Re: [lwip-users] shouldn't happen?

[Leon Woestenberg]

Tim,

Tim Newsham wrote:
> I checked out the CVS src tree today, and am playing around with it.
> in etharp.c:etharp_output there are some lines:
>
>   /* never reached; here for safety */
>   pbuf_free(q);
>   return result;
>
> reading the code, I dont see why it should never happen.  In
> fact, it is happening.  Worse, its getting called by the
> TCP retransmit function and happening twice, decrementing
> the reference count to zero, and then asserting when it
> tries to pbuf_free the buffer after its ref count is zero.
>
> Is this a bug in the code?  Is the current cvs tree unstable
> (should I be playing with a different version)?
I have investigated the etharp.c history:

This bug was introduced from 1.59 to 1.60, which involved a
major rewrite of the pbuf output flow.

In fact, etharp_output should never free any pbufs to start with.

Also, the "never reached" statement referred to the old code, it
should have been removed.

I have fixes to CVS HEAD.

Regards,

Leon.