We should be using git__malloc() here, or even git__calloc(). And yes, we should just bail out with the macro.

Am 2015-12-31 16:19, schrieb Carlos Martín Nieto:

In src/posix.c [1]:

@@ -62,8 +62,11 @@ int p_getaddrinfo(
ai = ainfo;

 for (p = 1; ainfo->ai_hostent->h_addr_list[p] != NULL; p++) {

•    ai->ai_next = malloc(sizeof(struct addrinfo));
  

•    memcpy(&ai->ai_next, ainfo, sizeof(struct addrinfo));
  

•    if (!(ai->ai_next = malloc(sizeof(struct addrinfo)))) {
  

We should be using git__malloc() here, or even git__calloc(). And yes,
we should just bail out with the macro.

Note however that this is a replacement function for getaddrinfo() that
also doesn't follow libgit2 semantics. For instance, GITERR_CHECK_ALLOC
will not clean up properly while the standard getaddrinfo() is supposed
to cleanup properly in a failure case. So should this really being used?
In posix.c there is quite a mixture of styles so it is not clear what
style should be really used. But if you say git__malloc() and
GITERR_CHECK_ALLOC then I will adjust my changes.

It's not 100% expected that the p_* functions that we write should set library error codes. These are often used as fallbacks when an actual posix function is not available. And - of course - the actual posix functions (like the system's getaddrinfo) will never set the library error codes.

Therefore p_getaddrinfo doesn't need to set the library error code. Instead, it needs to set something that p_gai_strerror can cope with (which is our reimplementation of gai_strerror. Indeed that does so - -1 is treated as an OOM scenario.

As a result, I think this change is correct.