Rewrite of mod_unique_id:

[atlesn]

The old implementation has multiple issues also when the latest counter-modification is taken into account.

The old module was a bit difficult to understand with a lot of extra logic doing byte per byte operations "manually". I'm proposing a new structure to the whole module which is easier to modify.

The actual problem with uniqueness cannot be solved as intended by the original authors 20+ years ago. This implementation used IP-addresses inside the unique ID, this has since been removed.

Maybe a configuration parameter could provide a server ID (set manually) which can be used inside the string.

I'm also unsure about this "Thread ID" parameter, I haven't found any reliable source for this parameter. Currently a truncated pointer is used.

• Use Process ID, Thread ID, Timestamp, 16-bit per thread counter and 16 random bits
• Use provided library function for encoding base64
• Use packed struct instead of addressing each member manually in and index
• Use global counter /only/ if threads are not being used
• General re-structuring of functions
• Initialize counters on demand, remove two init functions
• Straight forward to change the build-up of the unique ID parameters after discussion.
• TODO : Add endian conversion, find better thread ID

[atlesn]

Which byte swap functions should be used for the fixed width types that work on all platforms and that preferably are used elsewhere in the code?

[atlesn]

I've added the configuration parameter UniqueIdServerId which is used to ensure uniqueness across different servers. I've also implemented byte swapping and changed the method of retrieving a thread ID.

[wjcarpenter]

@atlesn You might be interested in some of the conversation taking place on this bugid: https://bz.apache.org/bugzilla/show_bug.cgi?id=65159

[atlesn]

@wjcarpenter Thanks, I can see that somebody linked in this PR there as well. I've also posted this diff on bugzilla https://bz.apache.org/bugzilla/show_bug.cgi?id=65307