[RFC] CONFIG_PM_LEGACY = compiler warnings for deprecation ... solution?

[RFC] CONFIG_PM_LEGACY = compiler warnings for deprecation ... solution?

[Scott Thompson]

w/ CONFIG_PM_LEGACY .config option set, you get following batch of
compile errors because it marks the functions as 'deprecated'. w/o 
it set, it just uses stubs. compiler warnings for refernece (from 
2.6.23-1, though appears to have been around since at least 2.6.11 
from another mail list post).

However, since these functions are required if you have
CONFIG_PM_LEGACY option marked...

I propose that within the include/linux/pm_legacy.h file, functions
in the CONFIG_PM_LEGACY aren't marked 'deprecated' and functions
outside the #if are marked deprecated.  Then the handful of calls
(12 or so) that actually call the pm_legacy apis can get wrapped in
the CONFIG_PM_LEGACY flag.

This is basically already being done in one case since as of
2.6.20.1 the pm_register function is only used in
drivers/serial/68328serial.c.

The pm_send_all is not wrapped around all uses yet, but that is
pretty easy to patch.  Then the compile will actually complain
should someone try to call it and not wrap w/ the CONFIG_PM_LEGACY
ifdef.

Happy to hear thoughts or other solutions.

------------------------------
Regards,
Scott Thompson / postfail@hushmail.com
------------------------------

--
Free information on vast array of home improvement projects.  Click here.
http://tagline.hushmail.com/fc/Ioyw6h4eNIBTzslyd6sFcem0DGn1xrajHGmTczUmHzuMeqA2wEMUHS/

Re: [RFC] CONFIG_PM_LEGACY = compiler warnings for deprecation ... solution?

[Rafael J. Wysocki]

On Wednesday, 1 August 2007 07:08, Scott Thompson wrote:
> w/ CONFIG_PM_LEGACY .config option set, you get following batch of
> compile errors because it marks the functions as 'deprecated'. w/o 
> it set, it just uses stubs. compiler warnings for refernece (from 
> 2.6.23-1, though appears to have been around since at least 2.6.11 
> from another mail list post).
> 
> However, since these functions are required if you have
> CONFIG_PM_LEGACY option marked...
> 
> I propose that within the include/linux/pm_legacy.h file, functions
> in the CONFIG_PM_LEGACY aren't marked 'deprecated' and functions
> outside the #if are marked deprecated.  Then the handful of calls
> (12 or so) that actually call the pm_legacy apis can get wrapped in
> the CONFIG_PM_LEGACY flag.
> 
> This is basically already being done in one case since as of
> 2.6.20.1 the pm_register function is only used in
> drivers/serial/68328serial.c.
> 
> The pm_send_all is not wrapped around all uses yet, but that is
> pretty easy to patch.  Then the compile will actually complain
> should someone try to call it and not wrap w/ the CONFIG_PM_LEGACY
> ifdef.
> 
> Happy to hear thoughts or other solutions.

We are going to remove the legacy PM altogether, AFAICS, although there still
are some obstacles.

Greetings,
Rafael


-- 
"Premature optimization is the root of all evil." - Donald Knuth

Re: [RFC] CONFIG_PM_LEGACY = compiler warnings for deprecation ... solution?

[Scott Thompson]

On Wed, 01 Aug 2007 06:57:25 -0400 "Rafael J. Wysocki" 
<rjw@sisk.pl> wrote:
>We are going to remove the legacy PM altogether, AFAICS, although 
>there still
>are some obstacles.
>

Thanks -- I had checked Documentation/feature_removal_schedule.txt 
for this, but looks like the obsoletion is referenced in 
Documentation/pm.txt.

I'll offer to patch in the solution I proposed to prevent the 
compile warnings in the interim if you're interested...

---------------------------------------
Scott Thompson / postfail@hushmail.com
---------------------------------------

--
Click to get 150,000 in student loans and find the lowest rates
http://tagline.hushmail.com/fc/Ioyw6h4d7Femgqznt3WSEFC1KSqiZZ3eudvCzwr9tOXpy1LdMQYzRK/