Recently I've been working on a set of basic compatibility macros for fasmg that would allow to assemble sources written for fasm with its standard Windows headers, with as few changes as possible. I have now completed a version that assembles most of the Win32/Win64 examples from fasm package and I'm attaching it here.

Note that I do think that any such headers for fasmg (or fasm 2, if such ever arrives) could make better use of the macro capabilities of the new engine, therefore I think that developing some entirely new macros with new and improved syntax could be a good direction. But converting the standard fasm's includes is an important first step.

Not every behavior of fasm's macros is preserved - for example, as I explained in the other thread, "proc" macro I prepared for fasmg creates a separate namespace for every procedure - because this is how it should have always been working, and it was only because of the limitations of fasm's macro engine that its "proc" did not really work like that (though it at least kept the local data labels limited in scope). This means that local labels inside "proc" no longer need to start with dot.

The "struct" and "union" macros I used are the plain ones I created for fasmg, and they do not reproduce the complete behavior of fasm's "struct". This should change later - I still plan to write a variant more compatible with fasm's one for the inclusion in this package, though in general I prefer the new one, for its simplicity and flexibility. Using the new version required some changes of syntax in fasm's examples, because this variant of "union" needs to end with "endu" and the values to initialize fields in structure need to be labeled instead of providing them in original order.

When converting the resource macros I was really amazed how old and fossilized they are, with weird syntax choices that date back to the time when fasm had only very basic macro abilities, and implementations that have not been touched in ages. I converted them all, preserving the compatibility, but I think that their syntax really could use some renewal. The same can be said about import macros.

Well done !

I also hope fasmg can replace fasm in time. It's a pity that some of my written apps fail to compile.

I have updated the package with converted Win64 headers and examples.

And another update, this time with extended "struct" macro which allows to initialize fields' values sequentially, like fasm's "struct" did. Both ways of providing values for fields are now allowed.
In fact, the "struct" macro in this package should now support all the documented features of its fasm's counterpart, while still allowing the new (named) style of field initialization.

Maybe it is better to use ":=" in the equates to ensure that these values will not be changed by accident.

VEG wrote:

Maybe it is better to use ":=" in the equates to ensure that these values will not be changed by accident.

Good idea! The equates were simply copied from fasm's package and fasm 1 did not have such feature, but now as it is available it would be wise to make use of it.

Code:

section '.rsrc' resource data readable         directory \                 RT_GROUP_ICON, icon_group_list, \                 RT_ICON, icon_data_list         resource icon_group_list, \                 200, LANG_NEUTRAL, main_icon_group         resource icon_data_list, \                 100, LANG_NEUTRAL, main_icon_16_data, \                 101, LANG_NEUTRAL, main_icon_32_data         icon main_icon_group, \                 main_icon_16_data, '16.ico', \                 main_icon_32_data, '32.ico'

The result:

Code:

flat assembler  version g.hs4p0 examples\dialog\dialog.asm [120]:         icon main_icon_group, \         main_icon_16_data, '16.ico', \         main_icon_32_data, '32.ico' macro icon [9]:         data file icon_file:position,size Processed: data file '32.ico':position,size Error: definition in conflict with already defined symbol.

The icon macros tries to redefine the "data" label when there are more than one sources of icons are provided.

Also I propose to add an additional "align 4" in the versioninfo macro (market with +):

Code:

macro vdata name,value                 local vs_data,vs_size,value_data,value_size                 align 4                 vs_data dw vs_size,value_size/2                 du 1,name,0                 align 4                 value_data du value,0                 value_size = $ - value_data                 align 4 ; + this one                 vs_size = $ - vs_data         end macro

At least I've seen that other compilers do an additional alignment here and this is why the wLength and the wValueLength are different fields here. The same with the macros for the FASM1, it also doesn't have this align directive here.

VEG wrote:

The icon macros tries to redefine the "data" label when there are more than one sources of icons are provided.

I have fixed the bug there.

resource.h has this line: "started = 1". But this variable isn't used anywhere and it is even not local. It seems that it is a leftover from some other algorithm and can be removed safely.

I have added the Challenger Interpreter to the examples, to demonstrate a bit more complex one assembling correctly. And I think that it is oddly fitting into a general theme of fasmg packages.

I found and fixed a couple of bugs in macros when assembling it for the first time.

How can i change the macros to declare proc with no "declaration"?

Code:

include 'win32w.inc' format PE GUI 4.0 entry start   start:     stdcall someProc     retn        proc someProc         ret   endp

Code:

E:\fas>fasmg src.asm src.exe flat assembler  version g.hy7gy src.asm [8]:         stdcall someProc macro stdcall [48] macro call [1] macro parse_jump_operand [11] macro parse_oper and [10] macro parse_operand_value [31]:         ns.imm = +op Processed: @dest.imm = +someProc Error: symbol 'someProc' is undefined or out of scope.

There is no error when i append something like "uses ebx" .
It caused because no match in proc32.inc , right?

Code:

match name declaration, statement

This has been fixed in the mean time. Please make sure you have downloaded the latest version (I have started adding the "update date" next to the file).

Hi, Thomasz:
There is one bug in the fastcall macro. The fastcall macro can't accept immediater >= 0x8000000 parameters. Please see below code in the examples\win64\template.asm:

Code:

examples\win64\template.asm: CW_USEDEFAULT = 80000000h invoke  CreateWindowEx,0,_class,_title,WS_VISIBLE+WS_DLGFRAME+WS_SYSMENU,CW_USEDEFAULT,128,256,192,NULL,NULL,[wc.hInstance],NULL

C:\fasmg\examples\x86\examples\win64>fasmg template.asm
flat assembler version g.hykpg
template.asm [25]:
invoke CreateWindowEx,0,_class,_title,WS_VISIBLE+WS_DLGFRAME+WS_SYSMENU,CW_USEDEFAULT,128,256,192,NULL,NULL,[wc.hInstance],NULL
macro invoke [1] macro fastcall [116] macro mov [92] macro store_instruction [86] macro simm32 [15]
Custom error: immediate value out of signed range.

The reason is that the CW_USEDEFAULT = 80000000h, so it will beyond the macro x86.simm32 imm size.

Thanks.

Thanks, I'm updating the file.

Hi, Thomasz:
It's okay with the newest fasm_win.zip. Thank you a lot

And please see this file:

Code:

examples\win64\dll\writemsg.asm: proc DllEntryPoint hinstDLL,fdwReason,lpvReserved         mov     eax,TRUE         ret endp

I think that it should use "rax" to instead of "eax" as the returned value.

Code:

proc DllEntryPoint hinstDLL,fdwReason,lpvReserved         mov     rax,TRUE         ret endp

The reason is that it's one 64bit DLL, not 32bit DLL.

Are you agree it?
Thanks.

IIRC TRUE == 1, so using either EAX or RAX will still put 1 into RAX, because the top dword of RAX is cleared to zero whenever EAX is written.

Perhaps ideally the code should look like this:

Code:

if TRUE shr 32 = 0                 mov     eax,TRUE         else                 mov     rax,TRUE         end if

revolution wrote:

Perhaps ideally the code should look like this: Code:         if TRUE shr 32 = 0                 mov     eax,TRUE         else                 mov     rax,TRUE         end if

Another approach could be:

Code:

assert TRUE shr 32 = 0         mov   eax,TRUE

It might also be worth noting for anyone not currently aware, and wondering why there is the extra complication, that using EAX instead of RAX produces a smaller code footprint because there is no REX prefix needed.

One could also write a MOV macro that would do this additional optimization automatically, similarly to the TEST optimization that was once discussed. Since in case of fasmg instructions are just macros anyway, we could add a package of such optimizations as an optional variant that would not have much overhead compared to the basic ones.