http://bugs.orocos.org/show_bug.cgi?id=852

Peter Soetens <peter [..] ...> changed:

What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
CC| |peter [..] ...
Resolution| |FIXED
Target Milestone|--- |next-bugfix

--- Comment #1 from Peter Soetens <peter [..] ...> 2011-03-30 17:28:37 CEST ---
(In reply to comment #0)
> Hi,
>
> I found out that bug when writing a std::vector into a port ten times.
> I did so by having the glibc malloc hook instrumented.
>
> The test shows that the 10th writing to the DataPort leads to
> an allocation of memory. That happens when the port is not read
> but only written.
>
> The analysis is that the 'operator=' of the std::vector class
> is used on a non-initialized receiver, thus the malloc.
>
>
> The bug is located in the DataObjectLockFree constructor (code taken from RTT
> 1.10 and 1.12
>
>
> // prepare the buffer.
> for (unsigned int i = 0; i < BUF_LEN-1; ++i) {
> data[i].data = initial_value;
> data[i].next = &data[i+1];
> }
> // BUG ! The following line is missing in the original code
> // --------------------------------------------------------
> data[BUF_LEN-1].data = initial_value;
> // --------------------------------------------------------
> data[BUF_LEN-1].next = &data[0];
> }
>
>
> I have quickly read the 2.3 version, and I think the bug is there, too,
>
> Adding the line above has solved my malloc issue.
>
> Cheers
> Thierry

Your analysis and fix are correct. Bug was present on both 1.x and 2.x
branches. I've patched both. Thanks a lot for reporting.

Peter