[OE-core] [PATCH v2] testimage: dont use DNS lookup for qemu based testimages

[Paul Eggleton]

On Wednesday 19 March 2014 13:06:51 Saul Wold wrote:
> On 03/19/2014 06:17 AM, Paul Eggleton wrote:
> > On Wednesday 19 March 2014 12:23:37 Stanacar, StefanX wrote:
> >> On Tue, 2014-03-18 at 17:31 -0700, Saul Wold wrote:
> >>> Signed-off-by: Saul Wold <sgw at linux.intel.com>
> >>> ---
> >>> 
> >>>   meta/classes/testimage.bbclass | 10 ++++++++++
> >>>   1 file changed, 10 insertions(+)
> >>> 
> >>> diff --git a/meta/classes/testimage.bbclass
> >>> b/meta/classes/testimage.bbclass index 691c7f6..ee028e1 100644
> >>> --- a/meta/classes/testimage.bbclass
> >>> +++ b/meta/classes/testimage.bbclass
> >>> @@ -45,6 +45,16 @@ TESTIMAGEDEPENDS_qemuall =
> >>> "qemu-native:do_populate_sysroot qemu-helper-native:d>
> >>> 
> >>>   TESTIMAGELOCK = "${TMPDIR}/testimage.lock"
> >>>   TESTIMAGELOCK_qemuall = ""
> >>> 
> >>> +IMAGE_POSTPROCESS_COMMAND_qemuall += "fix_dns_lookup_for_qemu"
> >>> +
> >> 
> >> I might be wrong, but this won't have any effect with the default
> >> (manual) way we use testimage. The testimage class isn't in the global
> >> inherit, so that  IMAGE_POSTPROCESS_COMMAND won't run as the image isn't
> >> reconstructed.
> >> 
> >> The AB does: bitbake core-image-sato on one step and then adds INHERIT
> >> +="testimage" in local.conf in the next step and runs bitbake
> >> core-image-sato -c testimage, no more do_rootfs here.
> >> 
> >> As it this it will only work with testimage-auto when you have
> >> TEST_IMAGE="1" in local.conf in the first build step so you only need to
> >> do bitbake core-image-sato and do_testimage gets run after do_rootfs,
> >> but we don't use this. And the fix should work for running the task
> >> manually too.
> > 
> > I have to say I agree; this is not the right fix.
> 
> Is there any reason that we can't include the INHERIT += "testimage"
> during the core build?

I don't want testimage.bbclass influencing how images are constructed. It's 
just as valid a workaround for people running qemu manually on their machines 
as it is for the automated testing.

> The other choice is to move the function into the openssh recipe and
> then the actual setting of the IMAGE_POSTPROCESS_COMMAND can happen in
> the local.conf or auto.conf

That can't work. The function has to be defined in the scope of the image not 
the openssh recipe or it can't be called from IMAGE_POSTPROCESS_COMMAND.

To be honest we already have code in image.bbclass for allowing empty 
passwords in OpenSSH / dropbear; this seems to me to be in the same category, 
so I would suggest putting at least the function there (with a more generic 
name that does not contain "qemu").

> >> I liked v1, I don't really see what the problem was there. No mattter
> >> the way you run our qemu images, you still have to set a proper DNS, and
> >> have network connectivity in the image (which usually involves the host
> >> too) so defaulting to UseDNS no for qemu images seems a sensible default
> >> to me.
> > 
> > Well, v1 was unconditional in the openssh recipe; I think it still needs
> > to be just for qemuall.
> 
> It was originally for qemuall, but was also for all images and was to
> broad as some may not want it disabled in this way.

OK, I missed that; but v1 still made openssh machine-specific which is not 
desirable.

Cheers,
Paul

-- 

Paul Eggleton
Intel Open Source Technology Centre