Patchwork [v2,0/1] busybox: update to 1.19.3

login
register
mail settings
Submitter Anders Darander
Date Nov. 22, 2011, 2:34 p.m.
Message ID <cover.1321972428.git.anders@chargestorm.se>
Download mbox
Permalink /patch/15249/
State Superseded
Headers show

Pull-request

git://github.com/darander/oe-core busybox-systemd

Comments

Anders Darander - Nov. 22, 2011, 2:34 p.m.
This updates busybox to the latest stable, 1.19.3.

Among other things, there should be rudimentary support in syslogd for
systemd, by enabling CONFIG_FEATURE_SYSTEMD.

It has been tested and verified on qemux86 and on our local arm-based system.

However, the patch busybox-udhcpc-no_deconfig.patch has been modified to
apply, but hasn't really been tested. As we're currently mostly running static
configurations, we've only verified that udhcpc works to the extent that we
get the network up and running, not how it behaves during dhcp renewal.
Thus I'm starting this as an RFC.

Changes:
v2: * Checked the new defconfig (removed settings implying CFLAGS and
    ARCH). The new defconfig should be as close as possible to the old one,
    with the exception of some new utils/options. 
    * Added one patch from upstream, to safeguard the use of CBAUD
    termios_init(). 
    * The other upstream patch from http://busybox.net/downloads/fixes-1.19.3/
    do only affect comments, thus I'm not adding it.

The following changes since commit dd2fdf9f5a3923c37e4ea2e46e347bb0657c2f5b:

  gcc-4.6: Fix gcc ICE on qt4-x11-free/armv7-a (2011-11-21 16:55:01 +0000)

are available in the git repository at:
  git://github.com/darander/oe-core busybox-systemd
  https://github.com/darander/oe-core/tree/busybox-systemd

Anders Darander (1):
  busybox: update to 1.19.3

 .../B921600.patch                                  |    0
 .../busybox-1.19.3/busybox-1.19.3-getty.patch      |   17 ++
 .../busybox-appletlib-dependency.patch             |    0
 .../busybox-udhcpc-no_deconfig.patch               |   66 +++--
 .../{busybox-1.18.5 => busybox-1.19.3}/defconfig   |  265 +++++++++++++-------
 .../get_header_tar.patch                           |    0
 .../run-parts.in.usr-bin.patch                     |   16 +-
 .../udhcpc-fix-nfsroot.patch                       |    0
 .../udhcpscript.patch                              |    0
 .../{busybox_1.18.5.bb => busybox_1.19.3.bb}       |    7 +-
 10 files changed, 245 insertions(+), 126 deletions(-)
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/B921600.patch (100%)
 create mode 100644 meta/recipes-core/busybox/busybox-1.19.3/busybox-1.19.3-getty.patch
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/busybox-appletlib-dependency.patch (100%)
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/busybox-udhcpc-no_deconfig.patch (62%)
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/defconfig (89%)
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/get_header_tar.patch (100%)
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/run-parts.in.usr-bin.patch (64%)
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/udhcpc-fix-nfsroot.patch (100%)
 rename meta/recipes-core/busybox/{busybox-1.18.5 => busybox-1.19.3}/udhcpscript.patch (100%)
 rename meta/recipes-core/busybox/{busybox_1.18.5.bb => busybox_1.19.3.bb} (82%)
Saul Wold - Nov. 22, 2011, 8:36 p.m.
On 11/22/2011 06:34 AM, Anders Darander wrote:
> This updates busybox to the latest stable, 1.19.3.
>
> Among other things, there should be rudimentary support in syslogd for
> systemd, by enabling CONFIG_FEATURE_SYSTEMD.
>
How much size does this add to busybox by having it enabled by default?
Is it possible to conditional add a config fragment if systemd is 
enabled ad the DISTRO/IMAGE_FEATURE level?

More info is required.

> It has been tested and verified on qemux86 and on our local arm-based system.
>
> However, the patch busybox-udhcpc-no_deconfig.patch has been modified to
> apply, but hasn't really been tested. As we're currently mostly running static
> configurations, we've only verified that udhcpc works to the extent that we
> get the network up and running, not how it behaves during dhcp renewal.
> Thus I'm starting this as an RFC.
>
> Changes:
> v2: * Checked the new defconfig (removed settings implying CFLAGS and
>      ARCH). The new defconfig should be as close as possible to the old one,
>      with the exception of some new utils/options.
Can you clearly enumerate what new utils and options and what their size 
impact on the busybox image is.

Thanks
	Sau!

>      * Added one patch from upstream, to safeguard the use of CBAUD
>      termios_init().
>      * The other upstream patch from http://busybox.net/downloads/fixes-1.19.3/
>      do only affect comments, thus I'm not adding it.
>
> The following changes since commit dd2fdf9f5a3923c37e4ea2e46e347bb0657c2f5b:
>
>    gcc-4.6: Fix gcc ICE on qt4-x11-free/armv7-a (2011-11-21 16:55:01 +0000)
>
> are available in the git repository at:
>    git://github.com/darander/oe-core busybox-systemd
>    https://github.com/darander/oe-core/tree/busybox-systemd
>
> Anders Darander (1):
>    busybox: update to 1.19.3
>
>   .../B921600.patch                                  |    0
>   .../busybox-1.19.3/busybox-1.19.3-getty.patch      |   17 ++
>   .../busybox-appletlib-dependency.patch             |    0
>   .../busybox-udhcpc-no_deconfig.patch               |   66 +++--
>   .../{busybox-1.18.5 =>  busybox-1.19.3}/defconfig   |  265 +++++++++++++-------
>   .../get_header_tar.patch                           |    0
>   .../run-parts.in.usr-bin.patch                     |   16 +-
>   .../udhcpc-fix-nfsroot.patch                       |    0
>   .../udhcpscript.patch                              |    0
>   .../{busybox_1.18.5.bb =>  busybox_1.19.3.bb}       |    7 +-
>   10 files changed, 245 insertions(+), 126 deletions(-)
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/B921600.patch (100%)
>   create mode 100644 meta/recipes-core/busybox/busybox-1.19.3/busybox-1.19.3-getty.patch
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/busybox-appletlib-dependency.patch (100%)
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/busybox-udhcpc-no_deconfig.patch (62%)
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/defconfig (89%)
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/get_header_tar.patch (100%)
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/run-parts.in.usr-bin.patch (64%)
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/udhcpc-fix-nfsroot.patch (100%)
>   rename meta/recipes-core/busybox/{busybox-1.18.5 =>  busybox-1.19.3}/udhcpscript.patch (100%)
>   rename meta/recipes-core/busybox/{busybox_1.18.5.bb =>  busybox_1.19.3.bb} (82%)
>
Anders Darander - Nov. 22, 2011, 8:47 p.m.
On 22 nov 2011, at 21:36, "Saul Wold" <sgw@linux.intel.com> wrote:
> On 11/22/2011 06:34 AM, Anders Darander wrote:
>> This updates busybox to the latest stable, 1.19.3.
>> 
>> Among other things, there should be rudimentary support in syslogd for
>> systemd, by enabling CONFIG_FEATURE_SYSTEMD.
>> 
> How much size does this add to busybox by having it enabled by default?
> Is it possible to conditional add a config fragment if systemd is 
> enabled ad the DISTRO/IMAGE_FEATURE level?

I'd love that. Any suggestions for a distro/image feature? As far as I know, we don't have any for init system selection.

I let the rather small systemd feature be enabled as we already package systemd unit files for a number of packages, and we also add --with-systemddir during configure or a number of packages. If we add a distro/image feature, we should(?) conditionally ship SysV init files also, or package them separately. (Although that is for another discussion...)

> More info is required.
> 
>> It has been tested and verified on qemux86 and on our local arm-based system.
>> 
>> However, the patch busybox-udhcpc-no_deconfig.patch has been modified to
>> apply, but hasn't really been tested. As we're currently mostly running static
>> configurations, we've only verified that udhcpc works to the extent that we
>> get the network up and running, not how it behaves during dhcp renewal.
>> Thus I'm starting this as an RFC.
>> 
>> Changes:
>> v2: * Checked the new defconfig (removed settings implying CFLAGS and
>>     ARCH). The new defconfig should be as close as possible to the old one,
>>     with the exception of some new utils/options.
> Can you clearly enumerate what new utils and options and what their size 
> impact on the busybox image is.

Sure, I'll come bak with a list, primarily focusing on any new features activated (most is deactivated though), although it could take a little time. Too much other things to handle...

Cheers,
/Anders
Anders Darander - Nov. 23, 2011, 8:59 a.m.
* Saul Wold <sgw@linux.intel.com> [111122 21:36]:
> On 11/22/2011 06:34 AM, Anders Darander wrote:
> > This updates busybox to the latest stable, 1.19.3.

> > Among other things, there should be rudimentary support in syslogd for
> > systemd, by enabling CONFIG_FEATURE_SYSTEMD.

> How much size does this add to busybox by having it enabled by default?

Enabling FEATURE_SYSTEMD in busybox costs 192 bytes in my tests in
qemux86.

> Is it possible to conditional add a config fragment if systemd is 
> enabled ad the DISTRO/IMAGE_FEATURE level?

> More info is required.

> > It has been tested and verified on qemux86 and on our local arm-based system.

> > However, the patch busybox-udhcpc-no_deconfig.patch has been modified to
> > apply, but hasn't really been tested. As we're currently mostly running static
> > configurations, we've only verified that udhcpc works to the extent that we
> > get the network up and running, not how it behaves during dhcp renewal.
> > Thus I'm starting this as an RFC.

> > Changes:
> > v2: * Checked the new defconfig (removed settings implying CFLAGS and
> >      ARCH). The new defconfig should be as close as possible to the old one,
> >      with the exception of some new utils/options.
> Can you clearly enumerate what new utils and options and what their size 
> impact on the busybox image is.

Apart from the FEATURE_SYSTEMD discussed above, these are the other new
options that I kept the new busybox default on (i.e. these are enabled,
while I turned of quite a few other options that automatically got
enabled). All costs are evaluted using qemux86, and the busybox binary
size is checked in the packages-split/busybox/bin directory.

I don't mind disabling any of these feature in a v3, if
desired/requested. Anyway, I'm running a completely custom config for my
normal uses...

FEATURE_RTMINMAX, support RTMIN[+n] RTMAX[-n] signals, claimed to cost
~250 bytes

FEATURE_REVERSE_SEARCH claimed to cost ~0.5k

FEATURE_SKIP_ROOTFS ignore rootfs entry in mount table (tools like df
etc will ignore it).

FEATURE_AR_CREATE, enable ar to create files, ~2.5k

FEATURE_SEAMLESS_XZ enable xz compression in tar, no measured cost.

XZ and UNXZ, enable xz compression, 8k

FGCONSOLE, print active console number, 128 bytes

FEATURE_LOADFONT_PSF2, FEATURE_LOADFONT_RAW, cost 576 bytes

FEATURE_VI_ASK_TERMINAL, last resort to find terminal size, 352 bytes

BLOCKDEV, perform some ioctls with block devices, cost 480 bytes

FDISK_SUPPORT_LARGE_DISKS, is disabled, it no longer wants to play with
LFS (both were enabled previously).

FEATURE_MOUNT_LOOP_CREATE, create new loopback devices if needed,
couldn't see any cost

FEATURE_WGET_TIMEOUT, no visible cost

FEATURE_SH_HISTFILESIZE, use $HISTFILESIZE (still max limited by
config), no visible cost

FEATURE_SYSLOGD_CFG, support syslog.conf, no visible cost

FEATURE_SYSLOGD_READ_BUFFER_SIZE, defines the size of the read buffer,
not measured

FEATURE_KLOGD_KLOGCTL, let klogd use the klogctl() interface, no visible
cost

Cheers,
Anders
Otavio Salvador - Nov. 23, 2011, 10:38 a.m.
On Tue, Nov 22, 2011 at 18:36, Saul Wold <sgw@linux.intel.com> wrote:

> On 11/22/2011 06:34 AM, Anders Darander wrote:
>
>> This updates busybox to the latest stable, 1.19.3.
>>
>> Among other things, there should be rudimentary support in syslogd for
>> systemd, by enabling CONFIG_FEATURE_SYSTEMD.
>>
>>  How much size does this add to busybox by having it enabled by default?
> Is it possible to conditional add a config fragment if systemd is enabled
> ad the DISTRO/IMAGE_FEATURE level?
>

If it is small it would be nice to have it enabled.
Anders Darander - Nov. 23, 2011, 10:42 a.m.
* Otavio Salvador <otavio@ossystems.com.br> [111123 11:39]:
> On Tue, Nov 22, 2011 at 18:36, Saul Wold <sgw@linux.intel.com> wrote:

>     On 11/22/2011 06:34 AM, Anders Darander wrote:

>         This updates busybox to the latest stable, 1.19.3.

>         Among other things, there should be rudimentary support in syslogd for
>         systemd, by enabling CONFIG_FEATURE_SYSTEMD.

>     How much size does this add to busybox by having it enabled by default?
>     Is it possible to conditional add a config fragment if systemd is enabled
>     ad the DISTRO/IMAGE_FEATURE level?

> If it is small it would be nice to have it enabled.

If I didn't copy my figures wrongly, it's just a couple of hundred
bytes. Thus it shouldn't really hurt more than e.g. shipping SysV-init
files on a systemd-image...

OTOH, I'll leave it up to you guys to decide on the defconfig, as I'll
always use my own custom defconfig.

Cheers,
Anders
Saul Wold - Nov. 29, 2011, 5:47 a.m.
On 11/23/2011 12:59 AM, Anders Darander wrote:
> * Saul Wold<sgw@linux.intel.com>  [111122 21:36]:
>> On 11/22/2011 06:34 AM, Anders Darander wrote:
>>> This updates busybox to the latest stable, 1.19.3.
>
>>> Among other things, there should be rudimentary support in syslogd for
>>> systemd, by enabling CONFIG_FEATURE_SYSTEMD.
>
>> How much size does this add to busybox by having it enabled by default?
>
> Enabling FEATURE_SYSTEMD in busybox costs 192 bytes in my tests in
> qemux86.
>
>> Is it possible to conditional add a config fragment if systemd is
>> enabled ad the DISTRO/IMAGE_FEATURE level?
>
>> More info is required.
>
>>> It has been tested and verified on qemux86 and on our local arm-based system.
>
>>> However, the patch busybox-udhcpc-no_deconfig.patch has been modified to
>>> apply, but hasn't really been tested. As we're currently mostly running static
>>> configurations, we've only verified that udhcpc works to the extent that we
>>> get the network up and running, not how it behaves during dhcp renewal.
>>> Thus I'm starting this as an RFC.
>
>>> Changes:
>>> v2: * Checked the new defconfig (removed settings implying CFLAGS and
>>>       ARCH). The new defconfig should be as close as possible to the old one,
>>>       with the exception of some new utils/options.
>> Can you clearly enumerate what new utils and options and what their size
>> impact on the busybox image is.
>
> Apart from the FEATURE_SYSTEMD discussed above, these are the other new
> options that I kept the new busybox default on (i.e. these are enabled,
> while I turned of quite a few other options that automatically got
> enabled). All costs are evaluted using qemux86, and the busybox binary
> size is checked in the packages-split/busybox/bin directory.
>
> I don't mind disabling any of these feature in a v3, if
> desired/requested. Anyway, I'm running a completely custom config for my
> normal uses...
>
> FEATURE_RTMINMAX, support RTMIN[+n] RTMAX[-n] signals, claimed to cost
> ~250 bytes
>
I can see these being useful

> FEATURE_REVERSE_SEARCH claimed to cost ~0.5k
>
Why is this needed?

> FEATURE_SKIP_ROOTFS ignore rootfs entry in mount table (tools like df
> etc will ignore it).
>
OK

> FEATURE_AR_CREATE, enable ar to create files, ~2.5k
>
> FEATURE_SEAMLESS_XZ enable xz compression in tar, no measured cost.
>
> XZ and UNXZ, enable xz compression, 8k
>
This is the one and the ar create above that sticks out, are these 
needed in the general case or just for your config?

> FGCONSOLE, print active console number, 128 bytes
>
> FEATURE_LOADFONT_PSF2, FEATURE_LOADFONT_RAW, cost 576 bytes
>
Seems resonable, but why did we not need this before, what changed?

> FEATURE_VI_ASK_TERMINAL, last resort to find terminal size, 352 bytes
>
Not sure about this and the FGCONSOLE above.

> BLOCKDEV, perform some ioctls with block devices, cost 480 bytes
>
Again is this useful in the general case?

> FDISK_SUPPORT_LARGE_DISKS, is disabled, it no longer wants to play with
> LFS (both were enabled previously).
>
OK

> FEATURE_MOUNT_LOOP_CREATE, create new loopback devices if needed,
> couldn't see any cost
>
OK

> FEATURE_WGET_TIMEOUT, no visible cost
>
OK

> FEATURE_SH_HISTFILESIZE, use $HISTFILESIZE (still max limited by
> config), no visible cost
>
OK

> FEATURE_SYSLOGD_CFG, support syslog.conf, no visible cost
>
OK

> FEATURE_SYSLOGD_READ_BUFFER_SIZE, defines the size of the read buffer,
> not measured
>
OK

> FEATURE_KLOGD_KLOGCTL, let klogd use the klogctl() interface, no visible
> cost
OK

>
> Cheers,
> Anders
>
Anders Darander - Nov. 29, 2011, 7:04 a.m.
* Saul Wold <sgw@linux.intel.com> [111129 06:47]:
> On 11/23/2011 12:59 AM, Anders Darander wrote:
> > * Saul Wold<sgw@linux.intel.com>  [111122 21:36]:
> >> On 11/22/2011 06:34 AM, Anders Darander wrote:
> >>> This updates busybox to the latest stable, 1.19.3.

> >>> Among other things, there should be rudimentary support in syslogd for
> >>> systemd, by enabling CONFIG_FEATURE_SYSTEMD.

> >> How much size does this add to busybox by having it enabled by default?

> > Enabling FEATURE_SYSTEMD in busybox costs 192 bytes in my tests in
> > qemux86.

> >> Is it possible to conditional add a config fragment if systemd is
> >> enabled ad the DISTRO/IMAGE_FEATURE level?

> >> More info is required.

Any more comment on this one? Otavio replied earlier that if it was a
small cost, he would prefer to have it on by default. Otherwise, any
suggestions for DISTRO/IMAGE_FEATURE?

> >>> Changes:
> >>> v2: * Checked the new defconfig (removed settings implying CFLAGS and
> >>>       ARCH). The new defconfig should be as close as possible to the old one,
> >>>       with the exception of some new utils/options.
> >> Can you clearly enumerate what new utils and options and what their size
> >> impact on the busybox image is.

I'll remove all features that you didn't OK (see below), unless we get
some more replies/wishes in the next day. (Thus I plan to send a v3 this
week, if times allow it).

Just to clarify it once more, the choice of which new features to enable
was done by trying to judge if they were reasonable, or seemed
helpfull/usefull enough. I'm not normally running with the defconfig
(apart from testing this patch), I'm using a heavily customized and
often minimized defconfig.

> > Apart from the FEATURE_SYSTEMD discussed above, these are the other new
> > options that I kept the new busybox default on (i.e. these are enabled,
> > while I turned of quite a few other options that automatically got
> > enabled). All costs are evaluted using qemux86, and the busybox binary
> > size is checked in the packages-split/busybox/bin directory.

> > I don't mind disabling any of these feature in a v3, if
> > desired/requested. Anyway, I'm running a completely custom config for my
> > normal uses...

> > FEATURE_RTMINMAX, support RTMIN[+n] RTMAX[-n] signals, claimed to cost
> > ~250 bytes

> I can see these being useful

> > FEATURE_REVERSE_SEARCH claimed to cost ~0.5k

> Why is this needed?

Not needed, just nice to have. But I'll remove it in v3.

> > FEATURE_AR_CREATE, enable ar to create files, ~2.5k

> > FEATURE_SEAMLESS_XZ enable xz compression in tar, no measured cost.

> > XZ and UNXZ, enable xz compression, 8k

> This is the one and the ar create above that sticks out, are these 
> needed in the general case or just vfor your config?

Shouldn't be needed. I just left it enabled as all other
FEATURE_SEAMLESS's was enabled. I'll disable these two in v3.

> > FGCONSOLE, print active console number, 128 bytes

> > FEATURE_LOADFONT_PSF2, FEATURE_LOADFONT_RAW, cost 576 bytes

> Seems resonable, but why did we not need this before, what changed?

Not really sure what the change is (normally I'm only working on
headless systems, thus no need for fonts). The two FEATURE_LOADFONT_*
options were not available in the old defconfig. I'll leave these two in
v3, as you thought the seemd reasonable.

> > FEATURE_VI_ASK_TERMINAL, last resort to find terminal size, 352 bytes

> Not sure about this and the FGCONSOLE above.

Sure, I'll remove these two.

> > BLOCKDEV, perform some ioctls with block devices, cost 480 bytes

> Again is this useful in the general case?

Same as previously, an arbitrary choice. I'll remove it in v3.
Phil Blundell - Nov. 29, 2011, 7:54 a.m.
On Tue, 2011-11-29 at 08:04 +0100, Anders Darander wrote:
> Any more comment on this one? Otavio replied earlier that if it was a
> small cost, he would prefer to have it on by default. Otherwise, any
> suggestions for DISTRO/IMAGE_FEATURE?

We've talked about this a few times in the past and it's never seemed as
though this would buy much in relation to busybox.  The busybox
configuration is so highly specific that each DISTRO is probably going
to end up wanting its own custom configuration file and it isn't obvious
that there is going to be much/any commonality between them.  That being
the case it seems as though it's easier for the distros to actually ship
the config files themselves, rather than try to translate every
conceivable option into PACKAGE_CONFIG and then express them that way.

I guess the other thing that naturally follows from this is that there
probably aren't very many people using the actual defconfig that oe-core
provides, and hence it probably doesn't matter all that much what is and
isn't enabled in it.  Enabling FEATURE_SYSTEMD seems like a reasonable
idea if there is already the mechanism in oe-core to make use of it, but
probably just a waste of space otherwise.

p.
Saul Wold - Nov. 29, 2011, 8:13 a.m.
On 11/28/2011 11:04 PM, Anders Darander wrote:
> * Saul Wold<sgw@linux.intel.com>  [111129 06:47]:
>> On 11/23/2011 12:59 AM, Anders Darander wrote:
>>> * Saul Wold<sgw@linux.intel.com>   [111122 21:36]:
>>>> On 11/22/2011 06:34 AM, Anders Darander wrote:
>>>>> This updates busybox to the latest stable, 1.19.3.
>
>>>>> Among other things, there should be rudimentary support in syslogd for
>>>>> systemd, by enabling CONFIG_FEATURE_SYSTEMD.
>
>>>> How much size does this add to busybox by having it enabled by default?
>
>>> Enabling FEATURE_SYSTEMD in busybox costs 192 bytes in my tests in
>>> qemux86.
>
>>>> Is it possible to conditional add a config fragment if systemd is
>>>> enabled ad the DISTRO/IMAGE_FEATURE level?
>
>>>> More info is required.
>
> Any more comment on this one? Otavio replied earlier that if it was a
> small cost, he would prefer to have it on by default. Otherwise, any
> suggestions for DISTRO/IMAGE_FEATURE?
>
Sorry for not being clear on this on, yes to the SYSTEMD change, it's 
small enough and something we are considering. As Phil points out in the 
follow-up to this, most are using custom defconfig's that we should just 
go ahead and enable this for future flexibility.

>>>>> Changes:
>>>>> v2: * Checked the new defconfig (removed settings implying CFLAGS and
>>>>>        ARCH). The new defconfig should be as close as possible to the old one,
>>>>>        with the exception of some new utils/options.
>>>> Can you clearly enumerate what new utils and options and what their size
>>>> impact on the busybox image is.
>
> I'll remove all features that you didn't OK (see below), unless we get
> some more replies/wishes in the next day. (Thus I plan to send a v3 this
> week, if times allow it).
>
> Just to clarify it once more, the choice of which new features to enable
> was done by trying to judge if they were reasonable, or seemed
> helpfull/usefull enough. I'm not normally running with the defconfig
> (apart from testing this patch), I'm using a heavily customized and
> often minimized defconfig.
>
>>> Apart from the FEATURE_SYSTEMD discussed above, these are the other new
>>> options that I kept the new busybox default on (i.e. these are enabled,
>>> while I turned of quite a few other options that automatically got
>>> enabled). All costs are evaluted using qemux86, and the busybox binary
>>> size is checked in the packages-split/busybox/bin directory.
>
>>> I don't mind disabling any of these feature in a v3, if
>>> desired/requested. Anyway, I'm running a completely custom config for my
>>> normal uses...
>
>>> FEATURE_RTMINMAX, support RTMIN[+n] RTMAX[-n] signals, claimed to cost
>>> ~250 bytes
>
>> I can see these being useful
>
>>> FEATURE_REVERSE_SEARCH claimed to cost ~0.5k
>
>> Why is this needed?
>
> Not needed, just nice to have. But I'll remove it in v3.
>
>>> FEATURE_AR_CREATE, enable ar to create files, ~2.5k
>
>>> FEATURE_SEAMLESS_XZ enable xz compression in tar, no measured cost.
>
>>> XZ and UNXZ, enable xz compression, 8k
>
>> This is the one and the ar create above that sticks out, are these
>> needed in the general case or just vfor your config?
>
> Shouldn't be needed. I just left it enabled as all other
> FEATURE_SEAMLESS's was enabled. I'll disable these two in v3.
>
>>> FGCONSOLE, print active console number, 128 bytes
>
>>> FEATURE_LOADFONT_PSF2, FEATURE_LOADFONT_RAW, cost 576 bytes
>
>> Seems resonable, but why did we not need this before, what changed?
>
> Not really sure what the change is (normally I'm only working on
> headless systems, thus no need for fonts). The two FEATURE_LOADFONT_*
> options were not available in the old defconfig. I'll leave these two in
> v3, as you thought the seemd reasonable.
>
For minimal, it's good to assume headless, then we should not need the 
LOADFONT. I thought it was needed for headless as well.

>>> FEATURE_VI_ASK_TERMINAL, last resort to find terminal size, 352 bytes
>
>> Not sure about this and the FGCONSOLE above.
>
> Sure, I'll remove these two.
>
>>> BLOCKDEV, perform some ioctls with block devices, cost 480 bytes
>
>> Again is this useful in the general case?
>
> Same as previously, an arbitrary choice. I'll remove it in v3.
>
>
Thanks for your efforts.

Sau!

>
Anders Darander - Nov. 29, 2011, 8:29 a.m.
* Saul Wold <sgw@linux.intel.com> [111129 09:14]:
> On 11/28/2011 11:04 PM, Anders Darander wrote:
> > * Saul Wold<sgw@linux.intel.com>  [111129 06:47]:
> >> On 11/23/2011 12:59 AM, Anders Darander wrote:
> >>> * Saul Wold<sgw@linux.intel.com>   [111122 21:36]:
> >>>> On 11/22/2011 06:34 AM, Anders Darander wrote:
> >>>>> This updates busybox to the latest stable, 1.19.3.

> >>>>> Among other things, there should be rudimentary support in syslogd for
> >>>>> systemd, by enabling CONFIG_FEATURE_SYSTEMD.

> >>>> How much size does this add to busybox by having it enabled by default?

> >>> Enabling FEATURE_SYSTEMD in busybox costs 192 bytes in my tests in
> >>> qemux86.

> >>>> Is it possible to conditional add a config fragment if systemd is
> >>>> enabled ad the DISTRO/IMAGE_FEATURE level?

> >>>> More info is required.

> > Any more comment on this one? Otavio replied earlier that if it was a
> > small cost, he would prefer to have it on by default. Otherwise, any
> > suggestions for DISTRO/IMAGE_FEATURE?

> Sorry for not being clear on this on, yes to the SYSTEMD change, it's 
> small enough and something we are considering. As Phil points out in the 
> follow-up to this, most are using custom defconfig's that we should just 
> go ahead and enable this for future flexibility.

Thanks!

As we're running systemd, I'm one of those in favour of getting systemd
into oe-core, at least now that is hopefully is getting slightly more
stable.

And yes, just as Phil points out, we're also running custom defconfigs.

> >>>>> Changes:
> >>>>> v2: * Checked the new defconfig (removed settings implying CFLAGS and
> >>>>>        ARCH). The new defconfig should be as close as possible to the old one,
> >>>>>        with the exception of some new utils/options.
> >>>> Can you clearly enumerate what new utils and options and what their size
> >>>> impact on the busybox image is.


> >>> FGCONSOLE, print active console number, 128 bytes

> >>> FEATURE_LOADFONT_PSF2, FEATURE_LOADFONT_RAW, cost 576 bytes

> >> Seems resonable, but why did we not need this before, what changed?

> > Not really sure what the change is (normally I'm only working on
> > headless systems, thus no need for fonts). The two FEATURE_LOADFONT_*
> > options were not available in the old defconfig. I'll leave these two in
> > v3, as you thought the seemd reasonable.

> For minimal, it's good to assume headless, then we should not need the 
> LOADFONT. I thought it was needed for headless as well.

Well, there is a risk of getting a few warning when booting, at least if
you also disable CONFIG_LOADFONT, unless you look through the boot
process. (We're running our system without CONFIG_LOADFONT, hence the
other LOADFONT-options becomes moot).

Cheers,
Anders