RawDoFmt() handling over SDI + VA_ARGS & Co

We have done such a cross-platform function just like done in example of SDI_stdargs.c:



On gcc 4.2.4 its just bring a warning

Quote:

But still compiles ok (line 25 is that RawDoFmt call at end).

On gcc 4.4.3 on the same code it change warning on error now, and :

Quote:

Can someone point out of how to make it warning/error less ?

All what i can find out its 1 link in google, where Cyborg meet with the same warning, and somehow fix it, but as code is no more available (that was Amirc), so i can see what he do. Only few links:

https://www.ohloh.net/p/amirc/commits/160123982 - problem
https://www.ohloh.net/p/amirc/commits/160123992 - kind of fix

@kas1e

Solution 1. Check your macros particularly VARARGS68K are defined correctly for each OS

(try compileing with -E option and see what is output)

Solution 2.

Utility the version 50 IUtility->SNPrintf() function


Somthing like

#define lsprintf(buffer,fmt,data, ...) SNPrintf(buffer, sizeof(buffer),fmt,__VARAGS__)

That macro is wrong but you can look up simalar ones to fix it.

@Andy
Quote:

Already done by SDI, and seems ok for all other times.

Quote:

Should be crossplatform (i.e. SDI)

Quote:

With defines there a lot of other problems like one function one it there, another there, everything done like it want it as function.

As i read on the Cyborg's posts, he fix it somehow by casting, but by what casting and where - dunno.

@kas1e

Make sure that <amiga_compiler.h> is included before any SDI includes.

If it's not the SDI definition of VARARGS68K will be used which IIRC is just:
#ifndef VARARGS68K
#define VARARGS68K
#endif

@Fredrik
Quote:
Yeah, that helps, thanks !

@all

Can anyone help with STDARGS to VARARGS + SDI conversion ?

Thats what we have originally:




As you can see even if name of functions kind of the same as in SDks, its still different with different params, so simply redefine non-go there.

So, we do such replacement:



But that didn't works. Can anyone point out how it should be done ? Should be exactly SDI to make it works on all platforms.

@kas1e

It doesn't work because the tag/tag1 parameter should be a part of the varargs parameter ('...') and not be defined separately.

@salas00
It works! Thanks a lot !

May i ask you please to convert also those ones (i anyway understand nothing with that varargs/stdargs crap, just will copy + paste it):



As well as those 3 which a bit different:



Currently we replace it like this:



but dunno about launch_error() as well as m_newObject and L_newobject should be different, as well as "defines" looks crappy and better to do everything via VARARGS68K way so to keep the same look of sources.

Since the L_NewObject() function has an unused first parameter (cl) which isn't passed on to NewObjectA() a more correct macro definition would be:
#define L_NewObject(cl, ...) NewObject(__VA_ARGS__)

As for launch_error() it's hard to say without knowing what the code in the '.....blablab..' part is.

Quote:

@salas00
Thanks a lot !

Now almost everything done, just a little strange moment, which maybe related to redefine of launcher_SystemTags().
I.e. once i run any programm from dopus5 (any native binary), and then exit from it, then i have crash on the very begining of launcher_exit_code() function (i.e. stacktrace point out not even on first code-string in functino, but just on { braket.

And that laucher_exit_code used only one time when we run process over that launcher_SystemTags. Thats how that part looks like:



And launch_erro_code() is:



Dunno if it just something need to be added to create proc (like some PPC/OS4 flags only, like NP_Child or NP_NotifyOnDeathMessage or kind), or its just redefine of that laucher_SystemTags varags done wrong ? Maybe NP_ExitCode different somehow ..

All what i can say, is that the same 1:1 code (i.e. with all our new redefines / rewrites to VARARGS) works fine when we compile it for os3, but when just recompile it for os4, on exit we have such crash.

According to the autodocs the NP_ExitCode function should look like this:

void SAVEDS ASM launch_exit_code( REG(d0, LONG return_code), REG(d1, LaunchPacket *packet)) {
...
}

@salas00
Replaced, it works ! Yeah !

I also rebuild the with the same 2 params and for os3: works as well. What mean that when it compiles for os3, seems sdk somehow put something there ? Or why it works on os3 ? (and the same os3 version with 1 param works on os4 too, i.e. like os3 ndk/sdk do something for).

edit: answer from itix:

Quote:

so now all clear

That quote from itix is particularly important for hook functions.