Cross-platform development - Delphi 2011: How to made a Windows-tied library cross-platform?

Question

As perhaps you know already, most probably the next version of Delphi will be cross-platform. Also, here are some polls on the matter.

While writing a cross-compiler isn't a thing which interests us very much now, porting a library which was/is Windows-tied to multiple platforms, certainly does.

You can think, for example at VCL (Delphi's standard library). While it was designed for Windows only, it has value in it, and, of course, there are huge codebases which depend on it.

The question is: Which would be the best approach to made an application / library cross-platform aware ensuring a smooth conversion / upgrade path (as much as possible of course)?

I stress it again, we are not interested which is the best way to do cross-platform development only (there were questions on this theme). We are interested also in yet another requirement: The old code base / installations management.

PS: Experiences and/or methodologies from similar situations with other languages (eg. C/C++) which are regarded as standard practices are welcomed.

Thanks in advance.

Answer 1

Experience so far has shown that the best way to get a Delphi app compatible with future versions is to stick to pure Delphi components, and use nothing third party. Such an app will probably suck, but that's how it seems to me. I use lots of third party components, and the apps are great and successful. But the chances of them moving to this future too are not certain, and that may cause problems with such changes, but I'd rather have a great app now and have the problem than have a poor app now and not need to worry about it.

Answer 2

Visual component developer's perspective:

Add levels of functionality to your code, so as to be able to add another platform without changing the "Core" of the component. The compiler hopefully will have a platform switch. (Preferable more than one, working in conjunction with each other. ex. Windows/ARM, Windows/386, OSX/Cacao/386, Linux/Gnome/386).

The Layout structure might look something like this.

• ComponentJ.pas
• Linux\ComponentJ.pas
• Linux\Gnome\ComponentJ.pas
• Linux\KDE\ComponentJ.pas
• OSX\ComponentJ.pas
• 386\ComponentJ.pas
• ARM\ComponentJ.pas

As an Application Developer:

I'll start by moving all WIN API calls in my code into a group of libraries in a Windows directory as to be able to IFDEF it at library level and translate it into another platform I'd like to support as soon as the compiler becomes available, but only as I come across them.) This will also add the possibility to add adapters easier for the new platforms. It in any case is good practice to remove possible dependencies into a central place.

Answer 3

IMHO you can't build a xplatform Delphi and ensure a smooth transition for current VCL applications. It won't work. VCL was (luckily, because it allowed for great applications) designed with Windows in mind, and trying to design a compatible library working on a different platform would just mean longer development cycles and lots of compromises. The outcome will be a library noone would wish to use. Look at what happened to VCL.NET: it was the wrong choice. And it was working on the same OS!

We know that targeting non-Windows platform with native applications needs a native GUI library. We don't care about creating a GUI from scratch, for our application it's the way to go, we don't need Windows GUIs with all their standards under a different OS using different standards - we need to be able to code a fully native GUI for the target OS. Other applications may survive a GUI porting, but in the long run you don't get a real xplatform tool - you get a tool that may compile for other platforms but brings one platform paradigms to others - and it will also be not welcome by "native" developers on other platforms. If you're a Linux or Mac developer, why should you learn how to work with a library that carries its Windows inheritance to your platform? You'd find it a pain in the ass. If Embarcadero wants to sell XDelphi outside actual developers base, it has to offer much more than a new CLX.

Answer 4

I will pull from some ancient experience in making a code base cross compilable between windows and dos (Delphi 1/Turbo Pascal 7). The rule of thumb was to separate code into multiple units. Try to code WITHOUT using windows, messages or any visual components. If you find you need to make a call to one of these, then place that call in another unit and write a proxy (abstract class that you descend from works well) to dispatch the calls through. When a cross compatible version is released, all that you should have to do is code the other side of the proxy for the new target.

If you're designing a form based system, then try to stick with as many of the standard components as possible. NEVER implement any "business" rules directly in an event, instead place them in another unit and call into the other unit to perform the logic.

Now, there will definitely be changes required to get your final project cross compatible, but by following these simple patterns you should be able to greatly reduce the amount of work it will take.

Answer 5

Well, aside the things said - thanks all - I do think that there we need some additional things:

• we need tooling to do the necessary conversions
• we need tooling to help us in programming against a (some form of) MVC pattern

Answer 6

Simply pick the latest 4.6 QT and add good integration betwen the Pascal and the QT library. They have done it before (in the Kylix times). The QT is such powerfull these days.

I believe that QT is even better then VCL and at least 10 times more frequently updated and fixed.

So the plan is simple:

1. make a cross-platform Pascal compiler
2. make a cross-platform RTL
3. put the QT on top

and you will have a first-class natively looking applications on all platforms.

Answer 7

Compromises should not be done too much to make VCL compatible with Linux and Mac. Windows is VCL's root. I'll prefer a new and very clean GUI framework, even though without any backward compatibility. Make VCL fatter and fatter isn't a good idea!

Answer 8

1. Make C/C++/Delphi compilers that targets OSX/Linux
2. Make C/C++ compiler that can be Boosted
3. Write new VCL-Presentation Foundation (VGScene/WPF alike) it should not be backward compatibile! Delphi IDE should be written with such VCL-PF
4. Component Library should stay as it is (but with improved Data-Binding)
5. Only provide VCL 64-bit for Win64

Is this a problem?

Answer 9

1. make a cross-platform Pascal compiler
2. make a cross-platform RTL
3. put the QT on top

Well, look at freepascal and lazarus

Answer 10

My opinion:

1. Make cross platform compiler (OS x/Linux/ embedded solutions?/ symbian?). Maybe add ability to compile/convert pascal code into portable c/c++ code to build then on embedded platforms.
2. RTL have to be separated into cross-platform layer and native layer (as for JCL).
3. Add new core components for cross-platform compatibility and native components for each supported platform (QT for ex)
4. Add translation utilities to create/convert between platform's components, for ex: to convert pure windows form into mac os x cocoa's form.
5. All windows hierarchy of components have to be only upgraded to support x64 with maximum backward compatibility. All cross-platform component have to be in parallel hierarchy.
6. Next version of cross platform solution can be refactored and can include migration/convertion utility. Due to minimum codebase of cross-platform solutions, hierarchy and classes for cross platform can be heavily changed from version to version to achieve best architecture.

sorry for my English - not a native language (Russian is)!

Answer 11

I don't get it. All .NET looks the same to me providing we don't use any third party. Delphi using standard control is already fully functional but your app would look like thousands of others.

I think Embar should go for PDA, IPhone, Andriod as Windows desktop already eat about 98% of the market.

Mac is expensive and Linux is no cost at all. No use to go for Mac and Linux. Not worth the investment.