[OE-core] [PATCH v5] do_image: Implement IMAGE_ROOTFS_EXCLUDE_PATH feature.

[Kristian Amlie]

On 23/08/17 15:19, Kristian Amlie wrote:
> On 23/08/17 14:47, Richard Purdie wrote:
>> On Wed, 2017-08-23 at 14:39 +0200, Kristian Amlie wrote:
>>> From: Kristian Amlie <kristian.amlie at mender.io>
>>>
>>> This is a direct followup from the earlier f6a064d969f414 commit in
>>> wic. It works more or less the same way: The variable specifies a
>>> list
>>> of directories relative to the root of the rootfs, and these
>>> directories will be excluded from the resulting rootfs image. If an
>>> entry ends with a slash, only the contents are omitted, not the
>>> directory itself.
>>>
>>> Since the intended use of the variable is to exclude certain
>>> directories from the rootfs, and then include said directories in
>>> other partitions, it is not natural for this variable to be respected
>>> for image creators that create multi partition images. Therefore the
>>> default is to ignore the variable, and image creators that create
>>> single root filesystems need to set
>>>
>>>   do_image_myfs[respect_exclude_path] = "1"
>>>
>>> in order to honor it. Specifically, "wic" and "multiubi" have not
>>> received this variable flag, while others have.
>>
>> I have to wonder if this is implemented the correct way around. Would
>> it not be better to mark multiubi and wic as not wanting this feature,
>> rather than requiring every other image creation method (which may even
>> be in other layers) to set this?
> 
> The idea was that people in downstream repositories (other layers) are
> more likely to have their own complete image creators than to have
> filesystem creators. So rather than requiring all of the downstream
> repositories to opt out of the feature, we opt in in the filesystem
> creators which are mostly contained in OE.
> 
> Or so was the thinking. Not sure if it's a good reason though. I can
> invert the behavior if you think it's better!

Since I heard nothing more I assume that it's better inverted, and I
kind of agree as well, the patch is much smaller. So here follows the
next patch, the changes from the last are:

* Inverted default of the respect_exclude_path flag, now on by default

* Added a debug message when it's turned off, to lessen confusion.

* Documentation is now stripped out as it should be, I will send it as a
separate patch as soon as we've concluded on this patch.

-- 
Kristian