The semantics of the GUID expression is inconsistent across platforms

[morozov]

Q	A
BC Break	no
Version	2.7.1

Summary

As a developer, I expect the GUID expression generated by AbstractPlatform::getGuidExpression() to have the same semantic (namely, UUIDv1) across supported platforms.

Current behavior

1. On MySQL platform, it's implemented as UUID() which generates UUIDv1 and is therefore correct.
2. On PostgreSQL platform, it's implemented it as UUID_GENERATE_V4() which generates UUIDv4. For some reason, UUID_GENERATE_V1() is unavailable on Travis CI:

   SQLSTATE[42883]: Undefined function: 7 ERROR: function uuid_generate_v1() does not exist

3. On Oracle platform, it's implemented as SYS_GUID() which returns RAW(16) which is not a UUID according to Stackoverflow.
4. On SQL Server and SQLAnywhere, it's implemented as NEWID(). From the documentation, it's not clear which UUID version it generates, but according to the test output, it's not UUIDv1:

   Failed asserting that 'DBB4D428-2338-4D4F-823F-4AA2ECEC985A' is greater than 'F4959EE5-02EC-4136-A21F-211813F67B63'

5. On SQLite, it's implemented using purely RANDOMBLOB() which makes it not compliant with the the UUIDv1 definition.

There's also a summary of the above on Wikipedia.

How to reproduce

Below is the pseudo-code. See a real example in #3162.

$previous = generateGuid();

for ($i = 0; $i < 99; $i++) {
    $value = generateGuid();
    self::assertGreaterThan($previous, $value);
    $previous = $value;
}

Expected behavior

All implementations generate UUIDv1 which looks impossible. Given the current state of the feature, would it make sense to deprecate it and encourage users to use application-generated UUIDs?

[Majkl578]

Handled in #3211 and #3212.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.