svn 23941 (comments on config symbol renaming).

[Denys Vlasenko]

On Friday 07 November 2008 19:35, Rob Landley wrote:
> Way way back in the dark ages, the original reason for naming symbols 
> CONFIG_APPLETNAME and CONFIG_FEATURE_APPLETNAME_BLAH was so that so that 
> things like "make allbareconfig" could could easily filter out features via a 
> sed invocation looking for "CONFIG_FEATURE".
> 
> This was a build test, because a few times we found build breaks when a 
> feature was _disabled_, because we tested allyesconfig and allnoconfig but 
> couldn't easily test enabling each applet but switching each sub-feature 
> _off_.

I have a script to generate and build test random configs
(in svn, scripts/randomtest). I use it to weed off broken configs
before each release.

> So I added "make allbareconfig" to make sure each applet feature  
> would at least compile with all its features disabled.  That build option 
> seems to have gone away since 1.2, any idea why?

Probably fell victim of build system rewrite.

> Ignoring that constraint, it does make more sense for config features to be 
> named CONFIG_APPLETNAME_SUBFEATURE, and that's what I did with toybox.  (I 
> also named global features "CONFIG_TOYBOX_SUBFEATURE"; again, easy to 
> spot 'em with config editing tools if you're consistent in the naming.)
> 
> Busybox had CONFIG_DEBUG_ as a prefix when defconfig was made via sed, because 
> all those needed to be filtered out.  These days, you can just as easily grep 
> for "_DEBUG" if you want to filter out debug features that way, and the 
> kconfig infrastructure's been updated to accept a driedfrogpills file to 
> switch the debug features off that way.  (It's sort of the opposite of a 
> miniconfig file, supplementing allyesconfig with symbols to be switched 
> _off_.)
> 
> I don't really know what the plans are for the config infrastructure.  (It 
> seems lots of the customizations we had got yanked out when it was resynced 
> with the upstream kernel infrastructure?)

Yes. A few got restored since initial breakage. libbusybox etc.

> Are things like allbareconfig still interesting?

I fear this may double the bandwidth of "our XXXXconfig is wrong"
discussions on the ml. :)
--
vda