[Buildroot] [PATCH v2 02/13] qt5: Convert to generic target install command

Arnout Vandecappelle arnout at mind.be
Thu Mar 14 11:20:14 UTC 2019



On 14/03/2019 12:06, Thomas Petazzoni wrote:
> On Thu, 14 Mar 2019 12:01:16 +0100
> Arnout Vandecappelle <arnout at mind.be> wrote:
> 
>>> One issue with this approach is that if you do a complete build, and
>>> then do "make qt5base-reinstall", then the date of the .stamp_built
>>> file of qt5base has not changed, and will be older than the one of all
>>> other packages that have been built after qt5base. Therefore, the rsync
>>> that you do will copy again not only the files from qt5base, but also
>>> the ones from all other packages that have been built after qt5base.  
>>
>>  I agree that this is a problem.
>>
>>  But I think it can be easily solved by using the pkg-file-list instead of find.
>> Since we already have a dependency of target on staging, we know that the
>> pkg-file-list is up-to-date.
> 
> True.
> 
>>> I continue to believe we want to find a solution where "make
>>> INSTALL_ROOT=$(TARGET_DIR) install" and "make
>>> INSTALL_ROOT=$(STAGING_DIR) install" works properly.  
>>
>>  I disagree. I think the pattern of installing to staging and then copying to
>> target is a good pattern.
> 
> We basically don't use this pattern anywhere.

 A quick grep gives me alsa-utils, dmalloc, libfuse, liblockfile, qt5*. So it is
rare, but "don't use it anywhere" is an overstatement.

> I would agree with the
> general direction of killing the staging vs. target difference, and
> simply produce "target" out of "staging" at the very end of the build,
> but that's a huge undertaking.

 So my point is: using this pattern for a few packages is a good step in that
direction, because it teaches us something about the implications.

 Regards,
 Arnout

> 
> Best regards,
> 
> Thomas
> 


More information about the buildroot mailing list