gh-114917: add support for AI_NUMERICSERV in getaddrinfo emulation

[maxbachmann]

• Issue: add support for AI_NUMERICSERV in getaddrinfo emulation #114917

[rruuaanng]

LGTM

[vstinner]

Can you write a test for this feature?

[maxbachmann]

I am not sure which currently tested target even runs this code. This is only included conditionally via:

#if !defined(HAVE_GETNAMEINFO)
#define getnameinfo fake_getnameinfo
#include "getnameinfo.c"
#endif // HAVE_GETNAMEINFO

[maxbachmann]

I added a test. However this only tests the fallback if we actually have a target using it that is under testing.
We use this fallback implementation on the playstation but we don't have unit tests running there.

[vstinner]

LGTM. I just have a minor coding style request.

[maxbachmann]

@vstinner is there still something stopping this from getting merged?

[vstinner]

@vstinner is there still something stopping this from getting merged?

I don't know Modules/getaddrinfo.c. Which platforms are using it? It looks like dead code to me. It starts with a big #if 0 and then most code is disabled unless the HAVE_NETDB_H macro is defined.

My pyconfig.h has:

/* Define to 1 if you have the <netdb.h> header file. */
#define HAVE_NETDB_H 1

but Modules/getaddrinfo.c doesn't include pyconfig.h nor Python.h.

[vstinner]

I tried to using this Linux with the patch:

diff --git a/Modules/socketmodule.c b/Modules/socketmodule.c
index 916ad35623e..c49d0d53042 100644
--- a/Modules/socketmodule.c
+++ b/Modules/socketmodule.c
@@ -448,6 +448,8 @@ remove_unusable_flags(PyObject *m)
 
 #endif
 
+#undef HAVE_GETADDRINFO
+
 /* I know this is a bad practice, but it is the easiest... */
 #if !defined(HAVE_GETADDRINFO)
 /* avoid clashes with the C library definition of the symbol. */

On Linux, I get many build errors:

./Modules/getaddrinfo.c:206:30: error: 'EAI_MAX' undeclared (first use in this function); did you mean 'AF_MAX'?
./Modules/getaddrinfo.c:285:17: error: 'EAI_BADHINTS' undeclared (first use in this function); did you mean 'EAI_BADFLAGS'?
./Modules/getaddrinfo.c:286:32: error: 'AI_MASK' undeclared (first use in this function)
./Modules/getaddrinfo.c:354:19: error: invalid type argument of '->' (have 'struct addrinfo')
./Modules/getaddrinfo.c:384:25: error: 'EAI_PROTOCOL' undeclared (first use in this function); did you mean 'SO_PROTOCOL'?
./Modules/getaddrinfo.c:512:10: error: implicit declaration of function 'getipnodebyaddr'; did you mean 'getnetbyaddr'? [-Wimplicit-function-declaration]
./Modules/getaddrinfo.c:512:8: error: assignment to 'struct hostent *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
./Modules/getaddrinfo.c:561:14: error: implicit declaration of function 'getipnodebyname'; did you mean 'getprotobyname'? [-Wimplicit-function-declaration]
./Modules/getaddrinfo.c:561:12: error: assignment to 'struct hostent *' from 'int' makes pointer from integer without a cast [-Wint-conversion]
./Modules/getaddrinfo.c:564:12: error: assignment to 'struct hostent *' from 'int' makes pointer from integer without a cast [-Wint-conversion]

How did you test your change? Are using using macOS?

[maxbachmann]

I don't know Modules/getaddrinfo.c. Which platforms are using it? It looks like dead code to me. It starts with a big #if 0 and then most code is disabled unless the HAVE_NETDB_H macro is defined.

It's only used when HAVE_GETADDRINFO isn't defined. In this case it's directly including the c file and so doesn't need the imports.

On Linux, I get many build errors:

I assume you didn't undef this for enough code. The include of addrinfo.h in line 423 would need this define to already be removed.

How did you test your change? Are using using macOS?

We are using it in our playstation port of CPython

[vstinner]

We are using it in our playstation port of CPython

Oh ok. A few lines to support PlayStation sounds reasonable, I merged your PR. Thanks.