A tevent function to propagate error info up the stack

[Simo]

On 05/01/2013 02:50 PM, Stefan (metze) Metzmacher wrote:
> Am 01.05.2013 15:24, schrieb Richard Sharpe:
>> On Wed, May 1, 2013 at 5:29 AM, Simo <idra at samba.org> wrote:
>>> On 05/01/2013 01:24 AM, Richard Sharpe wrote:
>>>> Hi folks,
>>>>
>>>> This is designed to allow us to propagate error locations up the stack
>>>> so that DEBUG messages can tell us where the error really came from.
>>>>
>>>> Does this seem like a reasonable approach?
>>> Not sure it is.
>>> It would require you to keep around the subreq, which is something we do
>>> not do in tevent_req style requests.
>>> Why don't you simply ask for the location string and carry it on yourself ?
>>> Maybe a pair of functions, one to get the location and one to set it
>>> explicitly instead of implicitly so you can simply chain up this function:
>>>
>>> bool tevent_req_get_error(struct tevent_req *req,
>>> enum tevent_req_state *state,
>>> uint64_t *error,
>>> char **location);
>>>
>>> with this new one, by passing the retrieved location on your local state.
>>> location would be allocated on req and needs to be stolen as the req you
>>> retrieve it from (the subreq) is going to be freed soon enough, and
>>> usually before you use location.
>>>
>>> #define tevent_req_error_loc(req, error, location) \
>>> _tevent_req_error(req, error, __location__)
>> I wasn't actually going to keep the subreq around, only use it to grab
>> the location pointer.
>>
>> However, we are going to need a function to get the location so what
>> you suggest might be the simplest approach.
> I had the idea of keeping a real error stack, but never finished a
> design for it.
>
> My idea was to have an opaque struct tevent_req_error_stack
> (or maybe just tevent_req_stack in order to make it more generic and also
> use it for success) and somehow pass it between the layers.
> special tevent_req_debug() functions could then also add additional
> information.
>
> The toplevel caller would initialize the error stack and pass it down
> and when it receives an error it could call a function to get ndr_print
> like representation
> of the error stack.
>
> Most times it's really the top level caller who has to decide if it's
> worth to print out a DEBUG message.
> For a SMB request from the client we would have such stack object, then
> we could decide to just
> debug anything if the whole request failed.
I would like this very much.

Simo.