Iteration order

[zorg]

But order in pairs/next should be consistent until changes to the table are made.

Does the reference state such a guarantee?

[MrFariator]

It doesn't. Since the behavior is not codified in any official reference or specification as far as I'm aware, that behavior may be subject to changes between versions of lua/luajit. Can think of it like undefined behavior, that may work for you in the moment, but not necessarily in different environments.

[BrotSagtMist]

'The behavior of next is undefined if, during the traversal, you assign any value to a non-existent field in the table. You may however modify existing fields. In particular, you may clear existing fields.'
Which i would read as, its defined until you add stuff.

[slime]

That's talking about a separate bit of undefined behaviour. Iteration order is not specified to be consistent during any point.

[BrotSagtMist]

You would not be able to iterate at all if that is the case.

[pgimeno]

I think Brot has a point. next(element) must return a value. If pairs() is based on next() then next(element) must be consistent for a certain element, as long as no elements are added to the table, because pairs() isn't expected to keep track of whether next() is visiting an already visited element or a new one. It's hard to devise a scheme where next(element) can change when the table has not been modified, except if designing it on purpose.

The only realistic scheme I can think of, where order can change between calls to pairs() when the table doesn't change, is: hash based on addresses; if a table is unlocked, then it can be freely moved in memory, and since the hashes are address based, the order can change; next(x) locks the table except if it returns nil, in which case it unlocks the table. It's still a quite artificial scenario, and would make next(element) change for a certain element without modifying the table, just because you asked for a different one in between.

That doesn't mean it needs to be consistent between runs though, even if the elements in the table and the order of insertion of the elements are the same. Hashes based on addresses, as mentioned above, are a possibility, and due to ASLR the addresses are random.

[BrotSagtMist]

Booth the lua and the luajit pages state the order changes when elements are added.
To my understanding that must mean it does never change when no elements are added.
Thats my thinking here.

[slime]

It doesn't say that order never changes until an element is added. It does say this:

The order in which the indices are enumerated is not specified, even for numeric indices. (To traverse a table in numeric order, use a numerical for or the ipairs function.)

The behavior of next is undefined if, during the traversal, you assign any value to a non-existent field in the table. You may however modify existing fields. In particular, you may clear existing fields.

This means you can't rely on it guaranteeing any order and also the entire function's behaviour (not just iteration order) becomes undefined if you add a key while iterating. The only guarantee is that it iterates through all key/value pairs in the table, there's absolutely no order guarantees and there's not even a guarantee that order will change if you cause undefined behaviour by adding a key while iterating.

Following "don't rely on behaviour specified to be unreliable" as a rule isn't too hard in this situation, IMO. There are lots of valid ways to iterate through something in a reliable order, they just don't use pairs or next to do it.

[BrotSagtMist]

Well i see it as guaranteed. Thats just my understanding of this text.