@ansd you are right, for this setting from now on we should simply pick the operator policy value and not the lower or higher one. So, we need to tweak the resolving function.

@ansd you are right, for this setting from now on we should simply pick the operator policy value and not the lower or higher one. So, we need to tweak the resolving function.

unfortunately it isn't a simple as that as the policy resolution code doesn't include queue arg values. i'd have to add some code to a do a resolution based on all of queue arg, policy and operator policy.

Alternatively we treat all values above some limit, say 65535, as unlimited rather than special casing -1. In this case the current resolution logic would work fine.

Alt we could just say that any policy would override the queue arg if the queue arg is negative.

@kjnilsson I like the idea with values above the (2^16-1) being treated as "unlimited". Because that's what the user who'd set it to, say, 100K, would arguably want.

Alt we could just say that any policy would override the queue arg if the queue arg is negative.

Yes, I think the "more conservative" value should always win for (x-)delivery-limit where "more conservative" means lower value except for -1 which means "unlimited".

Examples:

1. Queue argument = -1, Policy = 5 should result in 5
2. Queue argument = 5, Policy = -1 should result in 5
3. Queue argument = -1, Policy = -1, Operator Policy = 5 should result in 5
4. Queue argument = 0, Policy = -1, Operator Policy = -1 should result in 0

This allows operators to set guardrails via operator policies while applications can optionally choose a more conservative setting.

In addition, this resolution strategy won't break applications upgrading from 3.13 to 4.0.

Yes, I think the "more conservative" value should always win for (x-)delivery-limit where "more conservative" means lower value except for -1 which means "unlimited".

I went with this approach in the end. -1 isn't without precedence as means of turning some facility off so in this sense it maybe less surprising.