 02:44:43PM -0800, Dmitry Torokhov wrote:
> > > On Mon, Jan 28, 2013 at 02:09:31PM -0800, Joe Millenbach wrote:
> > > > On Mon, Jan 28, 2013 at 9:33 AM, Dmitry Torokhov
> > > > <dmitry.torokhov@...il.com> wrote:
> > > > > On Mon, Jan 28, 2013 at 06:46:15AM -0800, Greg KH wrote:
> > > > >> On Mon, Jan 28, 2013 at 08:44:24PM +1100, Stephen Rothwell wrote:
> > > > >> > Hi Greg,
> > > > >> >
> > > > >> > Today's linux-next merge of the tty tree got a conflict in
> > > > >> > drivers/input/keyboard/Kconfig between commit 6f2ac009f29b ("Input:
> > > > >> > goldfish - virtual input event driver") from the input tree and commit
> > > > >> > 4f73bc4dd3e8 ("tty: Added a CONFIG_TTY option to allow removal of TTY")
> > > > >> > from the tty tree.
> > > > >> >
> > > > >> > I fixed it up (see below - I am not sure if GOLDFISH_EVENTS needs TTY or
> > > > >> > not) and can carry the fix as necessary (no action is required).
> > > > >> >
> > > > >> > --
> > > > >> > Cheers,
> > > > >> > Stephen Rothwell                    sfr@...b.auug.org.au
> > > > >> >
> > > > >> > diff --cc drivers/input/keyboard/Kconfig
> > > > >> > index 078305e,008f96a..0000000
> > > > >> > --- a/drivers/input/keyboard/Kconfig
> > > > >> > +++ b/drivers/input/keyboard/Kconfig
> > > > >> > @@@ -479,16 -482,8 +482,18 @@@ config KEYBOARD_SAMSUN
> > > > >> >       To compile this driver as a module, choose M here: the
> > > > >> >       module will be called samsung-keypad.
> > > > >> >
> > > > >> > + if TTY
> > > > >> > +
> > > > >> >  +config KEYBOARD_GOLDFISH_EVENTS
> > > > >> >  +  depends on GOLDFISH
> > > > >> >  +  tristate "Generic Input Event device for Goldfish"
> > > > >> >  +  help
> > > > >> >  +    Say Y here to get an input event device for the Goldfish virtual
> > > > >>
> > > > >> Looks good, thanks.
> > > > >
> > > > > Greg,
> > > > >
> > > > > Please drop 4f73bc4dd3e8563ef4109f293a092820dff66d92, at least the parts
> > > > > related to input. As far as I know nothing except serport driver
> > > > > depends on tty and we do not need to introduce this kind of
> > > > > dependencie. Anyone needing slim config can simply try disabling
> > > > > input (or parts of it) without needing an artificial dependencies.
> > > > >
> > > > > Thanks.
> > > > >
> > > > > --
> > > > > Dmitry
> > > > 
> > > > Dmitry and Greg,
> > > > 
> > > > SERIO needs TTY, 
> > > 
> > > No it does not. There is only one (1) serio driver that needs the tty
> > > layer and that is serport.
> > > 
> > > > and the majority of the input changes are adding "depends on TTY" to
> > > > things that "select SERIO" as they break the dependency chain.  In
> > > > other words, enabling a component that selects SERIO will turn SERIO
> > > > on even when SERIO depends on TTY and TTY is disabled.
> > > 
> > > Except that it does not. Are you confusing SERIO with SERIAL by any
> > > chance?
> > 
> > A few serial drivers don't actually need the TTY layer.  However, most
> 
> Can you please tell me why you are talking about serial layer here?

Probably terminology sloppiness.  Insert noun for "things that depend on
SERIO" here.

> > do, including many that don't obviously appear to at first glance.  For
> > instance, MOUSE_PS2 doesn't *appear* to need TTY, but without the
> > dependency, having MOUSE_PS2 enabled and TTY disabled produces a kernel
> > that doesn't build.
> 
> Compile log please.

http://marc.info/?l=linux-kernel&m=134555498507747&w=1

IIRC, produced by dropping the "depends on TTY" from MOUSE_PS2, enabling
MOUSE_PS2, leaving TTY disabled, and changing nothing else.  (In
particular, not changing anything related to serport directly.)

> >  (Also keep in mind that many other headers include
> > <linux/tty.h>.)  Many of the drivers that don't actually need TTY
> > nonetheless won't typically appear on systems small enough to want to
> > compile out TTY.
> > 
> > In any case, it seems simple enough to whittle down dependencies on TTY
> > later on, but for a first pass, the conserative approach seems
> > preferable.
> 
> However my approach would be not to touch anything except serport
> driver (which indeed depends on TTY layer).

Quite a bit more than that depends on the TTY layer.

> > > In the future it would be nice if you CCed people involved in the
> > > subsystem you are changing.
> > 
> > Such as the maintainers of the TTY and serial layers?  Alan Cox OKed
> > this conservative addition of dependencies in the original version of
> > this change, and Greg had no complaints at the time.
> 
> Maintainer of _SERIO_ (not SERIAL, SERIO) layer, yours truly.

This change affected the dependencies of several hundred drivers, and
the output of get_maintainer.pl included hundreds of email addresses.
Spamming *all* of them for a one-line change to the dependencies of a
Kconfig option, when that change acts as a no-op except on systems
trying to use CONFIG_TTY=n, seemed imprudent.  Given that this was a
change to the TTY layer, and not to the SERIO subsystem other than its
dependencies (along with the dependencies of hundreds of other drivers),
SERIO didn't particularly seem to stand out here as one to keep the CC
for.  In any case, sorry you didn't hear about this sooner.

Given that the kernel without this patch effectively has CONFIG_TTY=y
hardcoded, and that CONFIG_TTY=y remains the common case, it seems
fairly reasonable to bias in favor of making sure that all
Kconfig-supported configurations will actually build rather than trying
to maximize the number of drivers buildable with CONFIG_TTY=n.  It seems
easy enough to drop dependencies on TTY later, but at the moment all the
dependencies on TTY came from either observed build failures without it
or fairly clear dependencies on TTY (such as the use of tty_* functions,
inclusion of linux/tty.h, or selection of a driver that does one of
those).

- Josh Triplett
--
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