Re: [PATCH] Makefile: Fix separate output directory build of kselftests

[Muhammad Usama Anjum <usama.anjum@collabora.com>]

Hi,

At this time:
make kselftest-all (works)
make kselftest-all O=/tmp (works)
make kselftest-all O=build (doesn't work, investigation was shared)
make kselftest-all O=build/build2 (works)

I'd shared my final thoughts in the last email. I don't see any other
solution. If anybody has any other thoughts on how to cleanly fix `make
kselftest-all O=build`, do share.

Thanks,
Usama

On 4/4/22 4:09 PM, Muhammad Usama Anjum wrote:
> From [Makefile](https://elixir.bootlin.com/linux/latest/source/Makefile):
> ```
> ifeq ($(abs_srctree),$(abs_objtree))
>         # building in the source tree
>         srctree := .
> 	building_out_of_srctree :=
> else
>         ifeq ($(abs_srctree)/,$(dir $(abs_objtree)))
>                 # building in a subdirectory of the source tree
>                 srctree := ..
>         else
>                 srctree := $(abs_srctree)
>         endif
> 	building_out_of_srctree := 1
> endif
> ```
> `ifeq ($(abs_srctree)/,$(dir $(abs_objtree)))` condition is setting
> `srctree` to `..`. This is wrong. This condition isn't considering that
> `header_install` doesn't depend on `abs_srctree and abs_objtree`. This
> condition needs to be tweaked or removed for the `install_headers` to
> work fine and fix this issue. I've added `KBUILD_ABS_SRCTREE=1` to the
> kselftest target which sets the `srctree` to `abs_srctree` and thus
> forcefully affecting only kselftest targets. This seems like the clean
> fix. Alternatively we should remove this condition `ifeq
> ($(abs_srctree)/,$(dir $(abs_objtree)))` but it'll affect other targets
> as well.
> 
> Complete details of investigation can be found here:
> https://github.com/kernelci/kernelci-project/issues/92#issuecomment-1087406222
> 
> On 3/17/22 11:08 PM, Masahiro Yamada wrote:
>> On Thu, Mar 17, 2022 at 7:49 PM Muhammad Usama Anjum
>> <usama.anjum@collabora.com> wrote:
>>>
>>> Reminder. Shuah is okay with this patch. Any thoughts?
>>
>> I do not think this is the right fix,
>> but something you just happen to find working.
>>
>>
>> The Make is working in a wrong directory, that is why
>> the relative path does not work
>> (and you use the absolute path to work around it)
>>
> `ifeq ($(abs_srctree)/,$(dir $(abs_objtree))) \ srctree := ..` has
> broken the `make headers_install` when called through
> selftests/Makefile. We can remove it or use the absolute path each time.
> 

-- 
Muhammad Usama Anjum