Re: DataObjectLockFree: fix facts in doxygen

On Tue, 11 Dec 2012, Sébastien Barthélémy wrote:

> On Tue, Dec 11, 2012 at 3:50 PM, Herman Bruyninckx <
> [email protected]> wrote:
>
>> Anyway, the Lock Free pattern has as pre-condition for its use that there
>> are
>> maximally N clients of the data, so that it can foresee N+1 buffer spaces
>> of the shared data.
>> Did you then still find problems with the code in cases where this
>> pre-condition is satisfied?
>
>
> Yes.
>
> Among the N clients there can be only one writer. This additional
> precondition is clear from the implementation and even more clear from the
> comments at line
> 119.<http://gitorious.org/orocos-toolchain/rtt/blobs/master/rtt/base/DataObjectLockFree.hpp#line194>
> Alas, it was not documented. Worst, the doxygen claimed the contrary (see
> the truth table).
>
>
>> Since that would be a big bug, I guess.
>>
>
> I do not know if having several writers for a non-buffered data port is a
> common scenario.

It should not! Because there can be only one "owner" of a particular piece
of newly created data. But I'm all about making these "hidden assumptions"
more clear (even though they are "natural"), so thanks for your noticing
these problems!

> But it would be a hard to find one certainly, since there is no lock nor
> warning, only random data corruption.
>
> Or are you trying to add more robustness into the implementation in case
>> the pre-condition is not satisfied?
>>
>
> no, I'm just fixing the doc so that it matches the implementation.
>
> Cheers

Best regards,

Herman Bruyninckx

-- 
Orocos-Dev mailing list
[email protected]
http://lists.mech.kuleuven.be/mailman/listinfo/orocos-dev