Re:How many ways are there to sort GUIDs?
by pdw on Tuesday April 30, 2019 @05:43AM
(#19786424)
I never understood why GUIDs were defined as
rather than simply
I get that traditional GUIDs have an internal structure, but it's entirely different than the struct (60-bit timestamp, 4-bit version, 16-bit sequence number, 48-bit MAC address).
[ Parent
| on HN ]
It's divided exactly as it is in string representation: 4 bytes time_low, 2 bytes time_mid, 2 bytes time_hi_and_version and then: other stuff - 8 bytes.
But the string representation has five fields, not four.
There might be an interesting story in tracing the history of GUIDs through the ages...
I found some 1987 code dealing with UUIDs here: https://stuff.mit.edu/afs/athena/astaff/project/opssrc/quota...
They seem to have originated as a 64-bit "uid"s:
struct uid_t {
u_long clockh;
u_int clockl: 4;
u_char extra: 8;
u_long node: 20;
};
Then the 1987 version seems to have been something like this (with the reserved field ignored in the string representation!) struct uuid_$t {
u32 time_high;
u16 time_low;
u16 reserved;
u8 family;
u8 host[7];
}
And I guess that's pretty close to the current struct...And to complete the story, I'm guessing that that old struct likely got serialized in the DCE protocol as dword/word/word/bytes, so the struct layout remained unchanged to maintain compatibility, even as the definition of an UUID evolved...
Nice find!
That seems like begging the question. Why it is divided like that in the string representation?
Re:How many ways are there to sort GUIDs?
by thunderbong on Tuesday April 30, 2019 @07:28AM
(#19786949)
I think if it was set like that many more programmers would be asking the reverse question - 'Why is it 16 bytes?' and it would be a Stackoverflow wiki question!
Maybe this is not the underlying reason, but in my opinion code should be written so that it is understandable and maintainable by future programmers