lib: fix AIX build issues

[vszakats]

• memdebug: replace keyword malloc with __malloc__ to
  not interfere with envs where malloc is redefined. Also apply
  the fix to alloc_size.
  Fixes:

  lib/memdebug.h:107:13: warning: unknown attribute 'vec_malloc' ignored [-Wunknown-attributes]
  CURL_EXTERN ALLOC_FUNC FILE *curl_dbg_fdopen(int filedes, const char *mode,
              ^~~~~~~~~~
  lib/memdebug.h:37:37: note: expanded from macro 'ALLOC_FUNC'
  # define ALLOC_FUNC __attribute__((malloc))
                                     ^~~~~~
  /usr/include/stdlib.h:753:16: note: expanded from macro 'malloc'
  #define malloc vec_malloc
                 ^~~~~~~~~~
  

• memdebug: always undef before defining.
  Also do this for the rest of functions redefined in the same block.
  Avoids warning on AIX:

  lib/memdebug.h:117:9: warning: 'malloc' macro redefined [-Wmacro-redefined]
  #define malloc(size) curl_dbg_malloc(size, __LINE__, __FILE__)
          ^
  /usr/include/stdlib.h:753:9: note: previous definition is here
  #define malloc vec_malloc
          ^
  

• easy: fix -Wformat warning on AIX by adding a cast.

  lib/easy.c:608:47: warning: format specifies type 'int' but the argument has type 'long' [-Wformat]
  "%" CURL_FORMAT_SOCKET_T ")", fds[i].fd);
  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~
  

• if2ip: silence compiler warning inside AIX system header.

  /lib/if2ip.c:219:19: warning: signed shift result (0x80000000) sets the sign bit of the shift expression's type ('int') and becomes negative [-Wshift-sign-overflow]
  if(ioctl(dummy, SIOCGIFADDR, &req) < 0) {
                  ^~~~~~~~~~~
  /usr/include/sys/ioctl.h:401:26: note: expanded from macro 'SIOCGIFADDR'
  #define SIOCGIFADDR (int)_IOWR('i',33, struct oifreq) /* get ifnet address */
                           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
  /usr/include/sys/ioctl.h:174:23: note: expanded from macro '_IOWR'
  #define _IOWR(x,y,t) (IOC_INOUT|((sizeof(t)&IOCPARM_MASK)<<16)|(x<<8)|y)
                        ^~~~~~~~~
  /usr/include/sys/ioctl.h:168:20: note: expanded from macro 'IOC_INOUT'
  #define IOC_INOUT (IOC_IN|IOC_OUT)
                     ^~~~~~
  /usr/include/sys/ioctl.h:167:28: note: expanded from macro 'IOC_IN'
  #define IOC_IN (0x40000000<<1) /* copy in parameters */
                  ~~~~~~~~~~^ ~
  

Ref: https://curl.se/dev/log.cgi?id=20240808180420-3809007
Assisted-by: Dan Fandrich
Closes #14464

[dfandrich]

This fixes the compile issue when building with ibm-clang, but I found a gcc on this machine and it doesn't need this workaround. The clang case set a _LINUX_SOURCE_COMPAT macros which triggers creation of the malloc et. al. macros that cause the problem. But, whatever the reason for _LINUX_SOURCE_COMPAT being set, this PR does fix things there. I'm still running a few more compiles to check it out further.

[vszakats]

The #undefs were done already for other redefinitions. I think it's safe to keep.

As for the other, what do you think of this?:

diff --git a/lib/memdebug.h b/lib/memdebug.h
index aba289f4e7..0e12cc27a0 100644
--- a/lib/memdebug.h
+++ b/lib/memdebug.h
@@ -33,7 +33,8 @@
 #include <curl/curl.h>
 #include "functypes.h"
 
-#if defined(__GNUC__) && __GNUC__ >= 3 && !defined(_AIX)
+#if defined(__GNUC__) && __GNUC__ >= 3 && \
+  !(defined(_AIX) && defined(_LINUX_SOURCE_COMPAT))
 /* AIX defines a macro named 'malloc', which breaks the line below */
 #  define ALLOC_FUNC __attribute__((malloc))
 #  define ALLOC_SIZE(s) __attribute__((alloc_size(s)))

[vszakats]

Hm, this might be most future-proof, if it doesn't have a side-effect:

#if defined(__GNUC__) && __GNUC__ >= 3 && !(defined(_AIX) && defined(malloc))
/* ibm-clang defines _LINUX_SOURCE_COMPAT which in turn defines a macro named
  'malloc', which breaks the line below */

[dfandrich]

Actually, how about #if defined(__GNUC__) && __GNUC__ >= 3 && !defined(malloc) That will fix things on any other strange platform or environment that defined a malloc macro, since it could easily happen outside AIX.

[vszakats]

Actually, how about #if defined(__GNUC__) && __GNUC__ >= 3 && !defined(malloc) That will fix things on any other strange platform or environment that defined a malloc macro, since it could easily happen outside AIX.

I wasn't brave enough to go that far :) But yes, if there is no side-effect, it seems even better.

The ultimate solution would be if compilers supported the __malloc__ keyword alternative. I can see it used in a GitHub search, but couldn't find it earlier in the gcc manual.

[vszakats]

Confirmed on godbolt that both __malloc__ and __alloc_size__ are accepted by gcc and clang. (The latter requires gcc 4.4 (possibly 4.3 or 4.2), just like its non-underscored counterpart.)

[dfandrich]

I suppose the simple case of a malloc macro simply redirecting to another function would be fine with this block, so an unconditional !defined(malloc) would be overkill. This PR at commit a1b7743, with the !(defined(_AIX) && defined(malloc)) clause instead, is working with ibm-clang.

FYI, I uploaded an AIX gcc snapshot autobuild log.

[dfandrich]

The PR commit 0b3c791 works fine on ibm-clang 17.1 and gcc 10.3.

[vszakats]

This still triggers, though a bit differently. Is it possible the cast is missing?:

/curl-8.10.0-20240808/lib/easy.c:607:17: warning: format '%d' expects argument of type 'int', but argument 3 has type 'long int' [-Wformat=]
607 | "call curl_multi_socket_action(socket "
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
608 | "%" CURL_FORMAT_SOCKET_T ")", fds[i].fd);
| ~~~~~~~~~

[dfandrich]

The autobuild logs are pure daily snapshots—no patches. With this PR, those warnings don't show up for me with either ibm-clang or gcc.