Bugzilla Bug 200367

Hello,
New version, many features.
Thanks!

We will probably bump to BusyBox 1.7.4 and not pick up the 1.8.x series till
1.7.x is dead.

Can I ask why?
You can have them both?
You can even mask >=busybox-1.8.0
But have both in tree.

(In reply to comment #2)
> Can I ask why?

Limited time/resources. Two weeks ago BusyBox 1.8.0 was the (unstable) branch. 
Not much has changed in the past few weeks that really warrant it being 
declared a stable series.. It's also stated there were changes to the build 
system (which is where we get into time/resources) as it will probably take 
some time to fwd port any patches and confirm the build system all 
works properly.

If you however want to bump the ebuild locally and see if it generally 
works and put it in the tree p.masked that would help.

Oh... I would not have opened this unless I used it.
I just renamed the build and it works on my target.
I can probably make some more tests if you don't have time.

What I don't understand:
1. Why you disable the "uncommon options" overriding the user config file... I
guess this part should go before the restore config, or only if you don't use
external configuration.
2. Have you reported the busybox-1.7.0-bb.patch patch? Why is it required, and
not merged to upstream?
3. Also the sed of CFLAGS, have you tried to ask upstream to be able to specify
it externally so these sed will not be required?

Thanks!

Created an attachment (id=137033) [details]
busybox-1.8.2.ebuild.diff

Hello
Please review.

Changes:
1. Add uclibc USE flag so that busybox will be used AS-IS.
2. Convert the sed to remove flags into a patch, so people know if something
wrong/modified.
3. Set option by USE flags even if user had savedconfig, so that emerge
--newuse actually do something.

Upstream changes:
CFLAGS now can be specified externally.

Thanks,
Alon.

Created an attachment (id=137034) [details]
busybox-1.8.2-flags-strip.patch

OK.
Added as masked.
Thanks!

Did a quick test to cross compile to powerpc and
I had to do this:
--- busybox-1.8.2/Makefile.org  2007-11-29 14:58:53.000000000 +0100
+++ busybox-1.8.2/Makefile      2007-11-29 14:59:05.000000000 +0100
@@ -499,9 +499,6 @@
 # warn about C99 declaration after statement
 CFLAGS += $(call cc-option,-Wdeclaration-after-statement,)

-# disable pointer signedness warnings in gcc 4.0
-CFLAGS += $(call cc-option,-Wno-pointer-sign,)
-
 # Default kernel image to build when no specific target is given.
 # KBUILD_IMAGE may be overruled on the commandline or
 # set in the environment

To make it compile with my 3.4.6 gcc crosscompiler.
Seems like the test is performed with the hosts gcc rather that
the cross compiler.

Don't have time to investigate futher this ATM

Fixed. Thanks!

lets see ...
 - ARCH setting is pointless seeing as how it doesnt get used by the build
system
 - HOSTCC is wrong (tc-getBUILD_CC() exists for a reason)
 - dont know why the USE=uclibc check is there, it's clearly wrong
 - USE=savedconfig is supposed to trump all other USE flags, this is by design

Thank you for reviewing.

(In reply to comment #10)
> lets see ...
>  - ARCH setting is pointless seeing as how it doesnt get used by the build
> system

Looking at the build system, ARCH is auto generated using uname, if you cross
compile it should be overridden, right?

>  - HOSTCC is wrong (tc-getBUILD_CC() exists for a reason)

OK.

>  - dont know why the USE=uclibc check is there, it's clearly wrong

Until now, the ebuild removed --gc-sections in anycase. If I understand
correctly, the busybox is designed to be built with uclibc, why change upstream
settings if this is used?

>  - USE=savedconfig is supposed to trump all other USE flags, this is by design
> 

I was not aware of this. It is unexpected....
# It is not uncommon to come across a package which has a very fine
# grained level of configuration options that go way beyond what
# USE flags can properly describe.  For this purpose, a common API

Also:
 - - savedconfig   : Use this to restore your config from
/etc/portage/savedconfig ${CATEGORY}/${PN}. Make sure you USE flags allow for
approprate dependencies

From this I understand that USE flags do matter... Only finer grained level is
controlled by the savedconfig.

Please consider modifying this description, clearly saying that savedconfig
override USE flags.

Thanks!

please review the build system rather than just scanning/grepping.  ARCH, while
set, is never actually used.

the gc-sections removal is because *glibc* is broken.  uclibc is irrelevant.

as the documentation states, you should make sure the USE flags match the
config you created.  if you have say USE=-pam but your custom config file does
enable pam support and your system doesnt have pam installed, then yes things
will break.  however, this is your problem, not the ebuild's.