[vdr] Re: thread safeness II (was: Re: Re: solved: segmentfaults from osdtelext plugin)

[Clemens Kirchgatterer <clemens@1541.org>]

Klaus Schmidinger <Klaus.Schmidinger@cadsoft.de> wrote:

> > > Too the problem does not show up when it is caused.
> > > For example the above implicit "free" may release memory that is
> > > still pointed by a task which is sleeping for 24h hours.
> > > So the segfault may occure 24h later, but only if meanwhile
> > > nothing alloced memory in the same area. Then that third, totally
> > > unrelated function may show "funny effects"...happy debugging...

> > i would like to hear a comment from klaus, if he is sure not to call
> > any of these functions with static locals from different threads.
 
> I am currently working in a totally different area (audio track
> handling, AC3 over DVB via firmware) and so I'm afraid I can't get
> involved in this discussion right now. I'll get back to this later.

<WildGuessing>

so, you are not sure about vdrs thread safety. this also seems to be
prooved by the fakt that only some people see random crashes and it is
_cured_ sometimes by attaching gdb. also that some plugins seem to
influence this behalfier is a good hint.

</WildGuessing>

anyway, IMHO the functions having static locals should be kept to an
absolute minimum in a multithreaded application and where they can't be
avoided (?) there must be mutex locks around them. the same applys to
globals. even if vdr itself does not introduce race conditions, who
knows what plugins do?

best regards ...
clemens