[rabbitmq-discuss] erlang client API: transactions

[Ben Hood]

Edwin,

On Tue, Sep 23, 2008 at 5:49 AM, Edwin Fine
<rabbitmq-discuss_efine at usa.net> wrote:
> I believe this is coming from the Erlang client, in
> amqp_channel:rpc_bottom_half/2. I can't tell you why your program is doing
> this, only where the error message originates. It seems to be happening in
> response to receiving a tx.select_ok, AFAICS.
>
> If you look at the code for rpc_bottom_half (reproduced at the end of this
> email for convenience), the line
>
> {{value, {From,_}}, NewRequestQueue} = queue:out(RequestQueue)
>
> will create the error message you are seeing if RequestQueue is empty. This
> can clearly be seen in the shell:
>
> 1> EmptyQ = queue:new().
> {[],[]}
> 2> {{value, {From,_}}, NewRequestQueue} = queue:out(EmptyQ).
> ** exception error: no match of right hand side value {empty,{[],[]}}
>
> I don't know if this is a bug or one of those "can't happen" things that
> just didn't give a nice error message. I suspect the latter, because the
> code just cannot continue without the value of "From" to send a reply to.
> Why should the function have been called with a reply if the reply queue is
> empty? Something's wrong somewhere.
>
> In addition, maybe I am too tired to see straight, but this looks really
> suspect:
>
>     catch case queue:head(NewRequestQueue) of
>         empty ->
>             ok;
>         {NewFrom,Method} ->
>             Do2(Writer,Method)
>     end,
>
> This is because queue:head/1 exits when it is empty and does not return
> 'empty', so the catch will return an EXIT:
>
> 1> catch queue:head(queue:new()).
> {'EXIT',{empty,[{queue,head,[{[],[]}]},
>                 {erl_eval,do_apply,5},
>                 {erl_eval,expr,5},
>                 {shell,exprs,6},
>                 {shell,eval_exprs,6},
>                 {shell,eval_loop,3}]}}
>
> I think this should be
>
>     catch case queue:head(NewRequestQueue) of
>         {'EXIT', {empty,_}} ->
>             ok;
>         {NewFrom,Method} ->
>             Do2(Writer,Method)
>     end,
>

Thanks for pointing this one out - it was a branch of the code that
had never been visited by any tests before :-(

I've applied this patch to the client - see
http://hg.rabbitmq.com/rabbitmq-erlang-client/rev/ef7db7bfc51a

Currently it is only on that branch - this will be merged back into
the mainline when the peer QA is done.

Note that this is not the solution for the TX issue - that will be
addressed separately.

Ben