Hand Coded GUI Versus Qt Designer GUI

Question

I'm spending these holidays learning to write Qt applications. I was reading about Qt Designer just a few hours ago, which made me wonder : what do people writing real world applications in Qt use to design their GUIs? In fact, how do people design GUIs in general?

I, for one, found that writing the code by hand was conceptually simpler than using Qt Designer, although for complex GUIs Designer might make sense. Large GUIs might be possible using Designer, but with time they might become very difficult to manage as complexity increases (this is just my opinion). I also downloaded the AmaroK source code to take a peek at what those guys were doing, and found many calls to addWidget() and friends, but none of those XML files created by Designer (aside: AmaroK has to be my favorite application ever on any platform).

What, then, is the "right" way to create a GUI? Designer or code? Let us, for this discussion, consider the following types of GUIs :

1. Simple dialogs that just need to take input, show some result and exit. Let's assume an application that takes a YouTube URL and downloads the video to the user's hard disk. The sort of applications a newbie is likely to start out with.
2. Intermediate level GUIs like, say, a sticky notes editor with a few toolbar/menu items. Let's take xPad for example (http://getxpad.com/). I'd say most applications falling in the category of "utilities".
3. Very complex GUIs, like AmaroK or OpenOffice. You know 'em when you see 'em because they make your eyes bleed.

Answer 1

In my experience with Qt Designer and other toolkits/UI-tools:

• UI tools speed up the work.
• UI tools make it easier to tweak the layout later.
• Raw code tends to result in more boilerplate that is used in every single project.

Complexity can often be dealt with in a UI tool by breaking the design into multiple UI files. Include small logical groups of components in each file and treat each group as a single widget that is used to build the complete UI. Qt Designer's concept of promoted widgets can help with this.

I'm not sure the scale of the project makes a big difference, but I would expect larger projects to benefit from UI tools. Maybe I'm wrong.

Ultimately, I think both raw code and UI tools can be effective. It probably depends a lot on the environment, the toolkit/UI-tool, and of course personal preference. I like UI tools because they get me up and running fast and allow easy changes later.

EDIT: Another nice feature of UI tools is that they make it easier/possible for non-programmers to work on the UI design.

Answer 2

Its strange that you're saying the writing code is simpler than manipulating objects in a graphical environment. It's a no-brainer.
The designer is there to make your life easier and in the long term it makes your code more maintainable. It's easier looking in the designer to see what the your UI looks like then reading the code and trying to imagine what it might look like.
With current Qt you can do almost everything from within the designer and the very few things you can't do, you can fix with very few lines of code in the constructor. Take for instance the simplest example - adding a signal-slot connection. Using the designer it's as simple as a double click. Without the designer you need to go lookup the correct signature of the signal, edit the .h file and then edit write your code in the .cpp file. The designer allows you to be above these details and focus on what really matters - the functionality of your application.

Answer 3

The organisation I work for has ported its GUI application to Qt several years ago. I think there are several aspects that are worth mentioning:

• Working with Qt Designer, at least at that point, was not a realistic option: there were too many features that couldn't be done with Qt Designer;
• Conventions and structure that had to be preserved prevented the use of Qt Designer;
• Once you've started without Designer, it is probably difficult to return to it;
• the most important aspect, though, was that the programmers were very much used to programming using vi or emacs, rather than using a GUI IDE.

My own experience, which goes back approx. 4 years, using Qt3.3, is that dynamic behavior in dialogs was not possible to realise in Designer.

Answer 4

I like to first turn to the designer to develop GUI widgets. As mentioned in the other posts, its faster. You also get immediate feedback to see if it "looks right" and isn't confusing to the user. The designer is a major reason I choose Qt over other toolkits. I mostly use the designer to make the one-off dialogs.

Having said that, I do the main window and any complex widgets by hand. I think this is the way Trolltech intended. QFormLayout is a class they provide to easily programatically create an input dialog.

By the way, the designer in Qt 4 is not an IDE like the one they had in Qt 3. It's just an editor for editing .ui files. I like it that way. The new cross platform IDE is going to be called Qt Creator.

Answer 5

Our experience with Designer started in Qt3.

Qt3

At that point, Designer was useful mainly to generate code that you would then compile into your application. We started using for that purpose but with all generated code, once you edit it, you can no longer go back and regenerate it without losing your edits. We ended up just taking the generated code and doing everything by hand henceforth.

Qt4

Qt4 has improved on Designer significantly. No longer does it only generate code, but you can dynamically load in your Designer files (in xml) and dynamically connect them to the running objects in your program -- no generated code however, you do have to name the items in Designer and stick with the names to not break your code.

My assessment is that it's nowhere near as useful as Interface Builder on Mac OS X, but at this point, I could see using the Designer files directly in a program.

We haven't moved back to Designer since Qt3, but still use it to prototype, and debug layouts.

For your problems:

1. You could probably get away with using the standard dialogs that Qt offers. QInputDialog or if you subclass QDialog, make sure to use QButtonDialogBox to make sure your buttons have the proper platform-layout.

2. You could probably do something more limited like xPad with limited Designer functionality.

3. I wouldn't think you could write something like OpenOffice solely with Designer but maybe that's not the point.

I'd use Designer as another tool, just like your text editor. Once you find the limitations, try a different tool for that new problem. I totally agree with Steve S that one advantage of Designer is that someone else who's not a programmer can do the layout.

Answer 6

Just to say I've written and maintained complex GUIs in Qt without using Qt Designer -- not because I don't like Qt Designer, but because I never got around to working that way.

It's partly a matter of style and where you're coming from: when I started on Qt, I'd had horrible experiences of Dreamweaver and Frontpage and other visual HTML tools,and far preferred writing code with HomeSite and resorting to Photoshop for tricky layout problems.

There's a danger with visual code IDEs that you try to keep within the visual tools, but end up having to tweak code as well -- in ways that aren't well understood.

Learning iPhone development, for example, I've found it frustrating to hit 'magic' visual stuff ('drag from the empty circle in the Connections inspector to the object in the Interface Builder window...') that would be simpler (for me) to understand in plain old code.

Good luck with Qt -- it's a great toolkit, however you use it, and Qt Creator looks like being a great IDE.

Answer 7

I'd add that one of the reasons for using graphical designer was the lack of layout managers in Win32, for instance. Only absolute positioning was possible, and doing that by hand would have just sucked.

Since I switched from Delphi to Java for GUI apps (back in 2002), I've never used designers any more. I like layout managers much more. And yeah, you get boilerplate code, but moving objects on a UI designer may take as much time as changing the boilerplate. Plus, I would be stuck with a slow IDE; that's for the Java/C# case, OK, while for Qt (especially Qt4) it doesn't apply. For Qt3, I wonder why one should edit the generated code - wasn't it possible to add code in other files? For which reason?

About the discussed cases: 1) Hand Coded GUI is likely faster to write, at least if you know your libraries. If you're a newbie and you don't know them, you may save time and learn less with a designer, since you don't need to learn the APIs you use. But "learn less" is the key factor, so in both cases I'd say Hand Coded GUI.

2) Menu bars are quite annoying to write code for. Also, think to details like accelerators and so on. Still, it depends on what you're used to. After some time, it may be faster to type that boilerplate than to point-and-click into designer to fix all those properties, but just if you can really type like into a typewriter (like those admins for which typing Unix commands is faster than using any GUI).

3) I'd extend the answer for case #2 to this one. Note that, for Win32 platforms, it may be possible that using designers which generate Win32 resources might be faster to load (no idea about that).

However, I'd like to mention a potential problem with using Qt Designer there. Real world case: it took some seconds (say 10) to load a complex Java dialog (the Preferences dialog box for a programmer's text editor) with a lot of options. The correct fix would have been to load each of the tabs only when the programmer wanted to see them (I realized that after), by adding a separate method to each preference set to build its GUI.

If you design all the tabs and the tab switcher together with a designer, can you do that as easily? I guess there might be a similar example where a hand coded GUI gives you more flexibility, and in such a big app, you're likely to need that, even if just for optimization purposes.

Answer 8

We're using the Qt Designer if anyone needs to create a Gui.
The thing is to create just little Widgets for certain tasks (like you would do in a class-design) and then get them together into a "parent-gui".

This way your widgets are highly reusable and could be used for Guis in a modular way. You just have to specify which signals each Widget is sending and which slots they provide.

We additionally are creating .ui-Files which than could be generated during build-process. Until now there was no need to edit those files by hand.