[Buildroot] [RFC PATCH v5 06/11] pkg-generic: add GLOBAL_INSTRUMENTATION_HOOKS for rpath sanitation

[Wolfgang Grandegger]

Hello,

Am 04.07.2017 um 10:57 schrieb Wolfgang Grandegger:
> Hello Arnout,
> 
> Am 04.07.2017 um 10:22 schrieb Wolfgang Grandegger:
>>
>>
>> Am 04.07.2017 um 00:30 schrieb Arnout Vandecappelle:
>>>
>>>
>>> On 03-07-17 23:47, Arnout Vandecappelle wrote:
>>>>
>>>>
>>>> On 30-06-17 10:37, Wolfgang Grandegger wrote:
>>>>> The hook calls the script "fix-rpath" at the end of the installation
>>>>> step.
>>>>>
>>>>> Signed-off-by: Wolfgang Grandegger <wg at grandegger.com>
>>>>
>>>>   Looks good to me. I'm not adding my reviewed-by yet because it 
>>>> will still
>>>> change when the fix-rpath arguments change.
>>>>
>>>>   Regards,
>>>>   Arnout
>>>>
>>>>> ---
>>>>>   package/pkg-generic.mk | 10 ++++++++++
>>>>>   1 file changed, 10 insertions(+)
>>>>>
>>>>> diff --git a/package/pkg-generic.mk b/package/pkg-generic.mk
>>>>> index 825ab0c..8812193 100644
>>>>> --- a/package/pkg-generic.mk
>>>>> +++ b/package/pkg-generic.mk
>>>>> @@ -139,6 +139,16 @@ define step_check_build_dir
>>>>>   endef
>>>>>   GLOBAL_INSTRUMENTATION_HOOKS += step_check_build_dir
>>>>> +define step_sanitize_rpath
>>>>> +    $(if $(filter install-host-end,$(2)-$(1)),\
>>>>> +        support/scripts/fix-rpath "host" 
>>>>> $($(PKG)_DIR)/.br_host_filelist)
>>>>> +    $(if $(filter install-staging-end,$(2)-$(1)),\
>>>>> +        support/scripts/fix-rpath "staging" 
>>>>> $($(PKG)_DIR)/.br_staging_filelist)
>>>>> +    $(if $(filter install-target-end,$(2)-$(1)),\
>>>>> +        support/scripts/fix-rpath "target" 
>>>>> $($(PKG)_DIR)/.br_target_filelist)
>>>
>>>   After reviewing the qt.conf thing, I realized that this is in fact 
>>> not enough.
>>> Indeed, staging-install may install things into host dir as well...
>>>
>>>   So I think that both for generating the lists and for doing 
>>> fix-rpath, we
>>> should not just look at the directory corresponding to the step, but 
>>> to all three.
>>>
>>>   Since that means that the same directory may be iterated over three 
>>> times, I
>>> think it's best to also make a total of 9 files:
>>>
>>> .br_host_install_host_filelist   -> things installed to host during 
>>> host-install
>>> .br_host_install_target_filelist -> things installed to target during 
>>> host-install
>>> .br_target_install_host_filelist -> things installed to target during 
>>> target-install
>>
>> Or we store the full path of installed files of all trees in one 
>> .br_install_filelist. fix-rpath could then act upon the path prefix 
>> (being equal to $HOST_DIR, etc.).
> 
> That would even simplify things (in pkg-generic.mk). Either I use the 
> full path or add the name of the tree before the relative file name:
> 
>    host "file-name"
>    staging "another-file-name"
> 
> Then fix-rpath would scan that file and sanitize the file according to 
> the tree type. I think using the absolute path is more efficient as
> I can scan in fix-rpath directly for the files with that name. If it's 
> an ELF file, it acts upon the rootdir prefix (host, staging or target).
> 
> What do you think?

Thinking more about it... Generating the list of all installed files 
over and over (for *each* packet *twice*) again is not very efficient.
Here some measurement on my rather fast i7 7700K PC with a rather big BSP:

$ cd <a-host-dir>
$ time find . -type f -print0 | xargs -0 md5sum | sort > ~/junk/filelist
real	1m43.422s

Doing it once at the end is likely more efficient. To avoid rpath 
sanitization with subsequent makes, we could maybe used some tags.

Wolfgang.


> 
> Wolfgang.