document non-reproducibility of objectid

[mhinsch]

At least for some types objectid usually varies across program runs (I assume because it's simply a pointer value). As that can break numerical reproducibility of a program it would be good to mention this in objectid's doc string.

[nsajko]

Make a PR? Just adapt the phrasing from the hash doc string:

The hash value may change when a new Julia process is started.

[vtjnash]

Do you have an example? We might just have a simple bug in some case or other

[oscardssmith]

objectid([1]) is the simple example here.

[Seelengrab]

[sukera@tower ~]$ julia -e 'println(objectid([1]))'
9761962143716435815
[sukera@tower ~]$ julia -e 'println(objectid([1]))'
4188474808741520571

[vtjnash]

If you make 2 different objects, of course those generally should have different objectid values, since they are different objects

[mhinsch]

If you make 2 different objects, of course those generally should have different objectid values, since they are different objects

Sure, if you think about it it makes perfect sense. If you don't it's very confusing, though. In my case I was using objectid to sort arrays of objects in a simulation, so that I could make sure there were no duplicates. It took me half a day to realise that that was the reason my simulations seemed to behave non-deterministically.

[vtjnash]

The objectid is not unique

[mhinsch]

The objectid is not unique

It's not, not even for objects of the same type? I thought it was essentially just the object pointer for mutable structs.

[nsajko]

The objectid doc string clearly implies the return value is not necessarily unique. My guess is any uniqueness is implementation-defined.

[disberd]

If you make 2 different objects, of course those generally should have different objectid values, since they are different objects

Does this imply that if I do use immutable objects not referencing other mutable objects the objectid should return the same value at least for the same Julia version?

[vtjnash]

Yes, objectid is likely surprisingly stable (though not guaranteed), even across machines and versions, since we have very little incentive to change the algorithm very often

$ ./julia1.13/julia -E 'objectid(1)'
0x5bca7c69b0d0ebca
$ ./julia1.12/julia -E 'objectid(1)'
0x5bca7c69b0d0ebca
$ ./julia111/julia -E 'objectid(1)'
0x5bca7c69b0d0ebca
$ ./julia1.10/julia -E 'objectid(1)'
0x5bca7c69b0d0ebca
$ ./julia1.9/julia -E 'objectid(1)'
0x5bca7c69b0d0ebca
$ ./julia1.8/julia -E 'objectid(1)'
0x5bca7c69b0d0ebca