[squeak-dev] About a new "Math" package ... (was: Proposal: Geometry Classes)

Marcel Taeumel marcel.taeumel at hpi.de
Wed Mar 10 11:23:14 UTC 2021


Hi Dave.

Just a quick thought. :-)

> Is there any reason to restrict classes to a single category. After all its primary use I believe is as an aid to navigating the classes via inspectors.

Well, in this case, it matters big time whether one proposes something for the "Kernel" or just "MathExtras" package. One goal here is to level the path for such Trunk contributions. :-)

Best,
Marcel
Am 10.03.2021 12:19:27 schrieb David Duke <duke.j.david at gmail.com>:
Interesting discussion. Throws up issues that I and possibly others have had getting started with Smalltalk in general.
What are the criteria by which classes are arranged into categories? Leaving an element of subjective choice may be
helpful, but some guidelines might not go amiss.
Whilst the design choice of single inheritance for classes has been done to death. Is there any reason to restrict
classes to a single category. After all its primary use I believe is as an aid to navigating the classes via inspectors.
And there as been work on UIs for multi-dimensional data/hierarchies. Take 'Polygon' as an example
I might  expect to find it  under:
- graphics, where its primary uses are?
- 2D Primitives?
- geometry, as it supports geometric operations such as testing containment
- Collections, as its a collection of points, and one might want to enumerate/select.  etc

`specific cases is an orthogonal issue. Is it useful to view
rectangle/triangle etc a optimised forms of Polygon?

Now if you look elsewhere 
- Java allows classes to have implement  interfaces
- Haskell allows types to be instances of multiple type classes?

So why not allow classes to belong to multiple categories and/or categories to be connected
via different kinds of links along the lines of concepts in a semantic network...
 In ay event it would be helpful if there were some general published guidelines for
what membership of a category means any  criteria for putting classes into categories or not
and then for each category the defining characteristics of its classes?

Just some thoughts..

regards,
David.




On Wed, Mar 10, 2021 at 10:45 AM Marcel Taeumel <marcel.taeumel at hpi.de [mailto:marcel.taeumel at hpi.de]> wrote:

That's right. I recall that Patrick's initial proposal was about reducing redundant implementations, most of them rather high-level.

(One reason why "Number" would remain in "Kernel" is that "Character" inherits from it. And "DateAndTime" is a "Magnitude". Hmmm...)

Hmm... so, if "Integer" would be in a "Math", would it make sense to keep "SmallInteger" in a "Kernel" package? I wouldn't have a problem with "eternally coupling" both Kernel and Math. Hmm....

For simplicity, let me focus on "SmallInteger" and "Float32Array" as two low-level optimizations. A minimal core needs minimal concepts, which should definitely include "Math" but maybe not "MathExtras"? After all, what value would a minimal core have if it cannot be used for anything meaningful? Then again, "anything meaningful" is so subjective ... Sigh. Mayber numbers aren't the best way to think about re-classification since most of them are used through literals and simple textual representations. Users do not bother with their classes.

Hmmm... back to "Geometry" again, I guess. :-) Point, Rectangle, Line, Polygon, ...

Best,
Marcel
Am 10.03.2021 11:22:23 schrieb Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com [mailto:nicolas.cellier.aka.nice at gmail.com]>:
Hi Marcel,
I see the POV of classification for understanding and teaching. But isn't there also the perspective of re-constructing a system from minimal core/kernel ?
Also, is SmallInteger really related to Math? I do not see it as a specialization, but rather as providing some kernel operations (primitives) onto which we can build/generalize Integer math.

Le mer. 10 mars 2021 à 09:48, Marcel Taeumel <marcel.taeumel at hpi.de [mailto:marcel.taeumel at hpi.de]> a écrit :

Hi Nicolas,

in the long term, we might want to re-design the entire "Kernel" perspective to be specialized attachments to other packages. I would like to see a clear separation of non-programming, high-level concepts and technical, low-level optimizations.

For example, "Integer" could be discoverable through "Math-Quantity" while "SmallInteger" could reside in "MathExtras-Kernel" or similar. Eventually, we would turn around the perspective from primarily technical to primarily conceptual. Note that the inheritance tree would still look the same.

Another example would consider RawBitsArray. "Rectangle" could reside in "Math-Geometry" while "Float32Rectangle" could be a subclass of "Float32Array" and hence reside in "MathExtras-Collections" or similar, optimized for FFI.

A similar example can be constructed for "Quanternion" and "Float32Quaternion".

Best,
Marcel
Am 09.03.2021 18:49:55 schrieb Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com [mailto:nicolas.cellier.aka.nice at gmail.com]>:
Wow, high level of brainstorming, not sure I can catch up ;)

I'm not sure whether Number subclasses should be put in Math, they are
so essential to the Kernel...

What could obviously go in some extra Math-something is for example
all the function extensions (inverse trigonometry, hyperbolic, ...)
For trigonometry, not sure, it's essential to geometry.
Also the accelerated large integer arithmetic would find its place in
some extra package (not required in Kernel).
No problem if you want Quaternion in trunk, if it can be useful for 3D
geometry, then good.

For RawBitArray, I'm not sure, it's more specific to programming than
math per se (the fact that we use bounded integers of some
byte-size...). RawBitsArray really shine when interacting with the
outside world (importing large data sets from some standard format
and/or passing them to FFI).

Le mar. 9 mars 2021 à 14:28, Marcel Taeumel a écrit :
>
> > non-ultimate partition
>
> The system evolves. Code changes. New insights will influence onward refactorings. That's always the baseline. ;-)
>
> Best,
> Marcel
>
> Am 09.03.2021 14:27:06 schrieb Thiede, Christoph :
>
> From the perspective of "good old baby steps" and with the notion of "Extras" as a non-ultimate partition in mind, this sounds very reasonable ... :-)
>
>
> Best,
>
> Christoph
>
> ________________________________
> Von: Squeak-dev im Auftrag von Taeumel, Marcel
> Gesendet: Dienstag, 9. März 2021 14:22:47
> An: squeak-dev
> Betreff: Re: [squeak-dev] About a new "Math" package ... (was: Proposal: Geometry Classes)
>
> First step would be to assess whether something is "core or not". And "SomethingExtras" or "SomethingExtension" is a familiar pattern for this. Once a "MathExtras" package severely lacks cohesion, we can split it up again.
>
> Best,
> Marcel
>
> Am 09.03.2021 14:18:39 schrieb Thiede, Christoph :
>
> Yeah, I see this point, but still ... "Extras" sounds kind of arbitrary to me. If you cannot find a precise name for a package, how high can its coherency be? :-)
>
>
> Best,
> Christoph
> ________________________________
> Von: Squeak-dev im Auftrag von Taeumel, Marcel
> Gesendet: Dienstag, 9. März 2021 14:05:58
> An: squeak-dev
> Betreff: Re: [squeak-dev] About a new "Math" package ... (was: Proposal: Geometry Classes)
>
> MathExtras might host low-level optimizations, as I exemplified:
>
> > And I would like to add that "number crunching" part around fancy graphics (e.g. OpenGL through FFI) to "Math-Collections" or maybe "MathExtras-Collections", looking at all the subclasses of RawBitsArray.
>
> :-) I assume that it may be beneficial to at least decide whether something is "core" or "extra". I would like to have this for "Math" from the beginning.
>
> Best,
> Marcel
>
> Am 09.03.2021 13:10:39 schrieb Thiede, Christoph :
>
> Referring to the MathExtras proposal ... What is your general idea of an extra package? I know Morphic-Extras as a collection of non-necessary tools, helpers, and demos, but the general idea sounds kind of vague to me.
>
>
> CollectionsExtras would be a place for Text enhancements such as attributes (so actually, we could also call it simply "TextSupport" or something like this), MathExtras would contain "math stuff that is not necessary" ...
>
> Do you have any more precise idea of what classifies an extra package or would "MathSmorgasbord" be a franker name for the package? If yes, this would make me think about the coherency of such a package ... Just my 2 cents, of course. :)
>
>
> Best,
>
> Christoph
>
> ________________________________
> Von: Thiede, Christoph
> Gesendet: Dienstag, 9. März 2021 12:56:04
> An: squeak-dev
> Betreff: AW: [squeak-dev] About a new "Math" package ... (was: Proposal: Geometry Classes)
>
>
> Hi Marcel,
>
>
> sounds interesting! I have a few thoughts regarding package dependencies.
>
>
> As far as I understand, the dependency structure would look kind like this:
>
>
> Kernel-Objects -> Math-Quantity (see senders of SmallInteger for example), Math-Analysis (see senders of #asFloat for example)
>
> Math-Quantity -> Kernel (superclass), Math-Analysis (see senders of #asFloat)
>
> Math-Analysis -> Kernel (superclass), Math-Quantity (maybe?)
>
> Math-Geometry -> Kernel (of course), Math-Quantity (senders of SmallInteger), Math-Analysis (see senders of #asFloat)
>
> Math-Collections would probably depend on all other packages?
>
>
> The question is which of these dependencies can be eliminated and which are a problem at all.
>
>
> What about Random? Do you want to keep it in Kernel, depending on Math?
>
>
> What is about dependencies from Number (Kernel-Objects) to Float (Math-Analysis), for example, #asFloat, but also sophisticated things such as #sin? Do we want to create a bunch of extension methods for this?
>
> In any case, I think the "math functions" protocol on Collection should become an extension protocol ("*Math-Analysis" or "*Math-Analysis-enumerating").
>
>
> Best,
> Christoph
> ________________________________
> Von: Squeak-dev im Auftrag von Taeumel, Marcel
> Gesendet: Dienstag, 9. März 2021 10:50:09
> An: squeak-dev
> Betreff: [squeak-dev] About a new "Math" package ... (was: Proposal: Geometry Classes)
>
> Hi all!
>
> I think that we would rather need a "Math" package with "Math-Geometry" being one of multiple categories.
>
> Please take a look at the following proposed classification:
>
> Kernel-Objects (= not Kernel-Numbers)
> Magnitude
> Number
>
> Math-Quantity
> Integer (+ subclasses)
> Fraction
> ScaledDecimal
>
> Math-Analysis
> Complex
> Float (+ subclasses)
> Quaternion
>
> Math-Geometry
> Point
> Line
> Rectangle
> Polygon
> Path
>
> Math-Collections
> Vector2 (from 3DTransform, CroquetGL, etc.)
> Vector3
> Vector4
> Matrix2x3
> Matrix4x4
> VectorArray
> ...
>
> It would involve some effort, especially to untangle "ST80-Paths" from graphics :-) Eventually, I would like to see Nicolas' efforts for "Complex" and "Quaternion" in Trunk. And I would like to add that "number crunching" part around fancy graphics (e.g. OpenGL through FFI) to "Math-Collections" or maybe "MathExtras-Collections", looking at all the subclasses of RawBitsArray.
>
> May I add "Math" and "MathExtras" packages so that we can slowly get started? :-)
>
> Best,
> Marcel
>
> Am 22.01.2019 18:32:10 schrieb patrick.rein at hpi.uni-potsdam.de [mailto:patrick.rein at hpi.uni-potsdam.de] :
>
> Hi everyone,
>
> during some recent clean-up an issue with geometry objects came up. As a result the following idea came up which I thereby would like to put out for an initial discussion.
>
> In case we find that this might be useful, I would continue by implementing a first prototype of the package as a foundation of a more in-depth discussion later on.
>
> Bests
> Patrick
>
> # Squeak Change Proposal - Geometry Package
>
> ## Why?
> Squeak implements basic geometry logic in too many different places. For example intersection of different kinds of geometric objects is implemented all over Morphic, ST80, Balloon, Graphics, and Etoys. Such extensive scattering across packages and classes impedes modularity, that is, readability and extensibility. An example for this recently came up when we discovered an issue with testing for graphical intersections between PolygonMorph and RectangleMorph. It was not possible to compute that overlapping area because the class Rectangle omits to provide an important method. A quick fix would entail unnecessary dependencies (here: Morphic -> Balloon) or duplicated code (see also http://bugs.squeak.org/view.php?id=7872 [http://bugs.squeak.org/view.php?id=7872]). Consequently, we might want to modularize the geometry objects and operations. As a side effect, dependent packages such as Morphic can be simplified a little bit (more).
>
> ## Scope
> The proposed package should cover basic 2D geometric objects and their operations represented by the following classes:
> - Point
> - Line
> - LineSegment
> - Polygon
> - Rectangle (as an optimization as it could be represented as a special Polygon already)
>
> Most classes could simply be moved from their previous packages. Afterwards the interfaces would need to be made consistent with each other to allow interoperability of all geometry classes within the new package.
>
> ### Affected classes:
> - All classes in ST80-Paths
> - LineSegement, Bezier2Segment, Bezier3Segment (Balloon-Geometry)
> - Rectangle, Quadrangle, Point (Graphics-Primitives)
> - LineIntersections, LineIntersectionSegment, LineIntersectionEvent (Etoys-Squeakland-Graphics-Tools-Intersection)
>
> ## Open questions
> - Should this become a single new package or a subcategory in the Graphics package?
> - Should the package contain an Ellipses class?
> - Should we model curved line segments as BezierLineSegments, CurvedLineSegment, or Arc?
>
> ## Risks
> - This would potentially deprecate the existing ST80 geometry classes (ST80-Paths)
> - Some of the new classes will cause name clashes with existing classes. For example Line is currently in ST-80 and represents a line segment, and the class LineSegment is a line segment but not in the geometric sense as it also incorporates arcs. Both names might then be used by new classes with different meanings. This might be mitigated by introducing a pre-/postfix for the names of the new classes.
>
>





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20210310/b1cf3a7f/attachment.html>


More information about the Squeak-dev mailing list