Why did oldconfig's behavior change in 2.6.15-rc1?

Why did oldconfig's behavior change in 2.6.15-rc1?

[Rob Landley]

On Saturday 12 November 2005 16:56, Rob Landley wrote:
> Linus says if we're going to test something, test -rc1, so I did.
>
> It went boing.
>
> I'm still trying to get -skas0 working on x86-64, but this was a standard
> x86 build...
>
> Rob

Very, very strange:

> make ARCH=um allnoconfig
> cat >> .config << EOF
CONFIG_MODE_SKAS=y
CONFIG_BINFMT_ELF=y
CONFIG_HOSTFS=y
CONFIG_SYSCTL=y
CONFIG_STDERR_CONSOLE=y
CONFIG_UNIX98_PTYS=y
CONFIG_BLK_DEV_LOOP=y
CONFIG_BLK_DEV_UBD=y
CONFIG_TMPFS=y
CONFIG_SWAP=y
CONFIG_LBD=y
CONFIG_EXT2_FS=y
CONFIG_PROC_FS=y
EOF
> make ARCH=um oldconfig
> grep SKAS .config
# CONFIG_MODE_SKAS is not set

Why did oldconfig switch off CONFIG_MODE_SKAS?  It didn't do that before.  
Hmmm...  Rummage, rummage...  Darn it, it's position dependent.  _And_ 
version dependent.

Ok, now I have to put the new entries at the _beginning_.  Appending them 
doesn't work anymore, it now ignores any symbol it's already seen, so you 
can't easily start with allnoconfig, switch on just what you want, and expect 
oldconfig to do anything intelligent.

That kinda sucks.  Oh well, I can have sed rip out the old symbols before I 
append the new ones.  Here's hoping it's not _that_ position dependent...

Rob

Re: Why did oldconfig's behavior change in 2.6.15-rc1?

[Sam Ravnborg]

On Sat, Nov 12, 2005 at 05:31:25PM -0600, Rob Landley wrote:
> On Saturday 12 November 2005 16:56, Rob Landley wrote:
> > Linus says if we're going to test something, test -rc1, so I did.
> >
> > It went boing.
> >
> > I'm still trying to get -skas0 working on x86-64, but this was a standard
> > x86 build...
> >
> > Rob
> 
> Very, very strange:
> 
> > make ARCH=um allnoconfig
> > cat >> .config << EOF
> CONFIG_MODE_SKAS=y
> CONFIG_BINFMT_ELF=y
> CONFIG_HOSTFS=y
> CONFIG_SYSCTL=y
> CONFIG_STDERR_CONSOLE=y
> CONFIG_UNIX98_PTYS=y
> CONFIG_BLK_DEV_LOOP=y
> CONFIG_BLK_DEV_UBD=y
> CONFIG_TMPFS=y
> CONFIG_SWAP=y
> CONFIG_LBD=y
> CONFIG_EXT2_FS=y
> CONFIG_PROC_FS=y
> EOF
> > make ARCH=um oldconfig
> > grep SKAS .config
> # CONFIG_MODE_SKAS is not set
> 
> Why did oldconfig switch off CONFIG_MODE_SKAS?  It didn't do that before.  
> Hmmm...  Rummage, rummage...  Darn it, it's position dependent.  _And_ 
> version dependent.
> 
> Ok, now I have to put the new entries at the _beginning_.  Appending them 
> doesn't work anymore, it now ignores any symbol it's already seen, so you 
> can't easily start with allnoconfig, switch on just what you want, and expect 
> oldconfig to do anything intelligent.
> 
> That kinda sucks.  Oh well, I can have sed rip out the old symbols before I 
> append the new ones.  Here's hoping it's not _that_ position dependent...

A much better way would be to put the values in a file named:
allno.config

With latest kconfig changes this will do the trick, and you will have a
valid config no matter what you put in.

	Sam

Re: Why did oldconfig's behavior change in 2.6.15-rc1?

[Roman Zippel]

Hi,

On Sat, 12 Nov 2005, Rob Landley wrote:

> Why did oldconfig switch off CONFIG_MODE_SKAS?  It didn't do that before.  
> Hmmm...  Rummage, rummage...  Darn it, it's position dependent.  _And_ 
> version dependent.

It's _not_ position dependent, but the behaviour with multiple equal 
symbols is undefined.

> Ok, now I have to put the new entries at the _beginning_.  Appending them 
> doesn't work anymore, it now ignores any symbol it's already seen, so you 
> can't easily start with allnoconfig, switch on just what you want, and expect 
> oldconfig to do anything intelligent.

Now you can put them in allno.config instead and allnoconfig will do the 
right thing.

bye, Roman

Re: [uml-devel] Re: Why did oldconfig's behavior change in 2.6.15-rc1?

[Rob Landley]

On Saturday 12 November 2005 17:49, Roman Zippel wrote:
> Hi,
>
> On Sat, 12 Nov 2005, Rob Landley wrote:
> > Why did oldconfig switch off CONFIG_MODE_SKAS?  It didn't do that before.
> > Hmmm...  Rummage, rummage...  Darn it, it's position dependent.  _And_
> > version dependent.
>
> It's _not_ position dependent, but the behaviour with multiple equal
> symbols is undefined.

I just want a minimally expressed config.  I've jumped through a number of 
hoops trying to get "allnoconfig plus these symbols" to work.

Setting .config to my symbol list and doing old config prompts a lot, doing 
yes "" | oldconfig sets lots of crap by default (since not all default values 
are n), and doing yes "n" | oldconfig goes into an endless loop every time 
it's prompting for a number or some such.  You have to start with 
allnoconfig.  Appending the new symbols and re-running oldconfig worked for a 
number of versions, until now.  Now you have to insert the new symbols at the 
beginning.

I could presumably also work around the new breakage via following allnoconfig 
with a sed -i invocation to remove the appropriate "# blah is not set" lines 
before appending the symbols.  (Of course it means that the config symbol 
list has to be fed into a for loop instead of being expressed in the same 
format as config, which sucks deeply.  Insert at the beginning works until it 
changes again...)

> > Ok, now I have to put the new entries at the _beginning_.  Appending them
> > doesn't work anymore, it now ignores any symbol it's already seen, so you
> > can't easily start with allnoconfig, switch on just what you want, and
> > expect oldconfig to do anything intelligent.
>
> Now you can put them in allno.config instead and allnoconfig will do the
> right thing.

Testing...

Ok, that worked.  Weird and totally unituitive name for a miniconfig, but oh 
well.

Is this documented anywhere?

Rob