⚙ D8334 eio: rely on efl_future_then to properly protect Eo object during the lifecycle of the future callback.

cedric created this revision.Wed, Mar 13, 2:58 PM

Herald added subscribers: committers, reviewers. · View Herald TranscriptWed, Mar 13, 2:58 PM

cedric requested review of this revision.Wed, Mar 13, 2:58 PM

cedric added a child revision: D8335: eio: path and filename property should always remind accessible even in case of error..

Harbormaster completed remote builds in B10263: Diff 20541.Wed, Mar 13, 2:59 PM

Rebase.

Harbormaster completed remote builds in B10304: Diff 20592.Thu, Mar 14, 2:28 PM

Rebase.

Harbormaster completed remote builds in B10350: Diff 20646.Fri, Mar 15, 4:58 PM

bu5hm4n added inline comments.Sat, Mar 16, 8:42 AM

src/lib/eio/efl_io_model.c
455	Just for me as a understanding point, this here could actaully also return NULL, right ?

cedric added inline comments.Sat, Mar 16, 10:30 AM

src/lib/eio/efl_io_model.c
455	It will always return an Eina_Value. Ideally one with a future, but if it can't create it, it will be an EINA_VALUE_EMPTY.

bu5hm4n added inline comments.Sat, Mar 16, 10:53 AM

src/lib/eio/efl_io_model.c
455	Okay, my question more clear: This future here in a eina value can also be just something else, since this future will not be assosiated with the promise that is resolved so `_eio_build_mime_now` is executed.

cedric added inline comments.Sat, Mar 16, 11:01 AM

src/lib/eio/efl_io_model.c
455	So it does just return a new future that will be resolved later using another job. The future as value actually is understood by the logic of Eina_Future and automatically plug this new future in the chain and will wait on it to resolve before continuing the chain. I am pretty sure I am hurting your brain again :-)

bu5hm4n added inline comments.Sat, Mar 16, 11:06 AM

src/lib/eio/efl_io_model.c
455	Can you point me to this logic ? Because the places i found are infact checking for a promise not for a future.

cedric added inline comments.Sat, Mar 16, 11:20 AM

src/lib/eio/efl_io_model.c
455	Oh, that's because there is no EINA_VALUE_TYPE_FUTURE only EINA_VALUE_TYPE_PROMISE. If you look at the code of eina_future_as_value in eina_promise.c, you will see how things are hooked via an intermediate promise.

We might want to consider renaming eina_future_as_value to something that does *not* suggest that the eina_value is a future :D

This revision is now accepted and ready to land.Sat, Mar 16, 11:35 AM

In D8334#150922, @bu5hm4n wrote:

We might want to consider renaming eina_future_as_value to something that does *not* suggest that the eina_value is a future :D

I don't know. It does take an Eina_Future and make it an Eina_Value. There is just some magic plumbing inside. I don't know if the user of the function has to know much about that magic.

In D8334#150925, @cedric wrote:

In D8334#150922, @bu5hm4n wrote:

We might want to consider renaming eina_future_as_value to something that does *not* suggest that the eina_value is a future :D

I don't know. It does take an Eina_Future and make it an Eina_Value. There is just some magic plumbing inside. I don't know if the user of the function has to know much about that magic.

Sure, but the function name indicates that the eina_value will be a future and not a promise, this is quite confusing to me. maybe @segfaultxavi has also an opinion on that ?

First off, thank you both for giving my brain the opportunity to hurt so much.

So, there is a method called eina_promise_as_value() which takes an Eina_Promise and wraps it in an Eina_Value.
And then there is eina_future_as_value() which indeed takes an Eina_Future and produces an Eina_Value. The mechanism, though, is obscure to me and I would appreciate an explanation. A proxy promise is created, and another callback is attached to the original future... but I get lost.

At any rate, eina_promise_as_value is very simple, whereas eina_future_as_value seems to be doing a lot of things. So yeah, until I understand better what is going on, I also think that "future as value" is misleading.

In D8334#151205, @segfaultxavi wrote:

First off, thank you both for giving my brain the opportunity to hurt so much.

You are welcome :-)

The idea for propagating a future into an Eina_Value without duplicating logic and adding a new type is to actually create a promise and put that into the Eina_Value. Then setting up a callback on the Eina_Future that was given and when it resolve manually trigger a resolve on the promise that was set inside the Eina_Value. This is like a manual pipe to get things across. Of course if the Eina_Value is freed, then it will cancel the callback. I hope this help.

Understood, but eina_future_as_value() does not wrap an Eina_Future in the value, it wraps an Eina_Promise, so... misleading :)

Anyway, that has nothing to do with this patch, right? I created another task for this: T7761.

Rebase.

Harbormaster completed remote builds in B10469: Diff 20808.Wed, Mar 20, 11:35 AM

Rebase.

Harbormaster completed remote builds in B10509: Diff 20880.Fri, Mar 22, 11:26 AM

bu5hm4n
zmike
barbieri
• stefan
segfaultxavi
stefan_schmidt
YOhoho

Diff	ID	Base	Description	Created	Lint	Unit
Base			Base
Diff 1	20541	48f1ed6		Wed, Mar 13, 2:58 PM	★	★
Diff 2	20592	823219b	Rebase.	Thu, Mar 14, 2:27 PM	★	★
Diff 3	20646	a1fe8a5	Rebase.	Fri, Mar 15, 4:58 PM	★	★
Diff 4	20808	a96ae58	Rebase.	Wed, Mar 20, 11:34 AM	★	★
Diff 5	20880	797f0a2	Rebase.	Fri, Mar 22, 11:25 AM	★	★

Status	Author	Revision
Needs Revision	cedric	D8450 elementary: make sure that the model parent being used is always the fileselector.
Accepted	cedric	D8449 elementary: prevent asynchronous properties change to believe target is ready when it is not in fileselector.
Accepted	cedric	D8448 elementary: fix fileselector entry support to define path manually.
Accepted	cedric	D8447 eio: guarantee that we will at least process one request per loop iteration for very slow system.
Accepted	cedric	D8375 elementary: destroy fileselector children when they are not itemized yet.
Accepted	cedric	D8374 eio: remove unecessary use of weak reference.
Accepted	cedric	D8373 eio: Efl.Io.Model should not make request when the object is invalidating itself.
Accepted	cedric	D8372 eio: add test to ensure proper lifecycle of Efl_Io object and futures.
Accepted	cedric	D8371 eio: work around the lack of integration between Ecore_Thread and Eina_Future.
Accepted	cedric	D8359 elementary: only apply text when the object is not invalidated and dying.
Accepted	cedric	D8429 Revert "efl_ui_layout: call efl_del instead of efl_unref when text part does not exist"
Accepted	cedric	D8358 eo: enforce auto_unref logic at the end of efl_unref execution.
Accepted	cedric	D8357 eo: refactor auto_unref logic used by efl_part.
Accepted	cedric	D8336 elementary: improve lifecycle of model object in the fileselector widget.
Accepted	cedric	D8335 eio: path and filename property should always remind accessible even in case of error.
Accepted	cedric	D8334 eio: rely on efl_future_then to properly protect Eo object during the lifecycle of the future callback.
Accepted	cedric	D8333 eina: prevent double cancel of ongoing dispatched future.
Accepted	cedric	D8332 elementary: split the smart callback event name from the Eo name when needed.

eio: rely on efl_future_then to properly protect Eo object during the lifecycle of the future callback.
AcceptedPublic
Actions

Details

Diff Detail

Revision Contents
Changeset List

Diff 20880

src/lib/eio/efl_io_model.c

eio: rely on efl_future_then to properly protect Eo object during the lifecycle of the future callback.AcceptedPublicActions

Details

Diff Detail

Revision ContentsChangeset List

Diff 20880

src/lib/eio/efl_io_model.c

eio: rely on efl_future_then to properly protect Eo object during the lifecycle of the future callback.
AcceptedPublic
Actions

Revision Contents
Changeset List