bpo-11063: Fix _uuid module on macOS

[vstinner]

On macOS, use uuid_generate_time() instead of
uuid_generate_time_safe() of libuuid, since uuid_generate_time_safe()
is not available.

https://bugs.python.org/issue11063

[vstinner]

I tested this PR manually on macOS Sierra: test_uuid pass, and is_safe attribute is unknown as expected.

macbook:master haypo$ ./python.exe -m test test_uuid 
Run tests sequentially
0:00:00 load avg: 0.75 [1/1] test_uuid
1 test OK.

Total duration: 337 ms
Tests result: SUCCESS

macbook:master haypo$ ./python.exe -c 'import uuid; print(uuid.uuid1().is_safe)'
SafeUUID.unknown

Note: I hacked _netstat_getnode() in Lib/uuid.py to add "-n" option to make tests faster, to prevent slow DNS resolution.

[pitrou]

I'd prefer if you'd return the (out, None) tuple here.

[vstinner]

I prefer to have very thin wrapper for C functions, and handle differences at the Python level.

[pitrou]

IMO it only makes both the Python code and the C code more complicated.

[pitrou]

One of the annoyances with modifying uuid.py is that there are so different code paths to take care of.

[pitrou]

There's no need to expose two different function names, IMO.

[vstinner]

There's no need to expose two different function names, IMO.

It's not the same API:

• _uuid.generate_time_safe() returns (uuid, is_safe).
• _uuid.generate_time() only returns the uuid

[pitrou]

I'm proposing that it's the same API (see above) :-)

[vstinner]

I simplified my PR to not add a new function with a different name in _uuid. Instead, I modified uuid.py to decide if we have or not uuid_generate_time_safe() using sys.platform.

[vstinner]

Instead of checking sys.platform, maybe we could have a constant in _uuid?

[pitrou]

Yes, I think it's better to put this in _uuid.

[vstinner]

Yes, I think it's better to put this in _uuid.

Ok, done. I added _uuid.has_uuid_generate_time_safe (int: 1 or 0).

[pitrou]

Thank you! :-)

[warsaw]

I'd rather see a configure check for the existence of uuid_generate_time_safe() rather than hard coding it to platforms !APPLE for two reasons. 1) If macOS ever adds this API in some future release, this ifndef will be incorrect, and 2) if some other platform comes along that doesn't have this API, it will still use the incorrect function.

[vstinner]

@warsaw: "I'd rather see a configure check for ..."

Please, go ahead and write it :-) I began to look at AC_CHECK_HEADER() but I don't see how to pass /usr/include/uuid/ to ask to test also this directory. Then I wanted to use pkg-config, but @pitrou showed me that it doesn't exist on macOS.

In short, modifying the autotools are out of my skills. I proposed a pragmatic solution to fix the _uuid module on macOS.

[ned-deily]

I agree with @warsaw. This should be a configure check.