[yocto] [PATCH] bb.build.addtask: add simple check for circular dependency

[Ulf Samuelsson]

> 18 jan. 2019 kl. 15:40 skrev Richard Purdie <richard.purdie at linuxfoundation.org>:
> 
> On Fri, 2019-01-18 at 15:16 +0100, Ulf Samuelsson wrote:
>>> 18 jan. 2019 kl. 12:29 skrev Richard Purdie <
>>> richard.purdie at linuxfoundation.org>:
>>> 
>>>> On Fri, 2019-01-18 at 05:30 +0100, Ulf Samuelsson wrote:
>>>> So if KNOWN_TASKS is defined in a configuration file in oe-core,
>>>> it
>>>> would be OK?
>>>> It needs to be extended to check for KNOWN_TASKS == None of
>>>> course.
>>> 
>>> No, this is too fragile. We then have to maintain two different
>>> lists
>>> of tasks. If we did that we may as well just remove the generic
>>> tasks
>>> mechanism and hardcode the tasks list in the metadata.
>>> 
>> Why do we need two lists?
> 
> If we change one of these "generic" tasks, we then need to remember to
> also update KNOWN_TASKS. The same information is being encoded in two
> places.
> 
> Maintaining the same information in two different places means one of
> those places inevitably ends up out of date.

We could insert a check if the KNOWN_TASKS is valid in a verification test.
When building core-image-minimal, the task order within core-image-minimal is analyzed, (dependencies on all other recipes is filtered away).
Then you find out if there are tasks added, missing, and/or in the wrong order.

Best Regards,
Ulf Samuelsson

> 
>> We need to know in what order the generic tasks are executed, that is
>> all.
>> 
>> Then we need to see if an addtask is violating that order.
>> If the ”after” or ”before” is not a generic task, the check cannot
>> catch that.
>> 
>> If the user decides to not use the generic tasks, and create new ones
>> with the same name, as the generic tasks, but rearranges the order,
>> there will be a problem, but I do not see any other problems.
>> 
>> Please enlighten me!
> 
> This has the problem that the code will find some subset of the bugs
> but not all of them. We'll then get users reporting that the tests
> failed and asking for the checks to be improved.
> 
> So no, we are not doing this, sorry.
> 
> Cheers,
> 
> Richard
> 
>