24 Nov
2011
24 Nov
'11
8:15 p.m.
On 11/24/2011 05:31 PM, Vincent Massol wrote:
On Nov 24, 2011, at 4:55 PM, Denis Gervalle wrote:
I sort of agree with Dennis on this one. Numbered priorities always have
conflicts, and in order to make sure you place yourself in the proper
place (before this stuff, after this other stuff) you need to check (in
the code? in a documentation about priority numbers?) the numbers that
stuff are using.
Or did I understand the proposal wrong and it's not about priority numbers?
It's really a good idea BUT, as we've discussed before, numbered
priorities are always very tricky, since you can never be sure what it
means when you put 100 or 0 or some other number x (what do other
listeners declare? will my listener be before or after a certain other
listener?).
I would think all the number-based priority levels endup being used as a
static list-based priority levels (there are a few well-known values
that are used), since you can't really use a continuous interval (you
need "before this standard listener" or "after this other standard
listener", if we're talking about non-standard listeners, numbers are
not much more helpful since you can never know what will be all the
priorities declared by all those listeners and where would yours be placed).
Thanks,
Anca
On Thu, Nov 24, 2011 at 16:25, Vincent
Massol<vincent(a)massol.net> wrote:
Yes it's static
dependency vs loose dependency. One is statically typed the other is documentation.
But that's the only way I know to not have modules depend on each other.
I like the static way of course but what you proposed while it's good for this
specific use case doesn't solve all other use cases, which is why I'm ambivalent
about it and why I think the generic solution is a bit better (I could be convinced
otherwise with enough arguing and if enough devs prefer the non generic way ;)). On Nov 24, 2011, at 4:06 PM, Denis Gervalle
wrote:
Well, I have
said +1 anyway, but...
On Thu, Nov 24, 2011 at 13:58, Vincent
Massol<vincent(a)massol.net>
wrote:
> Hi devs,
>
> Summary:
> ========
>
> I'd like to add the notion of Priority to Event Listeners. The reason is
> that in some cases it's important that some listeners execute before
others.
> The problem at hand:
> =================
>
> Here's a typical use case: When receiving the ApplicationStartedEvent,
we
> have lot of code that needs to initialize.
Initialization order is
> important (you can compare it to run levels in OS): for example some
init
> must happen after the Database initialization
has happened.
>
> Note that another solution exists for this use case: some
initializations
> could introduce their own events (such as a
DatabaseStartedEvent) and
other
> init could listen on those events instead of
the generic
> ApplicationStartedEvent. However I can see several drawbacks to this:
> * it's less generic than the priority solution
> * it means creating more events
> * but more importantly it means that modules will have strong
dependencies
> (at maven level) on each other whereas
it's not necessary and shouldn't
be
> the case. In our example use case: it means
that inits that must happen
> after database is started will need to depend on oldcore (which is
where DB
> is started ATM)
>
> Proposal:
> ========
>
> * Don't break backward compat in Observation module
> * Introduce a PrioritizedEventListener interface that adds a
getPriority()
> method
> * Modify ObservationManager implementation to take into account
priorities
> * In order to make it simple I propose to
have only a single priority
per
> Listener and not a priority per event
supported by a given listener
>
> General Context
> =============
>
> To give some context here's what I'd like to do on the medium term:
>
> * Step 1: Introduce notion of priority in EventListeners
> * Step 2: Refactor XWiki init to use an EventListener on AppStarted with
> low priority
> * Step 3: Refactor wiki macros to use an EventListener on AppStarted
with
> priority value lower than at step2
> * Step 4: Write an EventListener for the new UI Extensions with a
priority
> value higher than the one of step2<-- this
is the initial goal that
led
> me to make this proposal ;)
>
> WDYT?
>
Sounds good if not overkill for the goal.
An simpler alternative would be to have more than a single AppStarted
event, like there is more than one starting level in Linux.
Let say one level before XWiki, the one during, the one after ? is there
so
many other use case ?
Yes this is very close to the other solution I
explained above.
Surely.
But it's far less generic and introduces
knowing stuff you don't really
need to know. For me it's a "poorman" implementation of priorities.
Well, it depends on the pursued goal. What I mostly dislike in priority
systems, it the management of priorities. When you say that you are
listening on a let says DatabaseStarted event, you clearly explain what you
really need. On the other hand, when you say that you want to be run at
priority 100 of the AppStartedEvent, without another document saying that
starting at 100, database is ready, and ensuring this in the database
module, you do not really know what it means really.
Thanks
-Vincent
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
I
don't know about other use cases right now but I can imagine a lot of
them (for example, some extensions that needs open office to be
initialized, etc). We need to remember that we're developing a platform and
not just for XE's needs. The extra amount of work is negligible IMO not to
do it (btw I have a working version locally already, I just need to fine
tune it a bit for improved performance maybe).
Thanks
-Vincent
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs
--
Denis Gervalle
SOFTEC sa - CEO
eGuilde sarl - CTO
_______________________________________________
devs mailing list
devs(a)xwiki.org
http://lists.xwiki.org/mailman/listinfo/devs