Created attachment 2870 [details]
cpuinfo patch

Created attachment 2871 [details]
dynapi patch

Created attachment 2872 [details]
thread.h patch

Added, thanks!
https://hg.libsdl.org/SDL/rev/fed4345ff3f2
https://hg.libsdl.org/SDL/rev/a6da0a386f68
https://hg.libsdl.org/SDL/rev/97cd834a843e

(In reply to Ozkan Sezer from comment #0)
> 2. SDL_dynapi.h: disable dynapi for os/2.

This isn't super-urgent, but is there any reason this _can't_ work on OS/2, other than we still need to fill in some DosLoadModule/DosQueryProcAddr calls?

--ryan.

> This isn't super-urgent, but is there any reason this
> _can't_ work on OS/2, other than we still need to fill
> in some DosLoadModule/DosQueryProcAddr calls?

No, there is not.

>> This isn't super-urgent, but is there any reason this
>> _can't_ work on OS/2, other than we still need to fill
>> in some DosLoadModule/DosQueryProcAddr calls?
> 
> No, there is not.

Well I forgot that there is this: if I do the following in SDL_dynapi.c
and define SDL_DYNAMIC_API as 1, it builds yes, but nothing is exported
in the dll, so I gave up without a fight.  (The original 2.0.4 port by
Andrey possibly had the same issue.)

#elif defined(__OS2__)
#define INCL_DOS
#define INCL_DOSERRORS
#include <dos.h>
static SDL_INLINE void *get_sdlapi_entry(const char *fname, const char *sym)
{
    HMODULE hmodule;
    PFN retval = NULL;
    char error[256];
    if (NO_ERROR == DosLoadModule(&error, sizeof(error), fname, &hmodule)) {
        if (NO_ERROR == DosQueryProcAddr(handle, 0, sym, &retval)) {
            DosFreeModule(hmodule);
        }
    }
    return retval;
}

> Well I forgot that there is this: if I do the following in SDL_dynapi.c
> and define SDL_DYNAMIC_API as 1, it builds yes, but nothing is exported
> in the dll, so I gave up without a fight.

I've put this code in revision control: https://hg.libsdl.org/SDL/rev/6120f75935f9

Is BUILD_SDL defined when this code compiles? That's the magic in begin_code.h that tells OpenWatcom that DECLSPEC symbols should be exported, and should make SDL_DYNAPI_entry export in that file (at least, it _should_ do that...).

Also, while we're talking: one of our Linux buildbots has OpenWatcom available on it, so if all this OS/2 work gets far enough along, we can have it start doing OS/2 builds on each commit.  :)   I already do this with PhysicsFS, fwiw: https://physfs-buildbot.icculus.org/builders/physfs-os2

--ryan.

> I've put this code in revision control: https://hg.libsdl.org/SDL/rev/6120f75935f9

Well, it won't compile: I obviously pasted an old experiment
(it's a mess on my hard drive), here's a one that really builds:

diff -r 6120f75935f9 src/dynapi/SDL_dynapi.c
--- a/src/dynapi/SDL_dynapi.c
+++ b/src/dynapi/SDL_dynapi.c
@@ -24,6 +24,12 @@
 
 #if SDL_DYNAMIC_API
 
+#if defined(__OS2__)
+#define INCL_DOS
+#define INCL_DOSERRORS
+#include <dos.h>
+#endif
+
 #include "SDL.h"
 
 /* !!! FIXME: Shouldn't these be included in SDL.h? */
@@ -232,16 +238,13 @@
 }
 
 #elif defined(__OS2__)
-#define INCL_DOS
-#define INCL_DOSERRORS
-#include <dos.h>
 static SDL_INLINE void *get_sdlapi_entry(const char *fname, const char *sym)
 {
     HMODULE hmodule;
     PFN retval = NULL;
     char error[256];
     if (NO_ERROR == DosLoadModule(&error, sizeof(error), fname, &hmodule)) {
-        if (NO_ERROR == DosQueryProcAddr(handle, 0, sym, &retval)) {
+        if (NO_ERROR == DosQueryProcAddr(hmodule, 0, sym, &retval)) {
             DosFreeModule(hmodule);
         }
     }


> Is BUILD_SDL defined when this code compiles? That's the magic in begin_code.h
> that tells OpenWatcom that DECLSPEC symbols should be exported, and should
> make SDL_DYNAPI_entry export in that file (at least, it _should_ do that...).

It is indeed defined by the makefile. However, my first comment was not
completely correct, because :

1. SOME of the exports are missing, such as SDL_Log, e.g. testsuite fails
  like this:
Error! E2028: SDL_Log is an undefined reference
file testsprite2.obj(test/testsprite2.c): undefined symbol SDL_Log
file SDL_test_common.obj(src/test/SDL_test_common.c): undefined symbol SDL_Log

2. dynapi.c, is missing SDLCALL everywhere in its table defines, so
  it requires work if one really wants to build with dynapi enabled.


> Also, while we're talking: one of our Linux buildbots has OpenWatcom
> available on it, so if all this OS/2 work gets far enough along, we
> can have it start doing OS/2 builds on each commit.  :)

It would be cool. However, there are some fixes that needs to go in
before a build can succeed:

Bug #3740: otherwise testime.c won't build
Bug #3767: because watcom doesn't ship with scalbn(),
  except for the v2 fork at github

Optionals: bug #3769 and bug #3739

Besides, I don't know how you can build without the os/2 implementation
files themselves. Andrey still haven't responded to me.  (I can send to
your private email, though, if you want.)

> I already do this with PhysicsFS, fwiw:
> https://physfs-buildbot.icculus.org/builders/physfs-os2

OK.

OK, after several hours of pulling out my hair, found the culprit:
wlib is not case-sensitive by default, and it discarded SDL_Log()
in favor of SDL_log(). Using its -c switch when creating the import
lib fixes everything. Created bug #3771.

(In reply to Ozkan Sezer from comment #9)
> Well, it won't compile: I obviously pasted an old experiment
> (it's a mess on my hard drive), here's a one that really builds:

This patch is now https://hg.libsdl.org/SDL/rev/6e1d0d1e2a5b, thanks!

> Besides, I don't know how you can build without the os/2 implementation
> files themselves. Andrey still haven't responded to me.  (I can send to
> your private email, though, if you want.)

Yeah, I was just talking about the future.  :)

--ryan.