Re: [OE-core] [PATCH 1/1] classes/kernel*: allow disabling symlink creation

["Paul Eggleton" <bluelightning@bluelightning.org>]

On Wednesday, 2 December 2020 05:22:43 NZDT Martin Jansa wrote:
> I see, I was just worried that this case will block me pushing the last bit
> of [YOCTO #12937] which changes symlinks to hardlinks:
> https://git.openembedded.org/openembedded-core-contrib/log/?h=jansa/artifact
> s
> 
> Which makes the use case you're fixing worse (as even less systems will
> correctly handle hardlinks), so I support the idea to disable having the
> links. My current work around is to just remove the unnecessary files when
> rsyncing to file-server.
> 
> But the main issue which I'm seeing with my current implementation (and
> with webos_deploy this is based on) is that we need the version-less
> artifacts to be available, because e.g. image recipe searching kernel
> artifact when there is DATETIME or SRCPV in the kernel artifact-name won't
> be able to find it, because the image might be built at different DATETIME
> and doesn't know anything about kernel SRCREV/SRCPV. To resolve this, the
> build is using the version-less artifacts (always the latest one).
> 
> I guess your implementation might fail in such DATETIME cases as well.
> 
> With separate variable to disable the links we can somehow work around this
> or at least warn user that without the version-less filenames, the
> VERSION_SUFFIX variable needs to be stable (as no PV/SRCPV/DATATIME in it).

That is a good point and we have the same problem here - what I'm planning to
do is to set a bunch of the other variables such that there are no versioned
artifacts - i.e. the real files effectively appear in place of the symlinks and
the versioned ones disappear. It seems to work in my tests so far. With my
patchset this is what I set:

------------ snip ------------
IMAGE_LINK_NAME = ""
IMAGE_VERSION_SUFFIX = ""
IMAGE_NAME_SUFFIX = ""
IMAGE_MANIFEST_distroname = "${IMGDEPLOYDIR}/${IMAGE_NAME}${IMAGE_NAME_SUFFIX}.manifest"

KERNEL_ARTIFACT_NAME = "${MACHINE}"
KERNEL_ARTIFACT_LINK_NAME = ""
------------ snip ------------

I have found that I still need the plain type named kernel artifacts (e.g. vmlinux),
and since these don't actually use the link name variable I have changed those to
be written out even if the link name is blank (v2 patchset still in testing). We could
introduce another variable to disable those but I hadn't planned to do that
for the moment.

Cheers,
Paul