On 08/10/2011 06:33 PM, David Woodhouse wrote:

On Wed, 2011-08-10 at 10:34 -0400, Arnaud Lacombe wrote:

quoted

your point being ? I might as well tell you that I find the current
behavior of 'all*.config' just as broken wrt. to dependency
management.

You might indeed. And I would find that commentary just as irrelevant
and unhelpful.

quoted

quoted

If you have nothing relevant to say, just don't say anything.

maybe you can come with a detailed description of your proposal's
behavior, including how to manage case like this, instead of just
throwing patch around ?

How's this for a definition:

"The behaviour for unsettable (or unclearable) options shall be exactly
like it already is if you put them in all*.config, or if you manually
edit the .config file and run 'make oldconfig', as people have been
doing for years. There is nothing new to see here."

quoted

If I do:

# make CONFIG_WIRELESS_EXT=y allnoconfig

I expect either a success or an error, not a silent discard. And
*yes*, the problem already exists with "all*.config".

[dwmw2@i7 linux-2.6]$ make CONFIG_WIRELESS_EXT=y allnoconfig
scripts/kconfig/conf --allnoconfig Kconfig
#
# Could not set CONFIG_WIRELESS_EXT=y; perhaps it has unmet dependencies?
#
#
# configuration written to .config
#

I understand that my question is indeed neither wanted nor clever, but 
what's the point of trying to support "make CONFIG_FOO=y"?

Will we be expected to type a 42 meters long command line to compile the 
kernel instead of doing a menuconfig in the foreseable future? (and 
between typos, unmet dependencies and the myriad of other possible 
errors, I'm not sure I'll get more free time).

I don't get it. If the goal is to help the kernel hackers and if it 
really helps them it might be a thing to do - it might prove useful for 
very simple CONFIG_ options but I'm not sure this will stay true for the 
general case.

Best regards,

-- Emmanuel