[Pharo-dev] [ bloc ] I do not understand why some behavior is not in the right place

Nicolai Hess nicolaihess at gmail.com
Wed Apr 6 03:35:17 EDT 2016

2016-04-05 16:31 GMT+02:00 Igor Stasenko <siguctua at gmail.com>:

> On 5 April 2016 at 17:27, Igor Stasenko <siguctua at gmail.com> wrote:
>> On 5 April 2016 at 16:29, Aliaksei Syrel <alex.syrel at gmail.com> wrote:
>>> Now let's take a look at this code:
>>> drawOnSpartaCanvas: aCanvas
>>>> aCanvas
>>>>   clipPreserveBy: self shape during: [
>>>>   aCanvas paintGroup: [
>>>> aCanvas setPaint: self shape fillPaint.
>>>> aCanvas fillPreserve.
>>>> aCanvas paintMode source.
>>>> aCanvas setStrokePaint: self shape strokePaint.
>>>> aCanvas stroke ] ]
>>> You may be curious why it is so ugly :) Make it work - make it right -
>>> make it fast. We are on the first etappe, because I invested zero time in
>>> rendering stuff.
>>> What you see is the minimal amount of cairo primitive calls that are
>>> needed to render not overlapping fill and stroke. Clipping is needed to
>>> make sure that stroke does not get rendered outside of a path. Group is
>>> needed to have transparent target in order to make source paint mode work
>>> as expected. Compared to image_surface group, it in this case allows to
>>> preserve clip and current cairo state which is pushed to stack during
>>> push_group and restored during pop_group_to_source. fillPreserve allows to
>>> reuse the same path as used for clipping before saving cpu time on loading
>>> path.
>>> Yes, i understand that. You are forced to do that. And it is ugly not
>> because of all you listed above, it ugly because you could just use a
>> cascade:
>> aCanvas setPaint: self shape fillPaint;
>>   fillPreserve;
>>    paintMode source;
>>    setStrokePaint: self shape strokePaint;
>> stroke
>> (something like that)
>> but yeah.. that can wait .. since it is still work in progress. I agree.
>>> It is implemented in canvas specific method after dispatch though
>>> canvas, so we are allowed to use canvas specific api, for example groups.
>>> How to model stroke, fillPreserve and paintModein terms of Athens?
>> A good question. And i don't have an answer to it , ready for you.
>> 1. Stroke can be expressed as a special kind of paint. And that how its
>> done in Athens.
>> 2. What fillPreserve is, is can't find it in source code? Some
>> hot-swapping and preserving context state , i guess.
>> 3. As for paint mode, it is already in Athens, so why you asking? You
>> don't like how it is done or what?
> or maybe you meant how to group those operation and express them as
> command group?
> I have no simple answer here. Because this is root points of the core of
> graphics engine. From one side, you want such things be exposed to user,
> and from other you want a higher dimension concepts/operations to be
> allowed by combining those.
> There's no simple way. I would just stop at this level, letting user to
> decide how he wants to play with those pieces to achieve results he wants.

Grouping and Context save/restore are good additions - I think.

But it is true, that we should care about the api of Athens and not just
add things that happens to be possible, because we use cairo as a backend.

I made some fixes for AthensBalloon (not all are integrated yet, some parts
are just experimental and needs more tests), the idea is to have
AthensBalloon at least not crash or throwing  errors, even if not all
features are supported.
But at the moment, no one cares about non-cairo-athens. All users just
directly use AthensCairoCanvas/AthensCairoSurface.
I had a bug report for discussion about how to make some kind of factory
that would create the appropriate Athens backend.

Is there still some interest on AthensBalloon or to make athens more
independent from cairo?

>>> Cheers,
>>> Alex
>>> On Tue, Apr 5, 2016 at 3:15 PM, Aliaksei Syrel <alex.syrel at gmail.com>
>>> wrote:
>>>> Hello Igor
>>>> Thanks for extensive design explanation and effort!
>>>> Issues you mentioned in previous emails are important and need to be
>>>> addressed :)
>>>> fill(), stroke() fillPreserve() strokePreserve() need to disappear in
>>>> the end. We will come back to them later.
>>>> Let me tell a few words about Sparta.
>>>> Sparta implements Athens interface api (excluding some experimental
>>>> stuff to test possible performance boost in a few places) and does not have
>>>> task to remove Athens style and abstractions. Ideally Sparta will be
>>>> AthensCairo for bloc. I'm looking forward for your help :)
>>>> Here are some aspects in AthensCairo that Sparta tries to address in
>>>> first place:
>>>>    - *Clipping in local coordinates*. It is critical in Bloc. You
>>>>    implemented AthensCairo to have vector based rendering in Morphic and Pharo
>>>>    in general. Morphic lives in global coordinates, so your choice to clip in
>>>>    global coordinate is perfect! At the same time global clipping in bloc adds
>>>>    complexity. Sparta clips always in local coordinates (user space in cairo
>>>>    terminology).
>>>>    - *Clip by arbitrary path*. Athens and AthenCairo expect to see
>>>>    aRectangle as clipping region - your wise choice for morphic. In bloc I
>>>>    would have clipping by arbitrary path. clipBy:during: gets aPath.
>>>>    Rectangle/Color is polymorphic with path/paint in Sparta
>>>>    - *Support of groups*. (maybe user-level aspect? like shadows)
>>>>    Groups are powerful in cairo (do they exist outside of cairo?) and allow to
>>>>    draw both transparent fill and stroke without overlapping using only one
>>>>    path. On class side of BlElement there are examples (exampleCircle) that
>>>>    show such behavior.
>>>>    - *Do not maintain and set pathTransformation before each
>>>>    render-dependent action.* Questionable but what if Canvas will not
>>>>    maintain current state of pathTransform? Instead all transformations can be
>>>>    directly applied on cairo_t using native calls. If there is a need to get
>>>>    actual matrix we can ask cairo directly. From my perspective it simplifies
>>>>    transformation stuff a little bit.
>>>>    - *Benefit from cairo_save and cairo_restore.* AthensCairo
>>>>    maintains state manually by setting transformation matrix and clip. Instead
>>>>    we could save and restore state without caring about clip/matrix which
>>>>    simplifies code. Check SpartaCanvas>>#clipBy:during:
>>>> Cheers,
>>>> Alex
>>>> On Tue, Apr 5, 2016 at 2:12 PM, Igor Stasenko <siguctua at gmail.com>
>>>> wrote:
>>>>> Couple more words about that fill() function abstraction.
>>>>> Now you probably understand why there's no notion of stroke operation
>>>>> in Athens.
>>>>> Because instead of introducing it that way, by adding new kind of a
>>>>> function
>>>>> stroke(shape,paint)
>>>>> from our perspective, it falls into our more generic fill() function,
>>>>> except that
>>>>> instead of literally filling the shape we deciding to paint a stroke:
>>>>> fill(shape, strokePaint).
>>>>> As i said, there's nothing that tells that fill() function must affect
>>>>> only areas enclosed by the shape.
>>>>> For instance, you could imagine, that i'm in contrary, may want to
>>>>> fill everything , but the area(s) enclosed by given shape. And that still
>>>>> can be represented as invocation of our generic fill() function, except
>>>>> that we will use a different kind of paint, that will fill everything
>>>>> outside designated region, i.e.:
>>>>> fill(shape, fillOutsidePaint)
>>>>> On 5 April 2016 at 14:33, Igor Stasenko <siguctua at gmail.com> wrote:
>>>>>> On 5 April 2016 at 04:00, Ben Coman <btc at openinworld.com> wrote:
>>>>>>> On Tue, Apr 5, 2016 at 2:51 AM, Igor Stasenko <siguctua at gmail.com>
>>>>>>> wrote:
>>>>>>> >
>>>>>>> > Some more bashing today.. (don't take it personal, i may be wrong)
>>>>>>> >
>>>>>>> > BlPath hierarchy.. and BlShape.
>>>>>>> >
>>>>>>> > Why you redefining what is shape and what is path?
>>>>>>> > Of course, you are free to do it in Bloc..
>>>>>>> > But in terms of Athens, all of BlPath are actually - shapes..
>>>>>>> > And BlShape is some kind of encapsulation of shape, paints and
>>>>>>> transform.
>>>>>>> > It is a dumb state holder without any extra logic.
>>>>>>> >
>>>>>>> > My rule of thumb: do not produce dumb state holders. They has to
>>>>>>> be smart,
>>>>>>> > else it makes no sense in creating separate entity and designating
>>>>>>> it as
>>>>>>> > something else than any other bunch of data thrown into single
>>>>>>> clump,
>>>>>>> > sitting there deaf, blind, dead and silent until someone else will
>>>>>>> grab it
>>>>>>> > somewhere
>>>>>>> > and start using it for own purpose.
>>>>>>> >
>>>>>>> > Sure, i could understand, why you potentially may want such
>>>>>>> object(s)
>>>>>>> > around,
>>>>>>> > but it is not shape anymore and i wouldn't call it like that.
>>>>>>> Because shape
>>>>>>> > are shape, and has nothing to do with paints and transform,
>>>>>>> > it don't knows and don't cares whether it will be filled or
>>>>>>> stroked or both,
>>>>>>> >  and how many times, and if there will be single paint or thousand.
>>>>>>> > Such kind of properties is simply orthogonal to what shape
>>>>>>> existing for,
>>>>>>> > because it exists only to define geometry.
>>>>>>> >
>>>>>>> > I think all of that came from not understanding the roles of
>>>>>>> objects and how
>>>>>>> > they interact in Athens.
>>>>>>> Can you point us to documentation that describes Athen's architecture
>>>>>>> for these interactions?
>>>>>>> (sorry I haven't checked class comments, but I'm looking to start
>>>>>>> with
>>>>>>> something at higher level anyway)
>>>>>> No, i can't point it out. And you are right , this is nobody else's
>>>>>> fault than my own. I feel ashamed. Sure how i could demand that people
>>>>>> understand the concepts, if i didn't explained then anywhere (or if i did,
>>>>>> it is not in easily reachable place).
>>>>>> So, lets fix that. I will write it down here, and you can pick it up
>>>>>> and find suitable place for it.
>>>>>> ----------
>>>>>> Basic abstractions behind Athens.
>>>>>> Since Athens is about drawing graphics, we need a media where all
>>>>>> drawing operations will appear. We call that media a surface.
>>>>>> The surface is abstract. It can have set dimensions, or don't.  We
>>>>>> don't define if it representing some kind of physical surface (like part of
>>>>>> the display screen), or how it storing the data inside. We leaving an
>>>>>> introduction of such details to concrete surface implementation.
>>>>>> All that matters is that surface is a final target of all our drawing
>>>>>> operations.
>>>>>> Therefore, in Athens, a surface is usually a starting point where all
>>>>>> begins from, and you doing so by creating a specific surface.
>>>>>> It is surface's responsibility then, to provide user a means how he
>>>>>> can draw on it, and therefore there is a number of factory methods, that
>>>>>> allowing you to create a canvas, paints and shapes. All those three are
>>>>>> specific implementation of AthensCanvas, AthensPaint and AthensShape
>>>>>> protocols, suitable to be used with specific surface implementation that
>>>>>> you using.
>>>>>> Canvas.
>>>>>> Canvas represents a basic drawing context. We don't allow a direct
>>>>>> operations with surface, but instead we provide a context, that contains
>>>>>> and carries all information that represents a current stage of drawing
>>>>>> operations.
>>>>>> This includes things like, current coordinate transformation(s),
>>>>>> currently selected paint and shape, and paint mode.
>>>>>> In order to obtain canvas, one must use #drawDuring: message sent to
>>>>>> surface with block as argument. The given block receives an instance of
>>>>>> AthensCanvas as a single parameter. We intentionally enclosing all possible
>>>>>> drawing operations within a block to make sure that when we leave, we can
>>>>>> safely release all resources that was allocated, required to hold the
>>>>>> drawing context state. By exposing it in such form, we also making sure
>>>>>> that nothing can alter the surface outside a given block. That way, it
>>>>>> gives users a definitive answer, whether he finished drawing operations or
>>>>>> not, and if it safe to operate with surface for things like saving it to
>>>>>> file, or using it as a source for more complex operations, like acting as a
>>>>>> paint to fill area(s) inside another surface etc.
>>>>>> Paints and shapes.
>>>>>> A starting point is answering a question, how we can represent a
>>>>>> simplest, elementary drawing operation on a surface without putting too
>>>>>> much constraints.
>>>>>> We doing so by postulating that any elementary drawing operation can
>>>>>> be expressed by a function:
>>>>>> fill(paint, shape)
>>>>>> Please, note that 'fill' here is not a literally fill given shape
>>>>>> with given paint. We call it 'fill' for simplicity reason. It can anything
>>>>>> that altering the surface, but always taking into account given parameters:
>>>>>> paint and shape.
>>>>>> Then, from that perspective we can clearly say what are the roles and
>>>>>> responsibility of shapes and paints.
>>>>>> The shape defines the affected region, its geometry and location,
>>>>>> while paint defines how that region will be altered.
>>>>>> In this way, most of more complex operations can be expressed as a
>>>>>> series of such function invocations by using various paints and shapes.
>>>>>> Such representation also gives us a minimal set of roles, a building
>>>>>> bricks, that we need to introduce in order to represent any kind of drawing
>>>>>> operation we may need, as well as a minimal functionality in order to
>>>>>> implement such function(s). And therefore a minimal protocol(s), that all
>>>>>> paints and shapes should implement.
>>>>>> Since there potentially infinite number of various paint kinds and
>>>>>> shape kinds, we cannot make a single function that will implement all
>>>>>> possible permutations in order to fill shape with concrete paint.
>>>>>> To solve that we introducing a straight dispatch mechanism, where we
>>>>>> delegate the responsibility of implementing a concrete case, first to
>>>>>> shape, and then to paint.
>>>>>> The API representing this function in canvas by #draw protocol.
>>>>>> It takes currently selected paint and currently selected shape and
>>>>>> starting dispatch:
>>>>>> draw
>>>>>> "Fill the currently selected shape with currently selected paint"
>>>>>> ^ shape paintFillsUsing: paint on: self
>>>>>> So, first it goes to the shape, by sending #paintFillsUsing:on: ,
>>>>>> then shape dispatching it further to paint by sending appropriate
>>>>>> message
>>>>>> (be it #athensFillPath:on: or #athensFillRectangle:on: or anything
>>>>>> else, if you want to introduce new kind of shape representation and
>>>>>> implement it accordingly).
>>>>>> Such dispatch gives us an ability to easily extend the framework by
>>>>>> introducing new kind of shapes and paints , by implementing new kind of
>>>>>> fill() functions for them.
>>>>>> -----------
>>>>>> I hope that will make clear at least part of things what is there,
>>>>>> behind the scenes.
>>>>>>> cheers -ben
>>>>>> --
>>>>>> Best regards,
>>>>>> Igor Stasenko.
>>>>> --
>>>>> Best regards,
>>>>> Igor Stasenko.
>> --
>> Best regards,
>> Igor Stasenko.
> --
> Best regards,
> Igor Stasenko.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.pharo.org/pipermail/pharo-dev_lists.pharo.org/attachments/20160406/2d9644ee/attachment.html>

More information about the Pharo-dev mailing list