On 10/03/16 09:57, Tom Hacohen wrote:
> On 10/03/16 06:46, Jean-Philippe André wrote:
>> On 10 March 2016 at 15:05, Carsten Haitzler <ras...@rasterman.com> wrote:
>>
>>> On Thu, 10 Mar 2016 07:42:22 +0200 Daniel Zaoui <daniel.za...@samsung.com>
>>> said:
>>>
>>>> On Wed, 09 Mar 2016 16:23:04 +0000
>>>> Tom Hacohen <t...@osg.samsung.com> wrote:
>>>>
>>>>> On 03/03/16 10:22, Tom Hacohen wrote:
>>>>>> On 01/03/16 09:05, Tom Hacohen wrote:
>>>>>>> Hey,
>>>>>>>
>>>>>>> The Eo syntax is going to be changing once more, and this time, I
>>>>>>> really think/hope it'll be the last time. We plan on stabilizing
>>>>>>> Eo and all of the functions on top of it in the next few months,
>>>>>>> so that doesn't leave us much more time to change it again. :)
>>>>>>>
>>>>>>> These changes will remove the need for the eo_do family of
>>>>>>> functions. Functions will now look like normal C functions (which
>>>>>>> they are). There are many benefits to that, and we have many cool
>>>>>>> new ideas.
>>>>>>>
>>>>>>> For more info: https://phab.enlightenment.org/w/eo/
>>>>>>>
>>>>>>> I'm sending this email as an head's up, as I'll be starting to
>>>>>>> work on migrating to the new Eo syntax (and implementing it)
>>>>>>> today. Felipe and I have actually already started (needed to for
>>>>>>> the PoC), but I plan on pushing my changes to master soon.
>>>>>>>
>>>>>>> If you have any issues/suggestions/comments with the proposal,
>>>>>>> please let me know, either in pm, irc or just here.
>>>>>>>
>>>>>>
>>>>>> Changes are in! I still haven't migrated eo_add to the new syntax
>>>>>> (it uses a non portable gcc extension in the meanwhile), but
>>>>>> otherwise everything is in. Took me *much* less time than I thought
>>>>>> it would, so yay. :P
>>>>>>
>>>>>> I decided to push it now instead of letting it rest in my branch
>>>>>> for a while because literally every hour that passed introduced
>>>>>> more merge conflicts for me, so the benefits from stabilising it
>>>>>> more in my branch were diminished by the new conflicts and issues
>>>>>> that could arise.
>>>>>>
>>>>>> If you have an application that uses the Eo api, you can use my
>>>>>> script https://devs.enlightenment.org/~tasn/migrate_eo.py to
>>>>>> migrate your code. When using the script you should keep two things
>>>>>> in mind: 1. You are only allowed to run it *once* per source code,
>>>>>> because the changes to eo_add() would otherwise accumulate and your
>>>>>> code will be wrong. If you need to correct something you've done
>>>>>> wrong, reset the code to the previous state and run the script
>>>>>> again on the original code. 2. The migration script is not perfect.
>>>>>> In particular it can't deal with some corner cases like:
>>>>>> eo_do(obj, a_set(1),
>>>>>> /* b_set(2),
>>>>>> g_set(4), */
>>>>>> c_set(2));
>>>>>> Or abominations like:
>>>>>> eo_do(obj, if (a_get())
>>>>>> do_something());
>>>>>>
>>>>>> So please be aware of that and *manually* review your changes after
>>>>>> the script has run.
>>>>>>
>>>>>> If your code does have these cases, I recommend you either get rid
>>>>>> of them, or manually migrate that code before running the script
>>>>>> (remove the relevant eo_do).
>>>>>>
>>>>>> Follow the wiki page mentioned in the previous email for more
>>>>>> information about Eo and what else needs changing.
>>>>>>
>>>>>> Please let me know about any regressions (there shouldn't be any)
>>>>>> or any issues you may face.
>>>>>
>>>>> I'm now pushing my changes to eo_add. I'm pushing it now for the same
>>>>> reason I pushed the previous changes in.
>>>>>
>>>>> I created a new script that assumes the code has already been
>>>>> migrated with the previous (migrate_eo.py) script. This script is
>>>>> called migrate_eo_add.py and can be found at:
>>>>> https://devs.enlightenment.org/~tasn/migrate_eo_add.py
>>>>>
>>>>> When using the script you should keep two things in mind:
>>>>> 1. You are only allowed to run it *once* per source code, because the
>>>>> changes to eo_add() would otherwise accumulate and your code will be
>>>>> wrong. If you need to correct something you've done wrong, reset the
>>>>> code to the previous state and run the script again on the original
>>>>> code. 2. The migration script is not perfect. In particular it can't
>>>>> deal with cases like missing {} for if/for/while content so for
>>>>> example,
>>>>>
>>>>> if ()
>>>>> return eo_add(...)
>>>>>
>>>>> would break.
>>>>> 3. If you are fancy and use the same variable inside eo_add and
>>>>> outside, for example like:
>>>>> parent = eo_add(CLASS, parent);
>>>>>
>>>>> your code will break. I suggest you use a temporary variable.
>>>>>
>>>>> So please be aware of that and *manually* review your changes after
>>>>> the script has run.
>>>>>
>>>>> If your code does have these cases, I recommend you either get rid of
>>>>> them, or manually migrate that code before running the script (remove
>>>>> the relevant eo_do).
>>>>>
>>>>>
>>>>>
>>>>> Sorry, but C++ will break until the C++ guys fix it. I'm now in the
>>>>> process of migrating the rest of our applications. Hopefully this
>>>>> will be the last disruption of this sort.
>>>>>
>>>>
>>>> Sorry man but the new syntax is ugly. I still don't see why this change
>>> was
>>>> needed. Please enlighten me. It reminds me the wonderful eo_do_ret
>>> syntax :-)
>>>>
>>>> So yes Tom I vomit on your eo_add
>>>>
>>>>
>>>> BBEEEUUUUUUAAAAAHHHHH...
>>>>
>>>> %%%%%%
>>>> %%%% = =
>>>> %%C >
>>>> _)' _( .' ,
>>>> __/ |_/\ " *. o
>>>> /` \_\ \/ %`= '_ .
>>>> / ) \/| .^',*. ,
>>>> /' /- o/ - " % '_
>>>> /\_/ < = , ^ ~ .
>>>> )_o|----'| .` '
>>>> ___// (_ - (\ eo_add(&obj...
>>>> ///-( \' \\
>>>
>>> bwhahahahahha.
>>
>>
>>> the reason was eo add methods.
>>>
>>> obj = eo_add(..., text_set(obj, "x"), color_set(obj, 1, 2, 3, 4));
>>>
>>> because eo4 changes to pass obj into every method - that means obj has to
>>> be
>>> filled and defined with the RIGHT eo id before the extra text_set() is
>>> called
>>> because it passes it in, thus you have to pass a ptr to the eiod so it can
>>> be
>>> filled in first so it is correct for the following calls within the eo_add.
>>>
>>
>> I know you don't like it but I can see two solutions to that:
>> 1. a different macro for eo_add() that doesn't allow any function calls
>> before finalize (I believe it would be used quite often)
>> 2. use a tls to store the currently created obj and add a macro to get it
>> in those inlined function calls, eg.
>> obj = eo_add(CLASS, parent, do_something(eo_cur))
>>
>> wrt. 1. I wonder how the bindings will even be able to create objects and
>> call functions before finalize?
>
> As you explained to the vomiting boy somewhere else in the thread, the
> reason why I had to switch is portability.
>
> 1. I don't know how often that will be used.
> 2. You can't do that unfortunately. TLS is sucky and ugly and I would
> have hated using it, but I would have if it was possible, which is
> unfortunately not. You can't even use it in a single threaded
> environment (that's why we had the eo stack before). To see where it
> breaks, just consider an object that adds a subobject in the constructor.
>
> I don't see what you mean an what's the problem...
Btw, there are ways to make it work with the old syntax, for example, we
could reintroduce a stack for eo_add() like we had before (and make it
tls), and then have a magic macro EOID that returns the top of the stack
for every function. This is a question of what we prefer to compromise
on, complexity or syntax?
Choosing syntax has the advantage of allowing us to change to ({}) in
the future when it becomes part of the standard without any need to
change the API, or heck, even maybe autodetect (based on compiler) if we
can use ({}) so if we are allowed we automatically switch to this which
is faster than the hacky way with the stack. We can switch both on the
fly, some can use one and some can use the other no problem...
What do you think about that? do we want to go back to the old syntax
and have this magic behind the scenes that makes it fast on most
platforms (gcc + clang) and a bit slower on ones that don't support this
gcc extension?
I think that's a reasonable solution and would be happy to go forward
with it if people agree.
--
Tom.
------------------------------------------------------------------------------
Transform Data into Opportunity.
Accelerate data analysis in your applications with
Intel Data Analytics Acceleration Library.
Click to learn more.
http://pubads.g.doubleclick.net/gampad/clk?id=278785111&iu=/4140
_______________________________________________
enlightenment-devel mailing list
enlightenment-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
