[fpc-devel] OS/2 and DLLs

[Tomas Hajny]

On 17 Dec 14, at 19:27, Dave Parsons wrote:


Hi Dave,

> I needed to create an OS/2 DLL today & was disappointed by the compiler (v2.6.4) telling me that they were not supported.
> 
> http://www.freepascal.org/faq.var#os2-dll section 6.3 seems to agree but is rather old.
> 
> Can anyone confirm that this is still true?

Yes, it's still the case.


> Google didn't find much and it is not mentioned on  the Future Plans page.
> What is the reason?
>    Too difficult, no demand/request, no time or whatever.

The primary reason is the fact that creation of DLLs is not supported 
by the ancient EMX port of the GNU linker ld.exe. As far as I know, 
OS/2 ports of GCC (both the original EMX port and the recent ones) 
solve this issue by using the IBM linker (possibly masked by EMXOMFLD 
binary). However, this has several caveats (no special order):

1) Object files need to be translated/transformed from the a.out 
format supported by the GNU binutils (as.exe and ld.exe) to Intel OMF 
expected by the IBM linker (e.g. using the tool EMXOMF supplied with 
EMX), or they need to be assembled using a different assembler 
supporting this format (such as Watcom assembler or NASM) from the 
very beginning, or support for direct creation of object files in 
this format needs to be added to FPC directly. When I tried 
translating object files created by FPC (using as.exe) with EMXOMF 
some time ago, it failed to process some of them. I didn't manage to 
find out the reasons so that I could possibly fix them at that point 
in time (no we get to amount of time available mentioned by you).

2) Binaries linked with IBM LINK.EXE cannot be debugged using 
GDB/PMGDB any longer (and no other debugger either, in fact, because 
LINK.EXE would not support the stabs debug format which is currently 
the only alternative supported for the OS/2 target in FPC).

3) There are certain special features which would need to be solved 
differently if using a different linker (e.g. support for direct port 
access). These wouldn't prevent most common uses, but they would 
require spending some time on their resolution too.

Now back to your questions:

1) Is it possible to solve it? Yes, certainly.

2) Does the resolution require considerable amount of time? Yes, it 
probably does (especially if the only person working on it is me).

3) Has there been strong demand? Well, not really (although some 
questions already arose in the past).

I don't have much time for working on this task, but I'm willing to 
spend some time on it as soon as I finish my current work on finally 
properly supporting all UnicodeString Manager functions for OS/2 
(that requires some time, but it is quite simple from technical point 
of view). My motivation to work on the linker issue would be higher 
if someone else joins me in this effort. Feel free to contact me 
directly if you're interested.

Tomas