CONFIG_* defines

[Rob Landley]

On Thursday 08 September 2005 02:54, Bernhard Fischer wrote:

> >Not a clue.  Haven't got 64-bit hardware to test with on this end. 
> > (Anybody got any ideas?)
>
> I was talking about the "(null)", sorry if i made that not clear.

Ah.  I didn't know you were talking about a purely cosmetic problem, I thought 
it was something my (functional) testing had missed.

Um...  Okay, confirmed it here.  That prints out in confdata.c line 330, which 
is printing out "str" from menu_get_prompt(menu). "str" is not getting set by 
menu_get_prompt()...

This isn't actually that it's failing to print out the "testing" line.  Look 
down a bit.  It's just printing out extra crud due to the first of the two 
sections I uncommented.  (I was overzealous in making sure that it actually 
_printed stuff out_, darn it...)

> >What _exactly_ do you do to generate this file?  (Make yesconfig?  make
> >menuconfig?  Make allyesconfig && make menuconfig?)
>
> I see this for all of allnoconfig defconfig allyesconfig menuconfig.

Did this cause any actual problems?

Hmmm...   I've noticed another problem, which is that certain symbols (like 
CONFIG_TEST) get written out twice.  However, looking at coreutils/Config.in 
that's because it's defined twice (510 and 514), albeit in mutually 
incompatible circumstances.  Let's see, which symbols have this problem...

make allnoconfig
sort .config | uniq -d

#
# CONFIG_DATE is not set
# CONFIG_FALSE is not set
# CONFIG_FEATURE_FANCY_ECHO is not set
# CONFIG_FEATURE_IFUPDOWN_IP_BUILTIN is not set
# CONFIG_FEATURE_IP_ADDRESS is not set
# CONFIG_FEATURE_IP_LINK is not set
# CONFIG_FEATURE_IP_ROUTE is not set
# CONFIG_FEATURE_IP_TUNNEL is not set
# CONFIG_TEST is not set
# CONFIG_TRUE is not set

Hmmm...  The problem is that if both are set to different values (which they 
should be in the "y" case being mutually exclusive), then whichever one comes 
last in the config file wins.  Which is not what we want, we want boolean 
or....

This config stuff is a bit of a pain, isn't it?

Hmmm.  I know how to fix it, but this introduces an n^2 algorithm.  We only 
have a little over 600 total configuration symbols right now so that's not 
bad, but still...

Thinking...

Rob