Rewrite gendynapi into python

[1bsyl]

"gendynapi" tool seems more powerful in python and easier to extend.
Trying to add more possibility:
now there is an option "--dump" to dump all SDL api (parsed) into a json file.

(which can be useful for something like creating bindings #6337)

[1bsyl]

some status:

• it converts correctly all SDL function API.
• it converts correctly the callback, whereas gendynapi.pl is bugged on it.
• lots of comments
  -has a "--debug" flags to debug
• has a "--dump" to print a .json file like

some snapshot of the .json dump:

{
       "comment": "",
       "header": "SDL_stdinc.h",
       "name": "SDL_bsearch",
       "parameter": [
           "const void *REWRITE_NAME",
           "const void *REWRITE_NAME",
           "size_t REWRITE_NAME",
           "size_t REWRITE_NAME",
           "int (SDLCALL *REWRITE_NAME)(const void *, const void *)"
       ],
       "parameter_name": [
           "key",
           "base",
           "nmemb",
           "size",
           "compare"
       ],
       "retval": "void*"
   },

   {
       "comment": "/**\n * Compatibility function to initialize the SDL library.\n *\n * This function and SDL_Init() are interchangeable.\n *\n * \\param flags any of the flags used by SDL_Init(); see SDL_Init for details.\n * \\returns 0 on success or a negative error code on failure; call\n *          SDL_GetError() for more information.\n *\n * \\since This function is available since SDL 3.0.0.\n *\n * \\sa SDL_Init\n * \\sa SDL_Quit\n * \\sa SDL_QuitSubSystem\n */\n",
       "header": "SDL.h",
       "name": "SDL_InitSubSystem",
       "parameter": [
           "Uint32 REWRITE_NAME"
       ],
       "parameter_name": [
           "flags"
       ],
       "retval": "int"
   },

[slouken]

This seems like a good change to me, and will make it easier for people other than @icculus to extend it. ;)

[sezero]

Tested this by removing the late SDL_GetEventState addition to dynapi
headers and running the new python3 script to re-add it. Works but adds
a funny extra space, like this:

-SDL_DYNAPI_PROC(Uint8,SDL_GetEventState,(Uint32 a),(a),return)
+SDL_DYNAPI_PROC(Uint8,SDL_GetEventState,(Uint32  a),(a),return)

Nothing big, but thought that you'd want to know. This was with python-
3.6.6 built on an i686-linux.

[1bsyl]

Thanks, done, it makes it easier for comparing.

then, there is maybe some issue with:
extern DECLSPEC const SDL_AssertData * SDLCALL SDL_GetAssertionReport(void);

old:
SDL_DYNAPI_PROC(const SDL_assert_data*,SDL_GetAssertionReport,(void),(),return)
new:
SDL_DYNAPI_PROC(const SDL_AssertData*,SDL_GetAssertionReport,(void),(),return)

also with SDL_asprintf and "..."
SDL_DYNAPI_PROC(int,SDL_asprintf,(char **a, SDL_PRINTF_FORMAT_STRING const char *b, ...),(a,b),return)
vs
SDL_DYNAPI_PROC(int,SDL_asprintf,(char **a, SDL_PRINTF_FORMAT_STRING const char *b, ...),(a,b,c),return)
here, we should not generate the c, I guess

Not sure how we should handle array:
old:
SDL_DYNAPI_PROC(int,SDL_UIKitRunApp,(int a, char *b, SDL_main_func c),(a,b,c),return)
new:
SDL_DYNAPI_PROC(int,SDL_UIKitRunApp,(int a, char *b[], SDL_main_func c),(a,b,c),return)

this one seems incorrect:
extern DECLSPEC int SDLCALL SDL_WinRTRunApp(SDL_main_func mainFunction, void * reserved);
old has an extra "c"
SDL_DYNAPI_PROC(int,SDL_WinRTRunApp,(int a, char **b, void *c),(a,b,c),return)
vs
SDL_DYNAPI_PROC(int,SDL_WinRTRunApp,(SDL_main_func a, void *b),(a,b),return)

There are also ome SDL_OUT_BYTECAP differences, but the API is already written anyway.

To check that, I remove the whole dynproc.h file, regen, use "sort" and compare

[1bsyl]

This is the difference I believe that should be checked now:
OLD
NEW

if OLD is incorrect, it should be fixed ..

< SDL_DYNAPI_PROC(const SDL_assert_data*,SDL_GetAssertionReport,(void),(),return)
> SDL_DYNAPI_PROC(const SDL_AssertData*,SDL_GetAssertionReport,(void),(),return)

< SDL_DYNAPI_PROC(int,SDL_UIKitRunApp,(int a, char *b, SDL_main_func c),(a,b,c),return)
> SDL_DYNAPI_PROC(int,SDL_UIKitRunApp,(int a, char *b[], SDL_main_func c),(a,b,c),return)

< SDL_DYNAPI_PROC(int,SDL_WinRTRunApp,(int a, char **b, void *c),(a,b,c),return)
> SDL_DYNAPI_PROC(int,SDL_WinRTRunApp,(SDL_main_func a, void *b),(a,b),return)

< SDL_DYNAPI_PROC(SDL_assert_state,SDL_ReportAssertion,(SDL_assert_data *a, const char *b, const char *c, int d),(a,b,c,d),return)
> SDL_DYNAPI_PROC(SDL_AssertState,SDL_ReportAssertion,(SDL_AssertData *a, const char *b, const char *c, int d),(a,b,c,d),return)

[sezero]

SDL_assert_data <-> SDL_AssertData isn't actually wrong but is a left-over: the former still defined as the latter in SDL_assert.h. Same for SDL_assert_state <-> SDL_AssertState

[1bsyl]

I've applied, and may be this can be remove for SDL3:
https://github.com/libsdl-org/SDL/blob/main/include/SDL3/SDL_assert.h#L313 ?

[1bsyl]

a38ea6b
I've removed the old naming. Since it's been changed in 2.0.4, and that's not the most common API, it seems it doesn't worth an entry in WhatsNew nor README-migration ..
(cc @slouken )

I've put this in several commits if we want to cherry-pick them for SDL2

[1bsyl]

I've updated the array[] in SDL_UIKitRunApp.
I've tested with a function that has array, see below. "[]" are required, otherwise it won't even compile.

diff --git a/include/SDL3/SDL_render.h b/include/SDL3/SDL_render.h
index 1fac04152..a20aba885 100644
--- a/include/SDL3/SDL_render.h
+++ b/include/SDL3/SDL_render.h
@@ -168,6 +168,7 @@ typedef struct SDL_Texture SDL_Texture;
  */
 extern DECLSPEC int SDLCALL SDL_GetNumRenderDrivers(void);
 
+extern DECLSPEC void SDLCALL SDL_TestBuf(char *buf[]);
 /**
  * Get info about a specific 2D rendering driver for the current display.
  *
diff --git a/src/render/SDL_render.c b/src/render/SDL_render.c
index faa5c8c56..7f466bef6 100644
--- a/src/render/SDL_render.c
+++ b/src/render/SDL_render.c
@@ -851,6 +851,12 @@ static int SDLCALL SDL_RendererEventWatch(void *userdata, SDL_Event *event)
     return 0;
 }
 
+void SDL_TestBuf(char *buf[]) {
+        while (*buf) {
+            SDL_Log("%s", *buf);
+            buf++;
+        }
+}
 int SDL_CreateWindowAndRenderer(int width, int height, Uint32 window_flags,
                                 SDL_Window **window, SDL_Renderer **renderer)
 {
diff --git a/test/testgesture.c b/test/testgesture.c
index 004a0d87a..7c432b041 100644
--- a/test/testgesture.c
+++ b/test/testgesture.c
@@ -19,6 +19,7 @@
 #include <stdlib.h> /* for exit() */
 
 #include <SDL3/SDL.h>
+#include <SDL3/SDL_render.h>
 
 #ifdef __EMSCRIPTEN__
 #include <emscripten/emscripten.h>
@@ -276,6 +277,8 @@ int main(int argc, char *argv[])
     if (state == NULL) {
         return 1;
     }
+    char *buf[4] = { "aa", "bb", "cc", NULL };
+    SDL_TestBuf(buf);
 
     state->window_title = "Gesture Test";
     state->window_w = WIDTH;

[1bsyl]

Also updated the SDL_WinRTRunApp DYNAPI prototype because it's clearly different from the real api

[sezero]

Those were never caught because dynapi is actually disabled for ios and winrt

[sezero]

Those were never caught because dynapi is actually disabled for ios and winrt

And those two changes should be applied to SDL2 side if possible, for sake of correctness

[1bsyl]

Yes. I've applied the two patchs: faa7e3c 5efc9bd