Question whether to free message in process_message on error

[quinnj]

I couldn't find anything explicitly documented as such, but I'm trying to confirm a pattern I found in code. For example, here, if the message sending fails for some reason, the handler itself will free the message. So in my channel handler vtable process_read_message implementation, I should only free the message once I've successfully handle it, right? i.e. if I have an error condition for whatever reason, then I shouldn't free the message.

I ask because I had originally done the equivalent of:

try
    internal_process_read_message(message)
finally
    aws_mem_release(alloc, message)
end

to ensure I always handled the message freeing, but then ran into rare edge cases where my internal_process_read_message implementation would throw (and thus return an error code from my vtable process_read_message) and then, I guess due to a race condition, MY aws_mem_release on the message would fail an assertion because the sending channel handler would have already released it.

Am I understanding this all correctly? I could do a quick PR to the channel handler vtable docstring to clarify.

[waahm7]

@quinnj Thanks for creating the issue. Yes, you are right. You must only free the message if you have successfully handled it. Sorry for the confusing documentation. I have created a PR #633 to update it.