Suppose I have the following bugfix patch for a 3rd party machine conf
file I am using:

diff --git a/conf/machine/colibri-vf.conf b/conf/machine/colibri-vf.conf
index 3ddef79..ba47488 100644
--- a/conf/machine/colibri-vf.conf
+++ b/conf/machine/colibri-vf.conf
@@ -35,6 +35,6 @@ MKUBIFS_ARGS = " -c 8112 -e 124KiB -m 2KiB -F"
UBINIZE_ARGS = " -p 128KiB -m 2048 -s 2048"
UBI_VOLNAME = "rootfs"

-SERIAL_CONSOLE ?= "115200 ttyLP0"
+SERIAL_CONSOLE = "115200 ttyLP0"

MACHINE_FEATURES += "usbgadget usbhost vfat alsa touchscreen"

Is there a proper way to somehow temporarily add something to my layer
to apply this patch until it makes its way upstream and into the
branch I am using? Or should I make my own machine file, include the
original one and then change the variable?

So far I have not found any elegant way to otherwise force this
variable to its correct value in my image recipe, the only way I have
found to work around this bug is to .bbappend the inittab recipe where
this variable is actually used and change the value of another
variable (SERIAL_CONSOLES, note the S at the end) which is derived
from SERIAL_CONSOLE right there in this bbappend file. This seems to
help.

But while doing this I have also noticed a strange anomaly in the
output of bitbake -e:

* When I change the variable in my image recipe then bitbake -e will
show another "set" access and both variables SERIAL_CONSOLE and
SERIAL_CONSOLES will have the correct value as intended by me but the
produced image will have the **wrong** entry in its inittab.

* When I bbappend the inittab recipe to set SERIAL_CONSOLES right
there where it is used then my final image will have a correct inittab
and the serial console will work bit there is **no** mention of that
variable change in the output of bitbake -e

* Only when I change the machine conf file iitself to set the variable
then bitbake -e and the produced image both show the correct entry.
Why does it behave that way?

On Thu, Oct 12, 2017 at 3:32 AM, Bernd <prof7bit@...> wrote:

Suppose I have the following bugfix patch for a 3rd party machine conf
file I am using:

diff --git a/conf/machine/colibri-vf.conf b/conf/machine/colibri-vf.conf
index 3ddef79..ba47488 100644
--- a/conf/machine/colibri-vf.conf
+++ b/conf/machine/colibri-vf.conf
@@ -35,6 +35,6 @@ MKUBIFS_ARGS = " -c 8112 -e 124KiB -m 2KiB -F"
UBINIZE_ARGS = " -p 128KiB -m 2048 -s 2048"
UBI_VOLNAME = "rootfs"

-SERIAL_CONSOLE ?= "115200 ttyLP0"
+SERIAL_CONSOLE = "115200 ttyLP0"

MACHINE_FEATURES += "usbgadget usbhost vfat alsa touchscreen"

Is there a proper way to somehow temporarily add something to my layer
to apply this patch until it makes its way upstream and into the
branch I am using? Or should I make my own machine file, include the
original one and then change the variable?

That would be one way. You could also over-ride from any other global
config file, e.g. local.conf or the distro config file (if you have
one).

So far I have not found any elegant way to otherwise force this
variable to its correct value in my image recipe

Right, the variable is not used by the image recipe, so setting it
there won't change anything. Variables in an image recipe generally
only relate to how the image is constructed (what packages it contains
and format of the final image - tar.gz, squashfs, etc).

, the only way I have
found to work around this bug is to .bbappend the inittab recipe where
this variable is actually used and change the value of another
variable (SERIAL_CONSOLES, note the S at the end) which is derived
from SERIAL_CONSOLE right there in this bbappend file. This seems to
help.

For variables which are specific to one recipe then over-riding in the
context of that recipe is OK. However, SERIAL_CONSOLES is used in a
few different recipes, so over-riding from a global config file would
necessary to be sure of changing it consistently wherever it's used.

But while doing this I have also noticed a strange anomaly in the
output of bitbake -e:

* When I change the variable in my image recipe then bitbake -e will
show another "set" access and both variables SERIAL_CONSOLE and
SERIAL_CONSOLES will have the correct value as intended by me but the
produced image will have the **wrong** entry in its inittab.

If you run bitbake -e for inittab and for the image recipe and compare
the two it should show that setting the variable from within the image
recipe has no effect on the inittab recipe. That's not a bug.

* When I bbappend the inittab recipe to set SERIAL_CONSOLES right
there where it is used then my final image will have a correct inittab
and the serial console will work bit there is **no** mention of that
variable change in the output of bitbake -e

Similar to above, bitbake -e will give different results for each recipe.

* Only when I change the machine conf file iitself to set the variable
then bitbake -e and the produced image both show the correct entry.
Why does it behave that way?

Because the machine config is a global config file and any variables
it sets are seen by every recipe.

Hi,

On 12.10.2017 12:32, Bernd wrote:

Suppose I have the following bugfix patch for a 3rd party machine conf
file I am using:

diff --git a/conf/machine/colibri-vf.conf b/conf/machine/colibri-vf.conf
index 3ddef79..ba47488 100644
--- a/conf/machine/colibri-vf.conf
+++ b/conf/machine/colibri-vf.conf
@@ -35,6 +35,6 @@ MKUBIFS_ARGS = " -c 8112 -e 124KiB -m 2KiB -F"
UBINIZE_ARGS = " -p 128KiB -m 2048 -s 2048"
UBI_VOLNAME = "rootfs"

-SERIAL_CONSOLE ?= "115200 ttyLP0"
+SERIAL_CONSOLE = "115200 ttyLP0"

MACHINE_FEATURES += "usbgadget usbhost vfat alsa touchscreen"

Is there a proper way to somehow temporarily add something to my layer
to apply this patch until it makes its way upstream and into the
branch I am using? Or should I make my own machine file, include the
original one and then change the variable?

you can add to local.conf or distro.conf :

include conf/machine/${MACHINE}-extra.conf

then create in your meta layer or bsp layer :

conf/machine/beaglebone-extra.conf
conf/machine/rasberrypi-extra.conf
conf/machine/xyz-extra.conf

which contains overrides for your machine settings.
Note that is an include and not require, which means that if you machine has no *extra.conf then it's simply skkiped ( no overrides)

So far I have not found any elegant way to otherwise force this
variable to its correct value in my image recipe, the only way I have
found to work around this bug is to .bbappend the inittab recipe where
this variable is actually used and change the value of another
variable (SERIAL_CONSOLES, note the S at the end) which is derived
from SERIAL_CONSOLE right there in this bbappend file. This seems to
help.

But while doing this I have also noticed a strange anomaly in the
output of bitbake -e:

* When I change the variable in my image recipe then bitbake -e will
show another "set" access and both variables SERIAL_CONSOLE and
SERIAL_CONSOLES will have the correct value as intended by me but the
produced image will have the **wrong** entry in its inittab.

* When I bbappend the inittab recipe to set SERIAL_CONSOLES right
there where it is used then my final image will have a correct inittab
and the serial console will work bit there is **no** mention of that
variable change in the output of bitbake -e

* Only when I change the machine conf file iitself to set the variable
then bitbake -e and the produced image both show the correct entry.
Why does it behave that way?

--
Ayoub Zaki
Embedded Systems Consultant

Vaihinger Straße 2/1
D-71634 Ludwigsburg

Tel. : +4971415074546
Mobile : +4917662901545
Email : ayoub.zaki@...
Homepage : https://embexus.com

I have now combined the last 2 answers:

I did not want to put more customization than machine and distro into
my local.conf and since I already have my own distro conf I have now
put the include conf/machine/${MACHINE}-extra.conf into my own distro
conf file and made colibri-vf-extra.conf, both files are now inside my
own layer folder and under my project's version control so it won't
get lost. This works. Thank you

2017-10-13 10:56 GMT+02:00 Bernd <prof7bit@...>: