Hi Alex, On Thu, Jan 2, 2014 at 3:05 PM, Alexandre Bergel <alexandre.bergel@me.com>wrote:
I think that one example Stef has in mind, is the following:
GLMCompositePresentation new with: [ :a | a list display: [ :input | 1 to: input ]; act: [ :list | self halt ] entitled: 'Plip' ]; openOn: 42
what is âaâ ? How one should know that âaâ answer to #list ? And that the result understand #display: and #act:
Removing block, could be something like:
GLMCompositePresentation new add: (ListPresenter new display: [ :input | 1 to: input ] ⦠)
"with:" is a pattern that you can implement on Object actually so that you can continue cascading an object without using a separate local variable. I prefer this construct specifically when prototyping. "a" is just a variable name :). Especially when used in conjunction with "with:", you can read it as a sentence: "composite presentation with a list ...". The same goes when reading in the context of a transmission:. browser transmit from: #x; to: #y; andShow: [:a | a list ... ] ==> "browser transmit That is all. Of course, you can also do it the way you say it. And it can be even more explicit: composite := GLMCompositePresentation new. list := GLMListPresentation new. composite add: list. Is this better? Perhaps, if you see this code only once. But, let's look at it from a different perspective. If you use Glamour, you will likely end up with many browsers. So, you want to have the code fluent, and you might also want to recognize the code when you see it. It's much like we did for Roassal/Mondrian. In Moose, the "view" variable is pretty much reserved for visualization builders, and when we see methods called *On: we know they refer to visualizations. This is a pattern that emerged. I see the Glamour API in the same way. Cheers, Doru
Alexandre
On Jan 2, 2014, at 10:40 AM, Tudor Girba <tudor@tudorgirba.com> wrote:
Hi Stef,
Glamour does not overuse blocks any more than any caller to cull: does it. And if you look closely now, you will see that the amount of arguments have decreased significantly (you do not have to repeat the objects that are passed in by the transmission for every block). But, blocks are not essential in Glamour (and especially, if it really is a problem, we can enforce the arguments). I just showed how you can simply not use blocks at all for specifying widgets. So, my argument is that this "blocks worry" isn't truly justified. And I also showed how Spec takes a similar approach for scripting, but you do not see Spec as having the same problem.
I am a bit troubled by your remark that the component model is not plain Smalltalk. The component model has nothing to do with the blocks and it is formed by plain objects: http://www.themoosebook.org/book/internals/glamour/architecture
Could you explain?
(please note that I am not picking up a fight - I would just want to understand how people see these things)
Doru
On Thu, Jan 2, 2014 at 2:13 PM, Stéphane Ducasse < stephane.ducasse@inria.fr> wrote: Doru
You know well that Glamour has a component model that it not plain smalltalk model. So why are you telling me I do not know what? Now you also know that Glamour overuse blocks with multiple optional block arguments where the order is important. So I do not undersand why you are saying that.
Stef
Interesting comparison. It sounds to me as if this "component model" is
a bad thing and that VW is a gold standard when it comes to describing user interfaces. Also, when you say DSL you refer to the way the Smalltalk API was design to read like sentences, which might sound like it goes against the tradition of Smalltalk. As for blocks, some would go as far as considering them to be plain Smalltalk :).
I think a more detailed comparison would be more useful for newcomers.
So, here is one.
First, the block issue. The funny thing is that if you want to build a
dynamic browser with Spec you have to rely on blocks, too. Take a look at an example from Ben:
--- m := DynamicComposableModel new. m instantiateModels: #( button ButtonModel list ListModel). m list items: #(1 2 3 4 5). m button label: 'Plip'; action: [ self halt ]. m openWithSpecLayout: (SpecLayout composed newColumn: [: c | c add: #list ; add: #button height: 26 ]; yourself). ---
Glamour is actually written in plain Smalltalk, and you can simply use
subclassing, too.
For example, something like this: GLMCompositePresentation new with: [ :a | a list display: [ :input | 1 to: input ]; act: [ :list | self halt ] entitled: 'Plip' ]; openOn: 42
Could be done like this:
"Define a list presentation with an action in distinct classes to make
blocks disappear"
GLMGenericAction subclass: #MyPlipAction instanceVariableNames: '' classVariableNames: '' poolDictionaries: '' category: 'AAA'. GLMListPresentation subclass: #MyListPresentation instanceVariableNames: '' classVariableNames: '' poolDictionaries: '' category: 'AAA'. MyListPresentation compile: 'displayValue ^ 1 to: self entity'. MyListPresentation compile: 'actions ^ {MyPlipAction new}'. MyPlipAction compile: 'actOn: aPresentation self halt'. MyPlipAction compile: 'title ^ ''Plip'''.
"Use the list" composite := GLMCompositePresentation new. composite add: (MyListPresentation new). composite openOn: 42.
I doubt the blocks are an actual issue. Using blocks has the advantage of more compact code, but it is true that it comes with a price: when you store the block you cannot edit its code live while debugging (you actually can do that, too if you take care).
The main difference between Spec and Glamour is actually conceptual: the application space. Spec addresses widgets: with Spec you place widgets on a "canvas" and handle the interaction. So, in a sense, this places Spec at the same level as Morphic. Glamour is at a higher level and it is specifically dedicated to supporting browsing workflows. In its current state, it is not useful for laying out widgets, but it can be very concise for expressing browsers.
The way this works is via this "component model". Essentially, a browser can be described as a set of components that can pass data to each other. Glamour lets you model that.
It so happens that the inspector and the debugger are browsers and can be expressed at that level. It also so happens that most user interfaces that programmers use can be expressed as browsers, too. And Glamour was actually validated for hundreds of various browsers (literally), so it seems it can support a large set of use cases.
If you want to learn more about Glamour you can read the chapters from: - http://deepintopharo.org, or - http://www.themoosebook.org/book/internals/glamour
If you want to see where Glamour fits within Moose, you can read this post: http://www.humane-assessment.com/blog/moose-custom-analyses-made-easy
Doru
-- www.tudorgirba.com
"Every thing has its own flow"
-- www.tudorgirba.com
"Every thing has its own flow"
-- _,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;: Alexandre Bergel http://www.bergel.eu ^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;._,.;:~^~:;.
-- www.tudorgirba.com "Every thing has its own flow"