I thinkOneOf is less precise because it can be interpreted as "XOR".

@fancyweb 👍 : I didn't see it that way. But AtLeastOne suggest that many can succeed. But by the look of the code it stops as soon as a constraint has no violations. For me XOR equivalent would beExactlyOneOrNone

@Neirda24 there is nothing in the current implementation forbidding multiple ones to succeed. We stop the loop once we know the final result (the validation succeeds), because it is useless to run all constraint validators if we already know the answer. The fact that it stops is an internal implementation detail to optimize performance, but it does not change the result.

@stof : The fact that it stops once a constraint has no violations means we don't check for others. If we don't we can't know if more would have succeeded. Meaning that at this time the validator can only assure us that One and only one constraint did succeed. And regarding the fact that it is exactly what we want I thought that AtLeastOneOf was a bit redundant.

@Neirda24 even if we would be running them all, we would only report a violation if all of them failed, as that's the only case where theAtLeastOne constraint is violated.
AnAtLeastOne constraint does not care whether 1, 2 or 99999 constraints were satisfied to decide that it is valid.

The current implementation does not assure that one and only one constraint succeeded. To assure thatonly one part requires failing when 2+ are succeeding.

@Neirda24 I changed the initial name toAtLeastOneOf due to a suggestion in the PR. Asstof pointed out, the fact that it stops after one succeeds is an optimization. If the name is confusing in any way, maybe you could submit a PR to update the constraint's description in the documentation?

@przemyslaw-bogusz wedon't care if others succeed or not.At least one means exactly that at least one of them succeed. Theonly way this can fail is if all of them fail. So as soon as one of them did not fail, we already know the outcome.

I'm personally fine withOneOf as it means the same asAtLeastOneOf to me.
A XOR behavior would be extremely surprising so I couldn't expect this to mean that.

If we forget about the technical things for a moment, compare these two phrases:

Take one of these things.Take at least one of these things.

To me, "one of" and "at least one of" have completely different meaning.

Returning to technical discussions ... "one of" to me is equivalent to{1} in regular expressions and "at least one of" is equivalent to{1,} ... so they are not the same either. But maybe I'm missing something here 🤔

This is exactly my point. To me, the current validation does a{1} because (I understand it is for performance reasons) it stops checking other constraints as soon as one did succeed. If we run them all and then depending on whether at least one succeeded then{1,}.

In conclusion for me as long as we keep these optimization we change the behavior from At Least One Of to only One Of.

I don't know how it works. If there is a vote or something. So please just let me know if I can help in any way. (documentation as@przemyslaw-bogusz suggested).

IMOOneOf would be confusing exactly because of the{1} argument. It may lead users to think that they need to define a set of constraints where never more than one can pass at the same time.

I'm closing because this is not going to be accepted, let's move on. Thanks for proposing!

@Neirda24 the currently implementationdoes not do{1}. To do{1}, you have to report violations when 2 constraints (or more) are passing. The case where we can short-circuit the execution after one of them passes is precisely{1,} (where we only care about distinguishing between 0 and 1+ passing constraints).