I think that this is particularly true (having a themer per widget) when the themer is more a skinner (it changes behavior of the widgets), else we could have addressed the problem with a solution close to setting. Two years ago I played with theming and widgets hierarchy and I came to the same conclusion (that your blog entries).

Hi Esteban,

Sorry for the delayed response.

There are two parts in your mail:

1. the decomposition of theme to have one themer per widget

2. a centralized catalog of values

I certainly agree with 2. In fact, when I refactored the theme colors a long time ago, I introduced a rudimentary reuse mechanism on the class side of the Theme class. So, now, we should extract those separately.

About 1., I have a different opinion. The hierarchy you mention already exists in the Theme class as well, only it is hidden, and it is very hard to see what each methods impacts.

Here is how I reached this conclusion. Take a look at this picture I produced before we started with Brick (the one from Morphic). The gray circles are methods in UITheme, and the red circles are typical Morphs (like TextMorph, ListMorph). The gray lines show self calls inside the UITheme, and the red lines shod calls from the morph to the UITheme methods. What you see here is a very strong clustering of the methods around the widgets, and this indicates that having a theme object per widget is appropriate.

A further constrain is that we want to be able to change the appearance of a widget on an instance basis (that is, overriding the default theme setting), and having a per-widget themer object, we can achieve that as well.

A similar design appears in Bloc as well, and I think it is actually more scalable then the current UITheme, provided that you indeed have a way to either centralize the definitions of colors (and other properties), or inject those properties from the outside.

All in all, let��s go ahead and extract the palette :).

Cheers,

Doru

On Aug 1, 2016, at 11:13 AM, Esteban Lorenzano <estebanlm@gmail.com> wrote:

Hi,��

For one of my side-projects, I made a new theme for Pharo (still no name, I was planing to call it ��Dark Metal�� or��something like that. Is a variation on the Dark Theme but ��our�� dark theme is more brown and this one is more blue (see��attached)�� I wanted to publish it to push it but then I arrived to an unexpected problem: For Spotter and GTTools in��general, theming is not done following current theming approach. Instead, they made a full hierarchy of objects.��

IMO this is plain bad. I understand the attempt to decouple, but now that means if I want to create a new theme, I need��to create my theme object with colors I want and then also I need to create an undetermined number of classes (at least��one for each tool, but there is also a hierarchy of things there)�� anyway, this DOES NOT scale. Because each tool will��have to have a ��theme class�� for each existing theme��
How themes (skins, bah) work in all word is to have a color palette and then tools takes them (they can ��play�� a bit with��this palette, but need to always respect the palette).��

Then, I will commit a SLICE modifying the ��themer�� classes to take colors from the current theme (instead of have them��hardcoded).��
But of course, how theme work now is not good because they mix ��theme�� (how they display) and ��skin�� (color palette).��I will also extract the palette to where should have always been (some kind of a style sheet object)�� who also should��have been editable in settings so people can tweak their configuration.��

I didn��t wanted to touch this before, because this will supposedly change with brick, but honestly this will not be ready��for Pharo 6 and this is annoying (also, I want to publish my theme and I do not want to add overrides all around :P)

cheers,
Esteban

<Screen Shot 2016-08-01 at 11.00.15.png>

--
www.tudorgirba.com
www.feenk.com

"Innovation comes in the least expected form.��
That is, if it is expected, it already happened."