| wxWindows 2 for Windows FAQ |
See also top-level FAQ page.
wxWindows 2 is designed to make use of WIN32 features and controls. However, unlike Microsoft, we have not forgotten users of 16-bit Windows. Most features work under Windows 3.1, including wxTreeCtrl and wxListCtrl using the generic implementation. However, don't expect very Windows-95-specific classes to work, such as wxTaskBarIcon. The wxRegConfig class doesn't work either because the Windows 3.1 registry is very simplistic. Check out the 16-bit makefiles to see what other files have been left out.
16-bit compilation is supported under Visual C++ 1.5, and Borland BC++ 4 to 5.
wxWindows 2 for Windows will also compile on Unix with gcc using TWIN32 from Willows, although TWIN32 is still in a preliminary state. The resulting executables are Unix binaries that work with the TWIN32 Windows API emulator.
You can also compile wxWindows 2 for Windows on Unix with Cygwin or Mingw32, resulting in executables that will run on Windows. So in theory you could write your applications using wxGTK or wxMotif, then check/debug your wxWindows for Windows programs with TWIN32, and finally produce an ix86 Windows executable using Cygwin/Mingw32, without ever needing a copy of Microsoft Windows. See the Technical Note on the Web site detailing cross-compilation.
There is a linking problem with Symantec C++ which I hope someone can help solve.
Borland C++ is fine - and very fast - but it's hard (impossible?) to use the debugger without using project files, and the debugger is nowhere near up to VC++'s quality. The IDE isn't great.
C++Builder's power isn't really used with wxWindows since it needs integration with its own class library (VCL). For wxWindows, I've only used it with makefiles, in which case it's almost identical to BC++ 5.0 (the same makefiles can be used).
You can't beat Cygwin's price (free), and you can debug adequately using gdb. However, it's quite slow to compile since it does not use precompiled headers.
CodeWarrior is cross-platform - you can debug and generate Windows executables from a Mac, but not the other way around I think - but the IDE is, to my mind, a bit primitive.
Watcom C++ is a little slow and the debugger is not really up to today's standards.
However, the issues surrounding Unicode support have been looked into so we know what we need to do, and have some header files ready to use containing appropriate type definitions. Just about every file in wxWindows will need changes, due to the pervasive nature of characters and character arrays. Unicode support is needed for the port to Windows CE (see above), and will probably be added in time for version 2.1.
With a DLL approach, and with different versions and configurations of wxWindows needing to be catered for, the end user may end up with a host of large DLLs in his or her Windows system directory, negating the point of using DLLs. Of course, this is not a problem just associated with wxWindows!
Statically-linked wxWindows 2 programs are smaller than wxWindows 1.xx programs, because of the way wxWindows 2 has been designed to reduce dependencies between classes, and other techniques. The linker will not include code from the library that is not (directly or indirectly) referenced by your application. So for example, the 'minimal' sample is less than 300KB using VC++ 6.
If you want to distribute really small executables, you can use Petite by Ian Luck. This nifty utility compresses Windows executables by around 50%, so your 500KB executable will shrink to a mere 250KB. With this sort of size, there is reduced incentive to use DLLs.
VC++ 5's optimization code seems to be broken and can cause problems: this can be seen when deleting an object Dialog Editor, in Release mode with optimizations on. If in doubt, switch off optimisations, although this will result in much larger executables. It seems possible that the library can be created with strong optimization, so long as the application is not strongly optimized. For example, in wxWindows project, set to 'Minimum Size'. In Dialog Editor project, set to 'Customize: Favor Small Code' (and no others). This will then work.
Here are Vadim's notes:
To use these new makefiles, you don't need anything (but see below). However, you should NOT modify them because these files will be rewritten when I regenerate them using tmake the next time. So, if you find a problem with any of these makefiles (say, makefile.b32) you'll need to modify the corresponding template (b32.t in this example) and regenerate the makefile using tmake.tmake can be found at www.troll.no/freebies/tmake.html. It's a Perl5 program and so it needs Perl (doh). There is a binary for Windows (available from the same page), but I haven't used it, so I don't know if it works as flawlessly as "perl tmake" does (note for people knowing Perl: don't try to run tmake with -w, it won't do you any good). Using it extremely simple: to regenerate makefile.b32 just go to distrib/msw/tmake and type
tmake -t b32 wxwin.pro -o ../../src/msw/makefile.b32The makefiles are untested - I don't have any of Borland, Watcom or Symantec and I don't have enough diskspace to recompile even with VC6 using makefiles. The new makefiles are as close as possible to the old ones, but not closer: in fact, there has been many strange things (should I say bugs?) in some of makefiles, some files were not compiled without any reason etc. Please test them and notify me about any problems. Better yet, modify the template files to generate the correct makefiles and check them in.
The templates are described in tmake ref manual (1-2 pages of text) and are quite simple. They do contain some Perl code, but my Perl is primitive (very C like) so it should be possible for anybody to make trivial modifications to it (I hope that only trivial modifications will be needed). I've tagged the ol makefiles as MAKEFILES_WITHOUT_TMAKE in the cvs, so you can always retrieve them and compare the new ones, this will make it easier to solve the problems you might have.
Another important file is filelist.txt: it contains the list of all files to be compiled. Some of them are only compiled in 16/32 bit mode. Some other are only compiled with some compilers (others can't compile them) - all this info is contained in this file.
So now adding a new file to wxWindows is as easy as modifying filelist.txt (and Makefile.ams for Unix ports) and regenerating the makefiles - no need to modify all files manually any more.
Finally, there is also a file vc6.t which I use myself: this one generates a project file for VC++ 6.0 (I didn't create vc5.t because I don't need it and can't test it, but it should be trivial to create one from vc6.t - probably the only things to change would be the version number in the very beginning and the /Z option - VC5 doesn't support edit-and=continue). This is not an officially supported way of building wxWindows (that is, nobody guarantees that it will work), but it has been very useful to me and I hope it will be also for others. To generate wxWindows.dsp run
tmake -t vc6 wxwin.pro -o ../../wxWindows.dspThen just include this project in any workspace or open it from VC IDE and it will create a new workspace for you.
If all goes well, I'm planning to create a template file for Makefile.ams under src/gtk and src/motif and also replace all makefiles in the samples subdirectories with the project files from which all the others will be generated. At least it will divide the number of files in samples directory by 10 (and the number of files to be maintained too).
On the VC++ level, it's just the matter of calling _CrtSetDbgFlag() in the very beginning of the program. In wxWindows, this is done automatically when compiling with VC++ in debug mode unless wxUSE_GLOBAL_MEMORY_OPERATORS or __NO_VC_CRTDBG__ are defined - this check is done in wx/msw/msvcrt.h which is included from app.cpp which then calls wxCrtSetDbgFlag() without any #ifdefs. This works quite well: at the end of the program, all leaked blocks with their malloc count are shown. This number (malloc count) can be used to determine where exactly the object was allocated: for this it's enough to set the variable _crtBreakAlloc (look in VC98\crt\srs\dbgheap.c line 326) to this number and a breakpoint will be triggered when the block with this number is allocated. For simple situations it works like a charm. For something more complicated like reading uninitialized memory a specialized tool is probably better... Regards, VZ