Hi Nicolai, Thanks a lot for the feedback. Please letâs continue. See more inline.
On Aug 25, 2016, at 6:34 PM, Nicolai Hess <nicolaihess@gmail.com> wrote:
2016-08-25 8:47 GMT+02:00 Tudor Girba <tudor@tudorgirba.com>: Hi,
Hi Doru, some questions and feedback ( I am sorry for my tone or if this sounds negative, it isn't meant to be)
Over the last coupe of years Stefan Reichhart and the rest of the GT team worked on an implementation of examples. The work is inspired from previous work done by Markus Gaelli and Adrian Kuhn.
As one of the goals of GT is to offer a live programming environment,
Isn't this what we already have, a live programming environment, I think for newcomers (and maybe others) this needs to make be more clear, how is different, what gap are you trying to fill.
one important issue is how to move from the static code to live objects as fast as possible.
Isn't code always "static" and aren't objects always "live", how do play gtExamples any role in this?
What I meant is that I want to be as little as possible in the static code browser. Instead I want to write code in the presence of a bounded âselfâ which happens either in a debugger or in an inspector. The gap that examples fill is that when I look at a static code and I have examples about it, I can possibly jump to those examples and code against their result. So, instead of coding a method about a FileReference in the code browser, I will code it against an instance of a FileReference. Hence, we make live programming easier to achieve.
That is why we worked on this library as a solution to provide a link between the static code and live objects.
From my understanding, I think this "link" between code an objects is the real valuable new point. This is the great thing about gtExamples. link from code (the example methods itself or methods refering a class to an example object/instance that can be inspected (the raw object or an object specific inspector pane). This is what I would consider the real step forward. You see a method refering to class TextModel and Nautilus or any other tool not only offers a method to browse all Users of this class, but too, a dedicated list of "example methods" where every example has a view for this example instance that let the user show and interact (even for non-visual objects through the "evaluater pane", or just see the code creating this example.
Exactly.
We really miss examples, I often see questions on the mailing list (especially about spect) that can be explained easily with an example. And even worse, often the examples already exists, they just aren't as visible.
Exactly. These examples are particularly amplified by the fact that we have an inspector that can provide a reacher experience through different views.
Furthermore, this examples library also enables the definition of assertions on the examples, and this provides the possibility of rethinking the way we construct tests throughout our system. Tests are great as they help us create live objects and then to assert live properties.
This whole thing sounds as if Unit-Test were a good idea but not the way that they are used today, I strongly disagree. I don't see this as a "rethinking the way we construct tests", yes, we can augment the current set of tests with addtional assertions on live objects, but this is not a replacement. "Tests are great as they help us create live objects" This is not my only purpose for writing tests, often unit-tests cover methods and "private-apis" not even considered to be used on live objects. You can not (or I don't want to ) write tests only on "finished lived objects" sometimes we need tests for initialiazation/private code or exception handling I don't see how we can offer this only by using example instances (yes your "rethinking" sounds like "this is the better way to do testsâ).
I think there is a misunderstanding here. When I test, (1) I create one or more objects, (2) I assert against them and then (3) I potentially cleanup. At least the objects from step 1 are potentially interesting for documentation purposes as well. However, because tests are built in a completely different environment than examples are, and because they are not casually linked to the code they are about, we cannot exploit them to the maximum potential. The GT-Examples model offers a unification. This means that you can use the same mechanism for expressing both a test scenario and a documentation one. There is potential to be exploited here. For example, there is research that aims to take the all sorts of objects and try to infer types for code out of these. We could make this much simpler. I understand that this is a departure from the classic way of testing, but we have already expressed more than 1 thousand examples both from a documentation and from testing point of view, and it does seem to work.
However, they do not allow us to leverage the objects we create, and this can be a tremendous resource for understanding systems.
In our vision, examples should be everywhere and they should be explicitly linked to the static code they exemplify. That is why the library comes with an initial integration in existing tools (such as Nautilus, Spotter, Inspector).
The current solution works well and it is the result of several rewrites. We think that the solution is complete in terms of features, but there are still several things to improve and iterate on. To this end, I kindly ask you to take a look at it while distinguishing between the concrete implementation choice (e.g., the current extensive use of pragmas) and the conceptual benefits of the approach.
To ease the discussion, we put together a short documentation:
Everytime I see a gtExample method on a class I first think, shouldn't this go to a Help or Doc or Exampels class instead. I don't know how others thinks about this but this is my first impression. For example, the example on your page:
FileSystem class >> #createFileOnDisk
<gtExample> < description: 'Create a new file or override an existing file with some contents. Open and close the stream safely'
^ FileSystem workingDirectory / 'test.txt'
writeStreamDo: [ :stream | stream nextPutAll: self comment ]; yourself
Nice, now the user can see how to use FileSystem to create a file, open *and* close the stream safely. But for me, this method does *not* belong to the FileSystem class it just not make any sense to me, to have a method (*in this class*) that opens and closes a stream. Even if this is just an example, I would put it in a doc-page or a tutorial that can execute code. But again , this is just my point of view. I can not really explain it, having a example method on Morph or a widget ui or a widget model class, that opens an example morph or widget in the world, is for me something completly different and a valid example. Having the same for the method above - is not, at least not as executable code on the FileSystm class).
I do not understand this last point. Just because an object does not have a visual appearance like a Morph does, does not make it uninteresting from an interaction point of view. The inspector already can provide the views. We also have the possibility of adding custom actions that can be installed as menu items. Even for a morph, I sometimes want to not look at its default appearance, but at its submorphs. Thus, I do not see the confusion. Nevertheless, the example does not have to be on the class side. It can be in any class you want and you can associate it with a subject. For example, all Roassal examples are in dedicated classes. There are hundreds of methods, so putting them all on the class side of a domain class would not work at all. We showed the example on the class side because that is a pattern that people used for a long time and it is a reasonable place when you have only a handful of examples. The rationale is that an example is a way to instantiate a class, so having it on the class side is not far fetched. Also, if you put it on the class side, you get by default the class as a subject for the examples it contains which is quite natural.
http://gtoolkit.org/doc/Examples/examples.html
That being said, you can get it with the full GToolkit in Pharo 6.0:
Is this based on the recent Pharo 6.0?
GTExamplesReleaseTests are failing for me
Yes, these are yellow.
Where did you test this? I get some Object>>#name deprecation warnings when browsing for examples refering a class, for example on class FileSystem and menu entry "Browse Examples refering FileSystem" (maybe a Pharo 5.0 version?)
I tested in Pharo 6.0 (60188), but we just got a problem that was reported related to Epicea and Martin is looking at it.
The example on the examples.html side isn't actually in the image right?
Yes, itâs not there yet.
The browsing examples of a package (context menu on nautilus package pane) does not work or I don't understand why it does not find any examples at all. The World menu "Browse All Examples" does not contain the class FileSystem, although FileSystem>>gtExampleZip is a gtExample, this is because the example method is in an extension package, should all gtExample methods be class extensions ? This is handled differently for different packages.
Hmm. When I "Browse All Examples" I get a Nautilus with FileSystem class>>gtExampleZip in my image. But, indeed, in the latest GToolkit image, this example is missing.
The code pane context menu of a sendersOf Message browser is broken (debug menu).
I do not understand what menu item you refer to. Could provide a screenshot.
From the web-side:
"Furthermore, Nautilus, the World-Menu, all Rub-Text-Editors as well as Spotter and Inspector provide access to retrieve, browse and navigate examples from entities within the world" I can not find it, not in inspector, Rub-Text-Editors, only in Nautilus.
In Nautilus and in RubText you get it in the GT-Examples menu (Browse examples with subject â¦). The Inspector is not yet there, but we are adding it.
And the menu entries are ... unfortunate (see screenshot), what you put in, the whole source as menu label?
Hmm. Something is strange there. I get the name of a method, not the source code. What image are you in?
run "run the example and return its return-value"
debug "same run, but open debugger if the example failsâ returnValue "the return-value" Executing run/debug/inspect returnValue does not seem to make any different when called from nautilus. It always gives an inspector on a dictionary holding the gtexample and its gtexampleresult (is this a bug?)
Yes.
Glossary: "Example: an example is a tiny stub object representing a GT-Example. It holds the references to its dependencies, subjects and many other entities. " What are "many other entities" this is a bit unclear
Icon, Label, Provider, and others that you can add through custom annotations if you want to. This part is not yet clear.
"After-method: the after-method is a method that is performed right after the example." After the example ? I thought an example is a "tiny stub *object*", how can it run? How it is run after I run an example for inspection, after I closed the inspector?
Not yet. At this point, inspecting does not prevent triggering of the cleanup method, but it would certainly be interesting to get there. Cheers, Doru
http://gtoolkit.org/#install (easiest is to download the ready made image for now)
For those that are at ESUG, I will try to provide a short overview during the Show Us Your Project session from today.
Cheers, Doru
-- www.tudorgirba.com www.feenk.com
"Reasonable is what we are accustomed with."
<gtexample_menu.png>
-- www.tudorgirba.com www.feenk.com "No matter how many recipes we know, we still value a chef."