Imo, the problem with the 4-pane browsers are:
1 - The left-most 'package' pane is a flat list. And there are 10K + classes in e.g. Pharo. There are 900+ (!) packages in Pharo, in the pane, so its not easy to scroll through. This is solved 'manually' by have a hierarchical naming of packages. I think this pane should just have tree view.
2 - When developing an app, you may be working on classes in multiple packages. You want to have 'project' view with a simple way of switching between them, without having 10+ system browser windows open, like shown in the article. Dolphin Smalltalk has implemented this concept.
Shameless plug: SmallJS (https://small-js.org) has these too..
Rather than a fixed 4-pane browser, a potential solution is to use something like the macOS finder's "Column" view, where each layer of the hierarchy produces a new pane, as many as you need to drill down to the particular thing you're looking for.
Whenever I tried dabbling in Smalltalk, I would feel constrained by the system in terms of raw real estate. I never cared to view the system through the narrow slit of the method code editor. I would suffer the explosion of windows. All of it was off putting to me, and I never pushed through it to work with a project long enough in order for it to be comfortable.
But, let's contrast it to Java, which pretty much is in the same boat as a modern Smalltalk. A very large core system of packages, classes, and methods. Wrapped in a spider web of dependencies. And, granted, I've been working with Java a long time, and this is all quite comfortable to me. But at the same time, it doesn't suffer the issues that the Smalltalk browser presents.
Consider, regarding the parent point, the entirety of the JDK is not typically presented in the modern IDE. It's just back there. There's no real estate committed to it. There's your project (typically), with its classes.
I use NetBeans, so I have a project view of packages and classes, I have code tabs of individual classes, and a contextual method navigator within a class.
Its routine to potentially have a tab explosion, in contrast to a window explosion when exploring code. But when I do that, all I get is the code window, not the entire project or context of the class.
That is, in my code, I can click on, say, an ArrayList method, and jump to the JDK source code for the method within the ArrayList. I don't get its entire package hierarchy, I get its method list (uncatagorized, but alphabetical) while viewing the code file. So, the impact of viewing a new method is minimal -- it's just a new tab.
But as a benefit, I get to see not only the method I'm interested in, but I get to see the context of the code around the method. When I code, I don't write methods in a vacuum. I organize my code somewhat around aspect of the logic of the class. So, if there's a method that does one thing, chances are that the methods around it are doing similar or related things, or part of the logic of whats going on. Proximity in the source file tells part of the story not captured at the method level.
Sure, I can click around, in the navigator, in the IDE directly to visit other methods, but I can also scroll about, have the breadth of the class as a canvas and gather the ebb and flow of the logic within it. This is stark contrast to the ST "tiny" method browser.
I can also search globally within the source methods rather than just method signatures, while still having access to call trees and references and many of the other wonders of the ST browser.
Totally agree the only viewing methods in isolation,
is less valuable than viewing the methods around it "for free".
Indeed the methods in my classes are caryfully ordered,
to keep ones that are used together on the source together.
(And SmallJS is file based)
> But, let's contrast it to Java, which pretty much is in the same boat as a modern Smalltalk. A very large core system of packages, classes, and methods.
Java is used in many places. Smalltalk is not. The whole comparison falls short due to this already. TIOBE ranks Java #4 right now, which is not too bad, even though TIOBE also sucks.
Extending that, an outliner can be a great way to deal with this stuff. The problem becomes the self-referential nature of OO systems. Outliners are good when you get to the method of interest, but then you have the problem of browsing senders and implementers.
Still, there's nothing worse than a Smalltalk without a class category list. Smalltalk/MT, for example, just had a class hierarchy browser, and the category class member variable was left blank. I found it very hard to know where to start looking for behavior.
While I agree that the package column has room for improvement, some of the 'largesse' can be handled by the search field at the bottom of it, and once you get to doing something serious you could set up your own image and ditch a lot or most of the standard distribution.
I think I'd prefer a pane view over a tree view because the higher levels retain their scroll position even as I browse beneath. Is there a similar UX possible for tree views?
